Calven.ai Resources

What Is an AI Product Marketing Platform?

David Kolinek — Sat, 18 Jul 2026 00:00:00 GMT

An AI product marketing platform is one system that runs the entire product marketing function, from research through positioning, launches, and enablement, against a single shared knowledge base, so the work stays current without a person re-running it. It is not a competitive intelligence tool with a chat box bolted on, and it is not a stack of point tools sharing a login. It is the whole product marketing workflow, connected, so a signal found anywhere shows up everywhere it should. This page defines the category: what it covers, why it could not exist until now, and what makes it a platform rather than a bundle.

What is an AI product marketing platform?

Start with the plainer term inside it. A product marketing platform is one system that covers the full set of jobs a product marketing team owns, from research and audience through positioning, launches, campaigns, and enablement, instead of a separate tool for each. Sales has had this for decades in the CRM. Product marketing has not. What a PMM buys today is a shelf of point tools, each owning a corner and leaving the handoffs between them to a person.

The AI version adds the part that makes the coverage worth having. Software that only stores the work still leaves a person to move it between the pieces. On an AI platform the work runs against one shared base and stays current without anyone re-running it. Full coverage is what makes it a platform. Keeping that coverage connected is what makes it useful, and that connection is the part that only became possible recently.

What does a product marketing platform cover?

Start from the job, not the software. A product marketing team owns a connected chain of work, and a platform holds every link of it on one base. Those are the six functions in the diagram above, covered in one place instead of one tool each.

Research watches the market, the competitors, and the reasons deals are won and lost. Target audience defines who the product is for, the ICP and the personas behind it. Positioning and messaging decide how the product wins and turn that into the words that land. Product promotion takes it to market, from launches to the pages and collateral that carry it. Field enablement arms sellers with the battle cards and one-pagers they open inside a deal. And campaign strategy shapes the story the demand team runs on.

Point tools cover one of these and hand the seams back to you. A platform runs all six against the same base, so a claim that research turns up reaches the positioning, the messaging, and the field without a person carrying it across. That connection is the whole point, and holding it together by hand is exactly what never lasted.

Why couldn't this category exist before AI?

Here is the question that explains the timing. If the whole function held together is so clearly better, why did no one build a product marketing platform before? Sales got a CRM decades ago. Product marketing got a shelf of point tools and a person to hold the seams.

The answer is that a product marketing platform without AI holds together only by heroics, so no one built it as a product. Connect every function in one base and you have also created a coordination job. Someone has to read every call transcript and pull out what matters. Someone has to notice a competitor's move and decide which battle cards it touches. Someone has to check that the messaging a draft inherited is still the current one. A determined PMM can do all of this by hand, and plenty have, with a wiki and a lot of late nights. It just decays the moment they stop, because the whole function held together by one person is a second full-time job pretending to be software. The connected design was always the better one. It was simply unsustainable to run by hand, so the market settled for tools that each did one slice and asked a person to carry the rest.

Automation is the first thing that removes that coordination cost at scale. A Voice of Customer Agent reads every transcript, so the knowledge is captured instead of lost between calls. A built-in feedback loop pressure-tests each draft against the current positioning and personas, so the connected work validates itself instead of drifting out of sync. And because the base is reachable over MCP, from inside the AI tools a team already works in, like Claude, ChatGPT, or Cursor, the platform is not one more dashboard someone has to remember to open. Take any of those away and the platform collapses back into the manual coordination that made it decay before.

So the category is not new because someone finally thought of it. The idea is old. It is new because the thing that makes it hold together, work that runs the function and keeps it connected without a person carrying it, did not exist until now. The first platforms to cover the full function are arriving already AI-native, with no manual, pre-AI version to replace, and Calven is one of them.

For years the best parts of product marketing stayed buried under the busywork of keeping everything in sync. AI finally clears that away. What is left is the part I love: the thinking, the strategy, the storytelling.

— David Kolínek, Co-founder at Calven

What makes it a platform and not a bundle?

What makes it a platform is the shared knowledge base underneath it. Every function reads from and writes to the same base, so a change in one place moves to the others automatically, with no one carrying it across. A competitor's move flags the positioning for review. A shift in the positioning updates the battle card that still quotes the old line. That automatic propagation is the line between a platform and everything it gets confused with. A CI tool watches competitors and hosts cards. A messaging tool drafts copy. A martech listicle lists both and leaves the wiring to you. A platform is the whole function on one base, connected once and kept in sync on its own.

There is a second reason the category matters now. As Wharton's Stefano Puntoni argues, AI agents are beginning to act as buyers and shortlist vendors themselves (HBR, 2026). When a machine reads your positioning to decide whether you make the list, a version that is current and consistent everywhere stops being hygiene and starts being the whole game. That consistency is exactly what a connected platform protects and a shelf of point tools cannot.

What does "connected" actually look like?

Concrete beats abstract, so walk one chain end to end, with Calven as the example. A Voice of Customer Agent reads a win/loss call transcript and pulls the quotes worth keeping out of it, then sorts them by who needs them. The line where the buyer repeats a rival's pitch goes to the Competitive Intelligence Agent. The line where the buyer describes what they were actually trying to fix goes to the Personas Agent. Each agent takes its quotes and updates what it owns. The CI agent revises the battle card that rival appears on. The personas agent rewrites the persona's description to match how the buyer actually talks, and because that persona changed, the Messaging Agent reworks the messages that leaned on the old version.

One transcript, and the competitive card, the persona, and the messaging all move, each by the part of the platform that owns it, all against the same base. The edits land as changes you approve, or, once you have moved the dial, ones that ship on their own. That is the shape of the category, and the speed of it is what tracks to revenue. Teams that share competitive intelligence weekly or faster report revenue impact at 79%, against 41% for teams that share monthly or slower, per Crayon's 2026 State of Competitive Intelligence report. The gap is not in what they detect. It is in how fast what they detect reaches the deal. When you are ready to weigh one platform against another, the small-team buyer's guide turns this into the questions to ask and the red flags to watch. The test is the same one every platform has to pass: does a signal reach the work that depends on it, and can you set how much runs without you. Collection was never the hard part.

Where does the category go from here?

The category is early. Its definition is still unsettled, and most teams are still working off a shelf of point tools. What changed is that the technology to run the connected version finally exists, so the constraint that kept the platform theoretical is gone.

2026 is when that starts to land. The teams adopting first are not only swapping one tool for another. They are changing how the work moves: the base holds the positioning, the personas, and the competitive picture, and the updates travel across the function without a person shuttling them. That is a shift in how a product marketing team operates, not just what it buys. The platforms exist now. What is still being decided is how quickly teams reorganize their work around one.

We are glad to be at the front of that shift at Calven. If it goes the way we think it will, the win is not one more tool in the stack. It is product marketers spending their days on the parts of the job worth loving.

An AI Product Marketing Platform: A Small-Team Buyer's Guide

David Kolinek — Sat, 18 Jul 2026 00:00:00 GMT

Choosing an AI product marketing platform comes down to a short list of questions, not a feature comparison. The questions worth asking test whether a signal actually travels from where it is found to the rep who needs it, and whether you can control how much the agents do on their own. If the category itself is new to you, that explainer covers what one is and why it took AI agents to make it possible. This guide is the buyer's side of it. It matters most to the teams with the least slack. 44.3% of product marketing teams are one or two people (PMA 2025 State of Product Marketing), and 41% of PMMs hired in 2026 said the role's scope wasn't realistic for one person (Fluvio 2026 Hiring Trends).

Why do point tools leave a small team behind?

Because point tools automate collection, and collection was solved years ago. The value moved to what happens after a signal is detected, and that part still runs on a person. For a two-person team, that person is the whole team, and the carry is the work that never gets done at the same time as everything else.

Walk one signal through. A competitor reprices. The CI tool flags it, on time and accurately. Now someone has to decide the change matters, rewrite the battle card a rep pulls up mid-deal, adjust the messaging, and tell the field. None of that is collection. All of it is the carry, the part that stays on a person, and it is invisible on the order form. It is the same collection-versus-propagation test you would put to any competitive intelligence tool, applied to the whole workflow instead of one silo.

The speed of that carry, not the volume of collection, is what tracks to revenue. Teams that share competitive intelligence weekly or faster report revenue impact at 79%, against 41% for teams that share monthly or slower, per Crayon's 2026 State of Competitive Intelligence report. The gap is not in what they detect. It is in how fast what they detect reaches the deal. That speed is exactly what a small team has no hours to protect.

A small team doesn't need one more thing that collects. It needs the connections made for it, so a signal reaches the rep without anyone carrying it.

— David Kolínek, Co-founder at Calven

What should a small PMM team look for?

Six criteria, phrased as questions you can put to any vendor, including Calven. They are not a feature audit. Each one tests whether a signal survives the trip from where it is found to where a rep uses it, and whether you stay in control of that trip. Skip the demo choreography and ask these.

Two CI-specific criteria, relevance triage and time-to-propagate, live in the CI tools checklist; this table is the platform-level test that sits above them.

Criterion	The question to put to a vendor	A strong answer, and the red flag
Connected knowledge, one base	"Do the agents read and write into one shared base, or does each watch its own source?"	A competitive signal is visible to the messaging and persona work without anyone copying it across. Red flag: separate feeds you stitch together yourself.
Propagation over collection	"When something is found, does it reach the card, the messaging, and the rep, or just my inbox?"	Ask to watch the signal land inside an updated artifact. If the demo ends at the notification, the last mile is still your job.
Autonomy you can dial	"Can I set how much the agents do on their own, from approving every change to running end to end?"	Human in the loop by default, approvals you turn off as trust grows. Be wary of either extreme.
Grounding and citations	"When an agent makes a claim, does it cite the source or invent it?"	Every claim traces to a transcript, a page, a signal you can open. Red flag: fluent output with nothing behind it.
True cost, staffing math	"Once it detects something, who on my team authors and maintains the result?"	It removes propagation labor. If the honest answer is "a dedicated owner, a day or two a week," add that salary to the quote.
Reachable where you work	"Can I query and update this from inside the AI tool I already use, or only a dashboard?"	Answers inside Claude, ChatGPT, or Cursor over MCP. Red flag: knowledge locked in one more vendor UI you have to remember to open.

The grounding row is not academic. 72% of frequent AI users name inconsistent or low-quality output as their most-cited concern (Fluvio 2025 AI Trends). Ungrounded generation is the problem, not AI, and citations are how you tell the two apart. Run the staffing row hard, too. A platform that detects everything and updates nothing is a worse UI with a bigger invoice, and it belongs in the same total-cost conversation as the rest of your product marketing tech stack.

Can you control how much an AI agent does on its own?

Yes, and whether you can is the criterion that actually separates platforms. Autonomy is a setting, not a fixed property. The right buy runs human-in-the-loop by default, lets you keep approvals on while you build trust, then lets you turn them off one workflow at a time as the agents earn it, up to carrying a signal from the call to the card unattended.

Both failure modes are real, and a small team can afford neither. A platform that can never act alone leaves the carry exactly where it was, on your desk. One that acts with no way to keep a human in the loop is untrustworthy for a team with no margin to catch a confident mistake. The point is not how autonomous the agents are. It is that you hold the dial and can move it in either direction as your confidence changes. Calven runs a human in the loop by default and lets the approvals come off when you decide they should, but the criterion belongs to the category, not to any one vendor. Autonomy you cannot control is not a feature. It is a risk you did not agree to.

When does an AI product marketing platform pay off?

When the shared base is fed and someone acts on what it surfaces. Scope is not the deciding factor. Even a team whose only goal is current battle cards is better served by connected collection than by a card tool on its own, because a card is only as good as the signals behind it and how fast they reach it. A card built from one source and updated by hand is the one that goes stale. The multi-source collection and automatic propagation a platform runs is exactly what keeps that card current, so a narrow goal is a reason to connect the work, not to avoid it.

What a platform does need is a team that will use it. It earns its place when you feed it your calls, your positioning, and your competitors, and let it carry what it finds into the artifacts your team uses, with a human in the loop for as long as you want one. A team that will not keep the base current or act on what it surfaces gets no more from a platform than from the tools it already underused. The blocker there is never the technology. It is the willingness to let the work flow differently.

What does "connected" look like in a demo?

Concrete beats abstract. The pillar explainer walks one signal end to end: a single win/loss transcript that updates the battle card, the persona, and the messaging at once, each by the agent that owns it, all against the same base. That is the shape you are buying. Hold any platform, this one included, to the same test: does a signal reach the work that depends on it, and can you set how much the agents do without you. Collection was never the hard part.

Can You Use ChatGPT or Claude for Competitive Intelligence?

David Kolinek — Mon, 20 Jul 2026 00:00:00 GMT

Yes. For one band of competitive-intelligence work, ChatGPT and Claude are genuinely good, and any honest answer starts there. The ceiling isn't how smart they are, it's operational. A model is strong when you hand it a source, and solid on a one-off research pass. It gets shaky when the work has to run every week, reach past what is publicly crawlable, and stay accurate when most of what is public is marketing. The honest picture of AI in product marketing is a jagged frontier, and competitive intelligence sits right on the edge of it.

What ChatGPT and Claude are genuinely good at for competitive intelligence

They are good at two things. The first is synthesis of material you supply: paste a competitor's homepage, an earnings transcript, or a stack of win/loss notes into either model and it collapses that raw text into structured output faster than any analyst could. You hand it the source, so it is organizing something real, not guessing at what is true.

The second is a one-off research pass. Point either model at a competitor you know nothing about and it will browse the live web and hand back a serviceable first brief: the positioning, the rough product shape, the obvious talking points. As a starting point that is real time saved. The catch lives in both words. It is one-off, and on the open web it is only as accurate as whatever it happened to read. A source you trust and a record that stays current are exactly what go missing next.

A prompt set that actually works for CI

The prompts that hold up share one shape: give the model a role, separate the instructions from your pasted-in source with labeled tags, spell out the output you want, and forbid guessing. That last rule is what keeps it honest, and it matches Anthropic's prompt guidance. Paste each one in as written and drop your source where it is marked.

Start with a competitor's own words. It reads their site the way you would brief an analyst, and pairs with a structured competitor analysis.

Synthesize a competitor's positioning from their site copy

<role>
You are a senior B2B product marketer analyzing a competitor's positioning.
</role>

<task>
Read the competitor site copy in <source> and produce a positioning breakdown. Work only from the text provided. Add no outside knowledge and infer nothing beyond what is written.
</task>

<instructions>
1. Positioning: state their positioning in one sentence covering who it is for, the category they claim, and their stated differentiation.
2. Value propositions: list the top three they lead with, quoted in their own words.
3. Target persona: name the single buyer persona the copy is written for.
4. Weak claims: flag every claim that is vague, generic, or unsubstantiated.
</instructions>

<rules>
- Quote their exact wording for the value propositions.
- If the source does not support a field, write "not stated" instead of guessing.
- Report what their positioning is; do not judge whether it is good.
</rules>

<output_format>
Four labeled sections: Positioning / Value propositions / Target persona / Weak claims.
</output_format>

<source>
[PASTE HOMEPAGE AND PRODUCT-PAGE COPY HERE]
</source>

Swap in as many pages as you have. The 'not stated' rule is the anti-hallucination catch, keep it.

The same shape works on anything spoken, a transcript you already hold.

Summarize a competitor transcript you paste in

<role>
You are a competitive intelligence analyst preparing notes for a sales team.
</role>

<task>
Summarize the competitor earnings call, keynote, or webinar in <transcript>. Report only what is stated. Do not infer strategy that is not said out loud.
</task>

<instructions>
1. Strategic priorities: what they say they will focus on over the next 12 months.
2. Product, pricing, or packaging changes: any mentioned, with the detail given.
3. Market direction: any segment or market they say they are moving toward or away from.
4. Seller-usable quotes: verbatim lines, each with the speaker's name, that a rep could cite.
</instructions>

<rules>
- Attribute every quote to a named speaker; drop any quote you cannot attribute.
- If a section is not covered, write "not covered."
- Do not editorialize or add competitive commentary of your own.
</rules>

<output_format>
Four labeled sections matching the instructions above.
</output_format>

<transcript>
[PASTE TRANSCRIPT HERE]
</transcript>

Works on any transcript you already hold, an earnings call, a keynote, or a webinar recording. Drop it inside the tags.

And on your own win/loss calls, where a buyer names the rival and repeats their pitch.

Cluster win/loss notes into themes

<role>
You are a product marketer analyzing win/loss interview notes.
</role>

<task>
Cluster the closed-deal notes in <notes> into themes. Each note is tagged won or lost. Use only the reasons stated in the notes.
</task>

<instructions>
1. Group the stated reasons into themes.
2. For each theme, report: the theme name, how many deals it appears in, whether it skews won or lost, and one representative verbatim quote.
3. Rank the themes from most to least frequent.
</instructions>

<rules>
- Do not invent reasons that are not present in the notes.
- Keep quotes verbatim; do not paraphrase them.
- List a theme only if it appears in two or more deals; put one-off reasons under a separate "Outliers" heading.
</rules>

<output_format>
A table with columns: Theme | Deals | Skews won/lost | Representative quote. Then an \"Outliers\" list.
</output_format>

<notes>
[PASTE NOTES HERE, EACH TAGGED WON OR LOST]
</notes>

Paste the raw notes, tagged won or lost. Ranking by frequency surfaces the themes worth a battle-card update.

Every version works because you supply the source. Ask the model to go find the intel itself and it manages a decent one-off pass, but the limits start when that pass has to stay current, reach past what is public, and survive the fact that most of what is public is marketing.

What a raw LLM can't do well for competitive intelligence

The wall isn't intelligence. It is four operational jobs a raw chatbot does badly: staying current, reaching where CI actually lives, staying accurate when a deal is on the line, and keeping a living document correct over time.

Staying current

Ask a model to research a competitor today and it browses the live web and does a solid job. The problem is not reach, it is repetition, and a general chatbot is not built to keep a read current week after week. Scheduled runs close part of the gap: ChatGPT Scheduled Tasks and Claude Routines fire a saved prompt on a schedule, capped at roughly once an hour. But you script each task by hand, and every run starts fresh with no memory of the last, so nothing accumulates and nothing judges whether a change matters. Calven's Competitive Intelligence Agent monitors continuously, raising a change event only when a page actually changes, filtering for what is material, and keeping structured records that carry prior state across runs.

Reaching where CI actually lives

Most competitive intelligence sits behind surfaces a browsing session can't reliably open.

CI surface	Why a chatbot can't reach it
Ad transparency libraries (Meta, Google, LinkedIn)	Exposed through verified developer APIs, not open pages
G2, Capterra, TrustRadius	No public API, and Scrapfly documents the anti-bot defenses that block automated reads
LinkedIn and X	LinkedIn's User Agreement prohibits scraping and bots, exposing only limited APIs
Reddit	Behind an authenticated API and rate limits, not free-form browsing
Login-gated job boards	Behind logins and bot defenses a consumer session hits as a wall

Calven's CI Agent reaches these through verified data paths, official APIs and licensed datasets across ad libraries, review sites, social, hiring signals, filings, and partner directories.

Accuracy when a deal is on the line

Accuracy is where the stakes are highest, and a raw LLM gets a competitor fact wrong two ways. First, it makes things up, and browsing softens hallucination without removing it. The Tow Center found eight AI search engines gave wrong citations on over 60% of news-source queries and fabricated links, and Klue has documented ChatGPT inventing fake pricing during competitive research. Even purpose-built retrieval-backed legal tools still hallucinated 17% to 34% of the time in Stanford's testing, and general models missed on 58% to 88% of legal queries, a checkable stand-in for deal support.

Second, and worse for CI: when the model researches on its own, it reads what is public, and what is public is mostly marketing. Vendor sites overclaim by design, presenting what is merely planned as if it were built. So a model repeats claims a competitor cannot back, and a gap analysis against your own product does the same to you, crediting your marketing with capabilities the product does not have. The output looks thorough. It is built on assets written to persuade, not to be true.

An inaccurate competitive claim doesn't stay abstract. It lands on the person least able to absorb it.

The worst thing that can happen to a sales rep is that you arm them with a battle card that's inaccurate. It calls out some weakness of the competitor, or a missing feature the vendor actually has, that the AI research didn't find. That rep loses all their credibility, because to the buyer it looks like they have no idea what they're talking about.

— David Kolínek, Co-founder at Calven

Accuracy here isn't a nice-to-have, it is the whole job. The cure is not a sharper prompt. It is every claim tied to a source you can open, and every marketing claim checked against what the documentation actually supports. Calven's CI Agent works citation-first, and reconciles a competitor's marketing claims against their own product docs, flagging the ones the docs don't back.

Keeping a living document current

This is the one nobody explains. Point a scheduled routine at a battle card or a Google Doc and every run rewrites the whole thing, regenerating text that never changed. You can't fully prompt it away: asked to "update" a document, a one-shot LLM produces a new document, it doesn't diff and patch. The cost is version churn, overwritten hand-edits, and no clean trail of what actually moved.

An LLM routine told to "update" the doc	A diff-based system
Regenerates the entire document each run	Compares the new crawl to the last and captures only the diff
Overwrites hand-edits and reflows untouched sections	Leaves untouched rows untouched
Leaves no clean trail of what actually changed	Writes a dated, sourced row for each confirmed change

An agent can be built to diff and patch, which is what Calven does: it captures only what changed, writes only confirmed rows, and updates the record surgically rather than re-authoring it. Autonomy is a setting, human-in-the-loop by default, or switched off when you want the agent to carry the change end to end.

Gathering intel is step one. Making it useful is the job.

Collecting competitor intel is the easy half. The job is turning it into a battle card the field actually opens, a positioning update that ships, and a record that stays current without a person babysitting it. That is where a one-shot chatbot stops.

The best DIY setups make this concrete: you can wire Claude skills to watch a handful of signal types and it works, but you script every task, own the reliability, and still get gathered intel rather than an updated asset. The real test when you evaluate a CI system is that last mile, collection is solved, propagation is where tools separate. It is the difference between a chatbot you prompt and agents that operate the work continuously.

So the real question was never whether ChatGPT can gather competitive intel. It can. The question is who keeps that intel current and correct once it is gathered. Answer that, and you have answered whether a raw LLM is enough.

Product Marketing Should Own Your Company's AI Knowledge

David Kolinek — Sat, 18 Jul 2026 00:00:00 GMT

Nearly every product marketer now has a chat window open in a second tab, and almost none of them get more than a competent first draft out of it. Adoption is total. The payoff is thin. The honest answer to how a PMM should use AI comes in two parts, and most teams only see the first. Part one: treat AI as a fast first-drafter on the work you already own, and stop expecting a chatbot to know your company. What separates a useful draft from a generic one is not the prompt. It is grounding. Part two is the job most teams miss. AI is reshaping the role, and product marketing already owns the knowledge every other team's AI is starving for, so putting that knowledge in front of the whole company's AI is part of the job now.

Which product marketing tasks can AI actually handle today?

AI reliably handles first-pass structuring and drafting. It does not do the judgment. That line, drafting yes and judgment no, is the whole honest boundary, and it holds across nearly every use case a PMM reaches for. Adoption already reflects it: 96% of B2B marketers use AI, per Demand Gen Report's 2026 B2B Trends Research Report. The question stopped being whether to use it and became what to trust it with.

Hand it raw material and it shapes that faster than you can. Paste a competitor's homepage and pricing page in with a prompt like analyze their positioning and target segments, then flag where they claim an advantage we should counter, and you get a structured first-pass teardown in seconds. Ask for three positioning options using the competitive alternatives, unique attributes, and value, the components April Dunford's framework leans on, and it hands you drafts to react to faster than a whiteboard, which beats staring at a blank positioning statement. Drop in twenty call transcripts and it clusters the recurring pains and objections at a pace no human matches. Point it at closed-deal notes and it sorts them by win and loss reason into a first-pass categorization you sharpen instead of build. Even competitive research holds up as a starting point, which is why so many PMMs reach for ChatGPT or Claude to do it.

None of that is judgment. It is structuring, extraction, and a first draft. The PMM still decides which option is right, which theme matters, which objection is signal and which is noise. Use AI for the pass that used to eat an afternoon, and keep the decision for yourself.

Why do those prompts keep breaking down?

Every one of those prompts breaks at the same seam: it assumes knowledge the model does not have. Your real positioning, your actual ICP, a competitor's current pricing, the correction you made last week. These are not prompt-quality problems you can write your way out of. They are structural, and there are only a few of them.

The model has no grounding in your company, so the positioning it drafts is a plausible average any competitor could claim, and the persona it critiques your value prop as is a training-data stereotype, not the buyer you built from real calls. It has no memory across sessions, so last week's corrected framing is gone the moment you open a fresh chat. It cannot monitor anything, so it never notices a rival changed its pricing or a deal just closed, which is how a battle card goes stale three weeks before anyone checks it. And when it does not know, it does not stop. It invents a competitor feature, a pricing tier, a stat, and in front of a live deal that confident fabrication is a liability, not a draft. Not one of these is fixed by a better prompt.

After you have re-pasted the same positioning, the same ICP, and the same competitor notes into one fresh chat after another, you stop believing the problem is your prompt. The model is not short on intelligence. It is short on your context. Once that lands, the skill worth building stops being a cleverer prompt and becomes grounding the AI in what your company already knows, so nobody has to paste it in again.

— David Kolínek, Co-founder at Calven

There is a second, quieter failure underneath. Even a great prompt collapses the day you forget to paste the latest positioning, and the context does not travel. It lives in the PMM's head and nowhere else, so the sales rep, the BDR, and the CS manager who never had it get the ungrounded version by default.

What actually fixes it: grounding AI in one shared source of truth

If every failure traces to missing grounding, the fix is to stop supplying the context by hand and put it where the AI reads from every time. One place holding your real knowledge, positioning, messaging, personas, ICP, competitors, win-loss, so answers come from your truth rather than the training-data average. That shared base is also the memory the chat window never had: a correction made once persists, and every future answer inherits it.

Connecting any AI tool to that source is now standardized. Anthropic introduced the Model Context Protocol in late 2024, an open way to plug an AI tool into a knowledge source without a bespoke integration for each one, and within a year OpenAI, Google, and Microsoft had all adopted it. It is how a governed knowledge base becomes something any model can query, and querying verified context reins in a model's tendency to invent an answer.

This is the shape of what Calven builds. Its agents ground in a shared knowledge base, the One Brain, and answer from your company's real, current knowledge instead of the model's average. The Competitive Intelligence Agent tracks a rival's site and pricing and folds what changed into the dossier, so the teardown stays current. The Positioning Agent works from your actual market and audience intel, not a generic frame. It is the same principle Calven's own marketing runs on, agents working against a shared repository, the practice of running a team as marketing as code.

Why product marketing should own the company's AI knowledge

Product marketing should own it because it already owns the canonical version, and every other team's AI is describing the same company from a fragmented copy. Sales has one version of the pitch, support another, the website a third, a new hire's chatbot a fourth assembled from public scraps. Point an ungrounded model at that and it describes the company inconsistently, because there is no single truth to read. Enablement vendors like Highspot and Glean have named the same need from the sales side, then located that truth in the enablement platform. It does not live there. It lives in product marketing.

Look at what the function demonstrably owns: positioning, messaging, the personas and ICP built from real buyer conversations, the competitive intel, the win-loss patterns, the product truth. That is precisely the knowledge every team's AI needs and does not have, and product marketing is the only function that already maintains the correct version of it. Which makes the shared base a PMM asset, and making it AI-available a PMM responsibility. Not a stretch of the mandate. The most direct expression of it.

The objection writes itself: product marketing is already stretched thin, and this sounds like a second job on top of the first. It is not. It is the first job done once instead of every session, since the context a PMM re-pastes into chat after chat is the same context the base holds permanently. Governance cuts the same way: knowledge copied into a dozen tools nobody can audit is the exposure, and one governed source is the control. The first move is small. Put your own positioning and messaging where your AI reads them, start with a single agent, and let the rest of the company follow the proof.

What changes when every team works from the same brain

Once that knowledge is exposed through MCP, every team's AI answers from the same source instead of guessing, and the company stops contradicting itself. The content team's AI drafts on-brand because voice and messaging are inputs, not something it approximates. A sales rep's AI answers a competitive question with your real positioning and the current battle card, not a fabricated feature that falls apart on the call. A BDR's AI personalizes against the actual ICP instead of a generic segment. The product team grounds a roadmap call in real market and competitive intel rather than the loudest opinion in the room, and a CS manager's AI answers a renewal question with accurate product truth instead of last year's deck. Same brain, five teams, one company speaking with one voice.

Messaging moves, so holding that together is a continuity problem. When it changes, Calven flags the affected assets for review, so a battle card or landing page that has drifted off the current position gets caught instead of spreading the old line for another quarter. That is what turns a pile of connected agents into an AI product marketing platform rather than a folder of clever prompts.

Both jobs start with AI, and only one scales. The PMM who learns to prompt better optimizes a private tool that stops working the moment they forget to paste the context. The PMM who grounds the company's knowledge and exposes it becomes the reason every team sounds like the same company. Fluvio's 2026 research found 82% of product marketers now face an expectation to use AI, and 72% say it raised the output expected of them, per Fluvio's Product Marketing Hiring Trends Report. The bar rose. Prompting harder does not clear it. Owning the knowledge the whole company runs on does.

AI Can Run Your Win/Loss Interviews. It Can't Read Them for You.

David Kolinek — Fri, 17 Jul 2026 00:00:00 GMT

Standing up a voice agent that can call a lost buyer and hold a real conversation used to be a research-vendor engagement: a statement of work, a six-week timeline, an invoice. It is now an afternoon of wiring together three commodity parts. That collapse is why win/loss can finally run continuously instead of as a quarterly scramble. But the win isn't automating everything. It's drawing the line: automate the collection, transcription, and pattern extraction, and keep human the interpretation and the routing of each finding to its owner. This is the how. Why win/loss matters, and which questions to ask, live in the win/loss analysis guide.

Can AI actually run win/loss interviews now, and how hard is it to build one?

Yes, and the reason is unglamorous: the voice-agent stack turned into a commodity. A working interviewer is three parts in a loop, speech-to-text to hear the buyer, an LLM to decide what to ask next, and text-to-speech to answer, with turn-taking so it knows when the buyer has finished a thought. A non-engineer can wire one up in an afternoon. The horizontal platforms, from ElevenLabs to Vapi, Retell, and Bland, ship templates and a phone number you connect without code, and research-specific interviewers like Perspective AI, User Intuition, Outset, and Wynter package the same machinery for buyer conversations.

The durable point on cost is cents per minute. ElevenLabs' Conversational AI ran about $0.08 to $0.10 a minute as of early 2025, with the LLM billed separately (ElevenLabs, February 2025). Talk time is the cheap part. Add the outreach and retries it takes to land one completed conversation and vendor-published pricing still lands near $20 to $25 per completed interview, against the tens of thousands a managed program costs to cover a fraction of deals. Full ranges are in the companion guide.

Are buyers more candid speaking to an AI than typing to a form?

Often, yes, in two measurable ways: length and candor. When people speak an open answer instead of typing it, they say more. In a smartphone survey experiment on sensitive topics, spoken answers ran more than twice as long as typed ones and touched more topics (Höhne, Gavras & Claassen, 2024). And in a health-screening study, people who believed they were talking to a computer rather than a person reported lower fear of self-disclosure and lower impression management, and disclosed more (Lucas, Gratch, King & Morency, 2014). Take the human out of the loop and the reflex to manage what the other person thinks of you goes too, which is exactly what you're fighting when you ask a lost buyer why they chose someone else.

But there's a cost the vendors selling voice leave out. In that same Höhne study, 51% of the voice group abandoned the survey partway through, against 24% of the text group. Voice gets longer answers from the people who stay and pushes twice as many to quit. That trade-off is why the channel question has no clean answer.

Where AI interviews fall short, and where human interviews fall short too

AI isn't a free lunch. A recent evaluation found AI voice interviewers uneven: solid on structured questions, weaker on adaptive follow-up, reading emotion, and pushing for depth when an answer gets interesting (Tirumala et al., 2025). It's a fitness-for-purpose read, not a verdict that AI loses, and the same study finds these interviewers already outperform the automated phone systems research teams have used for years. Where a skilled human still wins is the unscripted turn, hearing the hesitation behind a polite answer and following it.

The part almost nobody says out loud is that human-led interviews aren't automatically better either. Moderators introduce bias: leading questions, unconscious cues, a warmer tone in the first interview than the tenth, and respondents manage their image harder in front of a person than a machine (social-desirability research, NCBI, 2022). So the honest difference isn't quality, it's cost. A human-led program is more expensive without being reliably better. The real decision isn't AI versus human in the abstract, but which buyer gets which.

There's no single right answer: it depends on your audience

No channel wins for every buyer, because comfort with talking to an AI isn't evenly spread. Voice is the least preferred way people interact with generative AI: no more than roughly 1 in 5 users in any generation call it their easiest method (PYMNTS Intelligence, 2025). Whatever a slick demo suggests, a large share of your buyers will never choose to talk to a bot, and the younger ones who reach for AI first are a reason to offer choice, not a rule to segment on.

Pair that with the Höhne breakoff and the design conclusion writes itself. Standardize on voice and you optimize richness on the deals you capture while shrinking how many you capture at all. Offer both a form and a voice option on the same program and let each buyer pick. Coverage beats richness on any single deal.

What to automate, and what to keep in human hands

Automate the operational and first-pass analytical layer. Keep the judgment layer human. The line is cleaner than most automate-everything pitches admit.

Hand to the machine	Keep in human hands
Scheduling and outreach on every closed deal	Designing the study: what you're actually trying to learn
Running the interview or survey at pipeline cadence	Interpreting what a pattern means, not just that it recurs
Transcription and first-pass theme coding	The high-stakes or executive conversation
Surfacing the patterns that repeat across deals	Routing each finding to the owner who can act on it

The collection and pattern work is what Calven's Win/Loss Agent and Voice of Customer Agent do. Standing up one interview is the afternoon project; running it on every closed deal and holding a single cumulative read on why you win and lose, tied to the real response counts and updated as each answer lands, is the part that doesn't assemble itself. The Win/Loss Agent runs the survey and the guided interview and keeps that rolling read; the Voice of Customer Agent mines your existing call transcripts for the same themes. The PMM keeps the interpretation and the routing, the same division AI is drawing across the rest of product marketing: the agent takes the grind, the PMM keeps the judgment.

Routing is where the judgment concentrates. The same surface pattern points to different owners. A recurring "we lost on the demo" can be a sales hard-skill gap enablement can train out, a soft-skill gap that belongs to sales leadership rather than a battle card, a real product gap for the roadmap, a messaging gap where the demo leads with the wrong thing, or an early-pipeline problem where the lead was never a fit, which belongs in your ideal customer profile, not your roadmap. Same five words from the buyer, five different owners, and reading a soft-skill gap as a product gap is the classic misroute. The machine can tell you the pattern is real and how many deals it spans. It can't tell you which of the five it is. That call is the job.

Running win/loss as a standing discipline, not a quarterly scramble

The payoff of cheap interviews isn't a thicker annual report. It's that win/loss stops being an event. When an interview costs cents and something reads every response as it lands, the program runs continuously, a standing instrument instead of a project that spikes before a board meeting and goes dark after.

Continuity only works if buyers talk, and the losses are where they clam up. A buyer who chose someone else rarely wants a call with the vendor they passed on, which is why in-house loss interviews are the hard ones to land, and why buyers are measurably more candid with a neutral party, the gap behind win/loss vendor Clozd's 70-versus-34 satisfaction split. A Calven Win/Loss Agent takes the rep who lost the deal out of the conversation, removing the sharpest source of face-saving even though the agent is still yours. It runs both the survey and the guided phone interview, so the buyer who would click a form and the one who would rather talk both have a way in. The questions it runs are in the win/loss template, and either way it's a recorded interview, carrying the same consent obligations as any other, handled up front.

The technology question is settled: cheap, continuous, candid enough. What's left is the operator's question, the one the machine can't answer. Which findings mean what, and who needs to hear them by Monday.

Competitor Analysis Template: 12 Fields and Where to Source Each One

David Kolinek — Mon, 06 Jul 2026 00:00:00 GMT

A competitor analysis template is the easiest deliverable in product marketing to build and the easiest to fill in badly. Twelve column headers in a spreadsheet is ten minutes of work. Then someone fills each cell with a guess, a copied marketing claim, or last quarter's pricing, the grid gets shared once, and it rots in a shared drive while the market moves underneath it. The template was never the hard part. Below is a template worth using, where to source every field, and how to keep the whole thing from going stale.

What is a competitor analysis template?

Two artifacts make up the template: a per-competitor profile (one per priority rival) and a you-vs-rivals comparison matrix, both ending in a mandatory "so what to action" column, with everything else optional. Underneath the format, the analysis itself is a structured, maintained view of who you compete with, how they position and price, where they win and where they're soft, and what each of those findings means for your own decisions.

The word "template" hides the work. A competitor analysis is a deliverable you produce and a discipline you keep, and most teams do the first and skip the second, which is why the average competitor doc describes a rival who no longer exists.

Table A: the per-competitor profile (12 fields).

#	Field	What good looks like
1	Name + tier	Direct, indirect, or aspirational. The tier decides how deep you go.
2	Company snapshot	Founded, HQ, headcount, funding stage, ownership. Signals of trajectory, not vanity metrics.
3	Target segment / ICP	Who they actually sell to best, which is usually narrower than they claim.
4	Product & key capabilities	Their real core capabilities mapped to buyer jobs, not a feature dump.
5	Pricing & packaging	Model, tiers, what's gated, published vs. "contact us."
6	Positioning & core claims	The category they claim and their differentiator, in their own words.
7	Strengths	Where they genuinely win, backed by reviews and win/loss, not assumption.
8	Weaknesses	Where they're soft, sourced from complaints and lost-to-them deals.
9	GTM motion	Sales-led, PLG, or channel. Where their pipeline comes from.
10	Recent moves	Funding, launches, repricing, key hires, messaging pivots. Dated.
11	So what to action	What it means for us, and the one decision it drives.
12	Sources + last updated	Where each claim came from, and when. Makes the doc auditable.

Table B: the comparison matrix. Rows are the dimensions buyers decide on, not every feature you ship. Columns are you plus each rival. Cells read "fully / partially / not supported," never a bare yes or no, and a final "why it matters to the buyer" column keeps the grid honest.

Dimension	You	Rival A	Rival B	Why it matters to the buyer
[Capability the buyer names]	Fully	Partially	Not supported	[The outcome they lose without it]

Free template: Competitor Analysis Template

A per-competitor profile and a you-vs-rivals matrix, with a prompt, a source, and a quality check under every field.

How do you do a competitive analysis?

Five steps. Define the decision the analysis serves, identify and tier the competitors, gather data against the profile fields, turn each finding into an action, and keep it current on a cadence. The method matters more than the grid, because the same twelve fields produce a decision tool or a dead spreadsheet depending on how you work through them.

Define the decision it serves. Analysis with no decision attached is a fifty-slide deck no one reads. Name the call it feeds first: a battle-card refresh, a positioning tweak, a pricing review.
Identify and tier the competitors as direct, indirect, or aspirational. Start from competitive alternatives, meaning what the buyer would use if you didn't exist. That is often a spreadsheet or nothing at all, not the funded startup you keep an eye on.
Gather data against the twelve profile fields, each from its right source (the sourcing table below).
Turn each finding into a "so what to action." A finding with no action is trivia.
Keep it current on a cadence set by how fast the field moves. Effort should follow tier: deep on the two or three rivals you lose to, light on the rest.

Step two is where most analyses go wrong before they start. The competitor to name is the one your buyer would pick instead of you, and for most B2B products that is the status quo, not a named rival. Get the set wrong and everything downstream inherits the error: a phantom rival on the tracker becomes a phantom rival in your positioning.

The first competitive alternative we need to beat is the status quo.

— April Dunford, Author, Obviously Awesome

Further reading: Obviously Awesome: How to Nail Product Positioning — April Dunford (Book)

The clearest thinking on why the competitor to name is the one your buyer would pick instead, including the status quo.

How do you fill in each field?

The source tells you where a field's data lives; the quality check tells you when you've filled it honestly. Most templates stop at the first. Three fields decide whether the whole profile is intelligence or fiction, and each has a check a rushed PMM skips.

Positioning & core claims: filled right when you can point to the source line. Capture the category and differentiator a rival leads with, verbatim, then check it against their own docs. A claim their documentation can't back up isn't a fact about their product. It's a battle-card opportunity.
Pricing & packaging: filled right when the gaps are marked, not guessed. When a rival hides enterprise pricing behind "contact us," the gap is the finding. A cell with an invented number in it is worse than an empty one, because it looks like knowledge.
Weaknesses: filled right when every entry traces to a review or a lost deal. This is the field teams fake, because the honest version means reading the reviews you'd rather skip and asking buyers why they picked someone else.

The weaknesses column is the one worth being strict about. Filled from imagination, it's how a battle card starts lying to your own reps, and they find out in front of the buyer.

— David Kolínek, Co-founder at Calven

Then the field that earns the whole exercise. The "so what to action" column is the line between a decision tool and a data-collection habit. Every row should end in one concrete move: a discovery question for reps, a claim to add to a battle card, a gap worth a positioning tweak. A profile with eleven fields full and the action column empty is a doc that gets built once and never opened again.

Where do you get competitor data?

Nine sources, each good for something specific and each with a trap. The template tells you what to record; the source tells you where it's true. Filling the grid from a single source, usually the competitor's own website, is how a profile ends up being their marketing copy retyped into your spreadsheet.

Source	Best for	The trap
Website + docs/changelog	Product, positioning, claims, launches	It's their story. Claims are not capabilities.
Pricing page	Model, packaging, what's gated	Often "contact us." Note the gap, don't guess.
Content & resources (blog, guides, webinars)	Who they target and how they position	Aspirational. Shows who they want, not always who they win.
Review sites (G2, Capterra, TrustRadius, Gartner Peer Insights)	Real strengths and weaknesses in customers' words	Read the three-star reviews, not the fives.
Win/loss + lost-deal reasons	Why buyers actually chose them over you	The CRM's competitor tag is unreliable. Get it from the buyer.
Sales-call transcripts (Gong, Chorus)	Competitive mentions, objections, how reps position	Volume. Impossible to read manually at scale.
Job postings	Where they're investing: sales vs. engineering hires	Directional, not definitive.
SERP / AI visibility	Their content and answer-engine footprint	Presence is not strength.
Funding / news	Trajectory and recent moves	Funding is not product quality.

Watch what a competitor publishes, not just what they build. The topics they invest content in, the personas their guides address, the pains their webinars lead with, are the clearest public signal of who they're really selling to and how they want to be seen. A vendor pouring resources into enterprise-security content is telling you where they're moving upmarket before their pricing page catches up.

Two rows carry the same warning: the truth is buried in more conversation than a human can read. Win/loss is the case in point. The competitor your CRM has tagged on a deal disagrees with what the buyer actually says in nearly 7 of 10 deals, an observation from win/loss vendor Clozd, which is why 36% of teams name win/loss insights from buyers and sellers as a top source in Crayon's 2026 State of Competitive Intelligence. The rep's guess and the buyer's reality are different data.

What are the most common competitor analysis mistakes?

Six, and most are variations on one root cause: collecting data without attaching a decision to it. Feature-checklist myopia treats the matrix as a race to the most checkmarks. Copying a competitor's claims uncritically poisons your battle cards with their marketing. Ignoring the status quo positions you against a rival the buyer was never choosing between.

Mistake	Why it's wrong
Feature-checklist myopia	Buyers decide on outcomes and fit, not checkmark count. A longer list is not a better product.
Copying competitor claims uncritically	Their marketing overstates. Treating claims as fact poisons your battle cards.
Ignoring the status quo	The real alternative is often "do nothing." Positioning against phantom rivals weakens you.
Trusting the CRM's competitor field	It disagrees with the buyer in nearly 7 of 10 deals. Get the real competitor from win/loss.
Data with no action	A grid full of findings and no "so what" is a spreadsheet, not intelligence.
Analyzing everyone equally	Effort should follow tier. A SWOT per rival across a full set is wasted work.

The most expensive mistake isn't on the list, because it happens after the analysis is done: the doc that rots. A profile built in Q1 and never touched again describes a competitor who repriced in Q2 and repositioned in Q3, and your reps are carrying it into live deals.

How often should you update a competitor analysis?

As often as your competitors move, which is continuously. A profile built once is usually wrong within a quarter. 57.5% of teams report more competitive deals than a year ago, per Crayon's 2026 State of Competitive Intelligence, and while that pressure rises, competitors reprice, relaunch, and reposition on their own schedule, not yours. Nearly 8 in 10 teams track 30 competitors or fewer, which is exactly why tiering matters: you cannot keep a hundred profiles current, so keep the ones you lose to current and let the rest age.

Maintenance is the real problem, not the template. The grid takes an afternoon. Rereading every competitor's pricing page, docs, and reviews every few weeks, folding the changes back into the right field, and re-checking the action column is the work that never ends, and the work most teams stop doing. This is where AI takes the grind while the judgment stays human.

That standing cost, the hours no one has and the doc that goes wrong, is the job the competitive intelligence agent Calven builds exists to carry. Today it identifies your direct and indirect competitors, writes the full per-competitor dossier, generates sourced battle cards, and watches competitor websites and docs, folding each change back into the right section instead of letting it age. But the template was never the hard part. Keeping it current, and turning every row into a decision, is.

Sales Calls Sharpen a Battle Card. They Don't Build One.

David Kolinek — Thu, 16 Jul 2026 00:00:00 GMT

A sales-call transcript captures a competitor the way a customer remembers them: as they were on the day that customer last looked. That memory is gold for one half of a battle card and wrong for the other. Point a pipeline from Gong straight into a card and you can auto-draft how a rival shows up in your deals, the objections, the exact buyer language. What you cannot auto-generate is what that rival charges, packages, or shipped last month. Transcripts sharpen a card. A separate live-research loop keeps it true.

What are sales-call transcripts actually the best source for?

Transcripts are the authoritative source for how a competitor shows up in a real deal: the words a buyer uses unprompted, the objection in the form it actually arrives, and which rivals reach the shortlist at all. A survey can tell you a competitor is "strong on integrations." A recording tells you the buyer said "we already run everything through their connectors and switching scares me." Only one of those is usable on a card.

They also tell you which competitors belong on it. As April Dunford puts it, a rival is not a true alternative if it doesn't show up on a shortlist, and calls are the evidence of who is actually on it. Timing is signal too: Gong Labs found competitors raised early in a cycle correlate with higher win rates than the same rival surfacing late. This is the work Calven's Voice of Customer Agent does against your recordings, pulling the buyer language, objections, and pains from each transcript into your why-won and why-lost analysis, where a buyer's stated reason beats a rep's guess.

Where does a transcript-built battle card get you wrong?

A transcript-built card gets the competitor's current state wrong, and current state is the fastest-moving part of a card. Pricing, packaging, a feature that shipped last month, an integration added or dropped: a customer describing a rival is describing the version they used, when they used it. A recent call with a current user can carry current signals. You just can't tell, from the transcript alone, how current any given mention is. Non-authoritative is not the same as wrong. It means you can't ship it unverified.

How fast is fast? PricingSaaS logged more than 1,800 pricing and packaging changes across the top 500 SaaS companies in 2025, about 3.6 per company, split 43% packaging, 40% pricing, and 17% usage limits, per Growth Unhinged, and Paddle finds the companies with the most pricing power re-evaluate roughly every three months. The fields a transcript is stalest on are the ones that move most. So the cost of a dated claim is not a quiet correction. A rep reads the card's "they can't do X" out loud, and the buyer, evaluating that rival precisely because it does X now, corrects them. As Crayon's own battlecard guidance puts it, when a rep gets called out for outdated information the trust that breaks is not only with the prospect, it is the trust sales had in you. A card is one wrong line away from never being opened again. The battle cards 101 guide has the full mechanic.

Which source should each battle-card field come from?

Source each field by how fast it decays. Language, objections, and how a competitor shows up come from transcripts, because those decay slowly and only a real deal reveals them. Pricing, packaging, features, and integrations come from live competitor research, because they move on the rival's cadence, not yours. Few templates make the split explicit, so here it is:

Battle-card field	Authoritative source	Why
Buyer objections and exact language	Sales-call transcripts	First-hand words; language decays slowly
How and when a competitor shows up	Sales-call transcripts	Only real deals reveal who's actually on the shortlist
Why you won or lost	Win/loss analysis and transcripts	The buyer's stated reason, not a rep's guess
Competitor pricing and packaging	Live competitor research	Fastest-moving fields; a call is a dated snapshot
Current features and integrations	Live competitor research	Ships and changes on the rival's cadence
Where they're genuinely strong	Live research reconciled with transcripts	The claim buyers test hardest, so it needs both halves
Your own differentiators and pricing	Your own product facts	Your side goes stale too, and you own the truth

I'll trust a transcript to tell me how a buyer talks about a competitor and which objections keep coming up. I won't let it decide what that competitor charges or what shipped last quarter. Those are the lines a rep says out loud in a live deal, and if the buyer already uses the product, a version-old claim is how you lose the room.

— David Kolínek, Co-founder at Calven

Every current-state row points at the same place: the live competitor analysis a card distills, refreshed against the competitor's primary sources, not a customer's memory of them.

How do you build the pipeline without shipping dated claims?

Extract the language and objections from your calls, draft the buyer-side half from them, then verify every fast-moving claim against the competitor's live site and docs before the card reaches a rep. Verification is the step, not a nicety. The version worth running this week: keep the fast-moving fields on a short checklist and re-check them only for the two or three rivals you actually lose to. Everything on that list is a claim a buyer can contradict in the room. Everything off it is safe to draft straight from the call.

Live pages settle one layer and not the one below it. Whether a feature exists is on the rival's site. How well it works, the throughput it holds, the rate limit its docs never advertise, is a claim the vendor's own marketing is the weakest source for. That depth comes from your own win/loss and deal evidence, not a page check. No single source is authoritative for the whole card. The tempting shortcut, pointing a raw model at a folder of transcripts, skips all of this: it drafts a clean card and inherits every version-bound error in those calls, stating the stale lines as confidently as the true ones. The useful shape of automation is narrower. Take the grind of extraction and drafting, and keep judgment human on what is true and what is sayable.

Why is keeping a battle card accurate genuinely hard?

Accuracy is a multi-source problem with no finish line, which is why effort alone never solves it. Three feeds, three cadences: buyer language from calls, which shifts slowly; competitor current state from the live web, which shifts fast; your own product facts, which you change and then forget the card still quotes. A card is true only while all three stay reconciled, and that reconciliation is the actual work, low-glory, never finished, the first thing a human week drops.

That is what a connected system of agents carries, one feed each. The Voice of Customer Agent reads the transcripts for buyer language, objections, and pains. The Competitive Intelligence Agent tracks each rival's site, docs, and pricing, keeps the dossier current, and generates the sourced battlecard from it. The Product Intelligence Agent keeps your own product facts canonical, so your side of the card doesn't go stale while you watch theirs. One brain holding three cadences at once. The template and the field list are the easy afternoon. The transcript gives you the buyer-language half, sharper than any other source can, and it cannot give you the current-state half. Build a card as though it can, and it ends up wrong in the one moment it was supposed to hold.

Message Map: What It Is and How to Build One (+Free Template)

David Kolinek — Sun, 05 Jul 2026 00:00:00 GMT

A message map is the document that turns your positioning into the words your whole go-to-market team reuses. The format is the easy part. A blank map takes ten minutes to draw, and every template site hands you one. The map is only as good as the buyer language inside it, and that is the part no format supplies. Below is a definition that resolves the map-versus-framework confusion, the components, a build method that starts where it should, a worked example, and a free template.

What is a message map?

A message map is an internal document that lays out a product's positioning as a hierarchy: one core message at the top, a handful of supporting pillars beneath it, and proof points under each, usually cut by buyer persona. It is a reference the whole team draws from, not customer-facing copy. Pragmatic Institute calls it an internal document that acts as a messaging structure for product communications. Crayon defines it as an overview of your company's positioning. Cascade Insights describes it as positioning laid out in the format of a map.

The three definitions agree on the load-bearing word: positioning. The map does not invent your message. It organizes the positioning decision you already made, and organizing is a core product marketing deliverable. The core message at the top is your positioning made concrete, turned from an internal strategy paragraph into a line a seller can actually say.

Map, framework, house, matrix: these are largely the same idea drawn different ways, and arguing about the name distracts from whether the language inside is right. The shared skeleton is a hierarchy of core message, pillars, and proof points. Cascade Insights has the clearest breakdown of the near-synonyms:

Term	What it emphasizes (per Cascade Insights)
Message map	Positioning laid out in the format of a map: a visual hierarchy
Messaging framework	Value props, pillars, and proof points in a grid-like format
Message house	Messaging organized in the outline of a house
Message architecture	The key messages each persona or stakeholder needs to hear
Messaging matrix	Distinct messaging by persona and channel

A map implies the visual, hierarchical snapshot; a framework is usually the more detailed grid it summarizes. Pick the layout your team will actually keep open.

What goes in a message map?

Four layers, top to bottom: a core message, message pillars, proof points, and a persona cut. Each layer supports the one above it, which is what makes it a map rather than a list of claims.

Component	What it is	The discipline
Core message	The single sentence everything ladders up to. Your positioning made concrete.	One sentence. Not a paragraph.
Message pillars	The value-prop themes that support the core message.	Three is the disciplined default.
Proof points	The evidence under each pillar: metrics, customer quotes, case studies, third-party validation.	Evidence, not a feature list.
Persona cut	How each pillar reframes for a specific buyer.	The map's job, not an afterthought.

Three pillars is the well-sourced default. Pragmatic Institute builds its guide around the rule of three, and three is what most disciplined teams ship. Some go to five. Treat that as a ceiling you reach when the product genuinely needs it, not a target, because a fourth weak pillar dilutes the three strong ones. Proof points are where maps most often go soft: Pragmatic recommends ordering them by revenue impact, which forces the question a feature list never asks, namely which evidence actually moves a deal.

How to build a message map, step by step

Build it from the buyer's words in, not the org chart out. The order below is deliberate: the first step does most of the work, and skipping it is why so many maps look complete and still fail.

Let your customer do the work for you – steal their language (in the nicest possible way, of course).

— Emma Stratton, Founder and CEO, Punchy

Start with customer language. Before you write a single pillar, pull the exact words buyers use from real conversations: call transcripts, win/loss notes, customer interviews, review sites. Not the words your team wishes buyers used. The words they actually said, verbatim, including the unglamorous ones. Crayon's method starts with personas for the same reason. Everything downstream inherits the quality of what you capture here.
Lock the core message. Your positioning made concrete, in one sentence a seller can say out loud without wincing.
Draw the three pillars from the themes in the language. Cluster what buyers actually care about into three, and let those clusters become your pillars. Not your feature categories. Not your engineering roadmap.
Attach proof points. Evidence under each pillar, ordered by what moves revenue.
Cut by persona. Reframe each pillar in the terms the specific buyer uses, because the CFO and the practitioner value the same product for different reasons and in different words.

A map assembled from internal jargon is decoration. It will be well-formatted, internally consistent, and useless, because the buyer never sees themselves in it.

Step one is also where AI helps most. Rather than combing through call recordings by hand, point a model at your transcripts through a sales-intelligence MCP and have it surface the buyer's exact phrasing, clustered into the themes that become your pillars. A prompt to start from:

Extract buyer language from sales calls

# Role
You're a product marketing analyst building a message map from the voice of the customer. You work only from what buyers actually said, in their own words, never ours.

# Context
I'm writing the pillars of a message map and need the raw language first: the exact words buyers use for their problems, the outcomes they want, and the alternatives they weigh. I'll cluster your output into three to five themes and turn those into pillars, so the honesty of the source language matters more than polish.

# Source data
Pull the transcripts of our sales calls across the whole company, the most recent 12 months (or the last few hundred calls, whichever is larger). Favor prospect and customer conversations over internal ones. [Use your sales-intelligence MCP here, e.g. "use the Gong MCP" or "use the Chorus MCP", to fetch the transcripts.]

# Task
From those transcripts, pull verbatim buyer language into four buckets:
1. Problem: how buyers describe the pain they're trying to solve, including the blunt, off-message wording.
2. Outcome: the result they say they want, in their terms.
3. Objections: their hesitations, risks, and reasons to say no.
4. Alternatives: what they call the competitors, incumbents, and the status quo (spreadsheets, "doing nothing").

# Output format
Give me one table per bucket, with three columns: the verbatim quote; the speaker's role and segment (for example "VP Marketing, mid-market"); and the call date. Below the tables, cluster the recurring phrases into three to five themes, and for each theme list the three most representative quotes. End with one line on which theme came up most often.

# Rules
Quote verbatim, never paraphrase or clean up grammar. Never invent a phrase, a speaker, or a source that isn't in the transcripts, and if a bucket is thin, say so. Prefer patterns that many buyers repeat over a single vivid quote.

Swap the bracketed placeholder for whichever tool holds your calls. The same prompt works against any sales-intelligence MCP, so "use the Gong MCP" becomes "use the Chorus MCP", and so on.

Free template: Message Map Template

A blank message map skeleton: the core message, three pillars with a supporting line each, proof points under every pillar, a persona cut, and a prompt plus a quality check under each blank.

The template gives you the empty hierarchy. What it can't give you is the positioning decision the core message is supposed to make concrete. That decision has to be made before the map has anything true to organize, and it is a different, harder piece of work.

Further reading: Obviously Awesome: How to Nail Product Positioning — April Dunford (Book)

The positioning decision that has to be settled before a message map has anything true to organize.

A filled-in message map example (reconstructed from public messaging)

No company publishes its internal message map. Like a positioning statement, it is a strategy artifact, not marketing copy. What follows is a reconstruction assembled only from one company's public homepage and product messaging, not a leaked internal document. It shows the shape filled in: a core message, three pillars, proof points, and one persona cut.

Illustrative example: Notion

Layer	Content
Core message	The connected workspace where teams and agents think together.
Pillar 1: One workspace, not twelve tools	Docs, wikis, projects, and databases in one place, so context stops fragmenting across apps. Proof: replaces separate doc, wiki, and project tools `[illustrative]`.
Pillar 2: Bend it to how your team works	Flexible building blocks assemble into whatever a team needs instead of a rigid template. Proof: databases, custom views, and templates configurable without code `[illustrative]`.
Pillar 3: AI that knows your context	AI that answers from your own workspace, not the open internet. Proof: search and generation grounded in your existing docs and data `[illustrative]`.
Persona cut (team lead)	Stop paying for and stitching together five tools. One workspace your team will actually keep open.

Read it top to bottom and the ladder holds: each proof point supports its pillar, each pillar supports the core message, and the persona cut restates the whole thing in a buyer's terms. Every metric here is marked [illustrative] on purpose. In a real map, each one would carry a source or be cut, because an unsourced proof point is a claim, not proof.

Who captures the buyer language, if not you?

Every step of the method above depends on one thing the format can't supply: the exact language your buyers use, captured from real conversations rather than invented at a whiteboard. Getting that language out of hours of call recordings and interview notes and into a usable record is the job of Calven's Voice of Customer Agent. Today it reads uploaded transcripts, interviews, and win/loss notes and folds the signals into an evidence-backed knowledge base with the quotes attributed to their source. The skill that captures the exact terminology buyers use, the one most directly tied to filling a map, is on the near roadmap. A map also goes stale the moment the product, market, or competitors move, so that capture has to be continuous, the kind of monitoring AI is starting to take on in product marketing, not the annual refresh that always slips and leaves sellers quoting pillars the buyer no longer recognizes.

That record is meant to feed the Messaging Agent Calven is building to write and maintain the messaging framework, grounded in real customer language instead of internal jargon. That agent isn't shipped yet, and saying so matters: the buyer's language is the work, and it's worth capturing well whether an agent or a person does the listening.

Value Proposition Template + 8 B2B Examples, Honestly Scored

David Kolinek — Tue, 14 Jul 2026 00:00:00 GMT

Sit through enough messaging reviews and you learn to spot the value proposition that lost. It lists eight benefits and differentiates on none. Every stakeholder got their favorite feature in, someone softened the verbs, and what survived is a line that could belong to any of four competitors. It reads fine. It commits to nothing. Below is the Value Proposition Canvas that forces you past that failure, the frameworks behind it, a five-test rubric to grade the result, and eight live B2B value propositions run through it.

What is a value proposition?

A value proposition is a clear statement of the specific outcome a specific buyer gets, and why your offering delivers it better than the next best alternative. It is the external promise a customer reads, distinct from three things it gets confused with. It is not a positioning statement, which is the internal strategic decision behind it. It is not a tagline, which is a memory device, not a claim you can check. And "unique value proposition," or UVP, is the same artifact under another name.

Writing, sharpening, and defending that promise is core product marketing work. The trouble is that most value propositions fail at exactly the point they should do the most: they try to say everything, and end up saying nothing that separates you from the alternative.

Why do most value propositions fail?

Most value propositions fail because they are all-benefits lists that skip differentiation and proof. The definitive framework here comes from Anderson, Narus and van Rossum, whose 2006 paper Customer Value Propositions in Business Markets sorts B2B value propositions into three kinds, and almost no ranking template cites it.

The first kind is all benefits: list everything the product might do for anyone. It feels safe and thorough. It is neither. The authors call the trap it invites benefit assertion: claiming advantages for features the customer doesn't actually value. The second kind, favorable points of difference, is better. It says how you differ from the next best alternative. The third, and the one to aim for, is resonating focus: the one or two points of difference the buyer values most, substantiated with evidence.

Every PMM has watched the failure happen in a room. A value proposition gets built by consensus. It has to survive every stakeholder's pet feature, so it absorbs all of them, and a line that offends no one differentiates against no one. Resonating focus is the discipline of cutting back to what a buyer would actually change vendors over, then proving it.

The value proposition canvas: your fill-in template

The template most teams actually need is the Value Proposition Canvas, from Alexander Osterwalder and Strategyzer. It is a two-sided worksheet that forces the fit the last section demanded. You map what the buyer wants on one side and what you offer on the other, and the value proposition falls out of the match instead of a brainstorm.

One side is the Customer Profile, and you fill in three things about the buyer:

Jobs: what they are trying to get done, practical, social, and emotional.
Pains: the frustrations, risks, and obstacles in the way.
Gains: the outcomes and benefits they actually want.

The other side is the Value Map, your offering set against that profile:

Products and services: what you give them.
Pain relievers: how you remove the specific pains above.
Gain creators: how you produce the specific gains above.

Free template: Value Proposition Canvas Template

A copy-ready value proposition canvas: the customer profile (jobs, pains, gains) and the value map (products, pain relievers, gain creators), with a prompt under every block and a fit check.

You have fit when the Value Map answers the most important items in the Customer Profile. Not all of them, the two or three the buyer would switch vendors over. That one rule is what turns the canvas from a form into a discipline, and it is exactly where most teams cheat: they list every pain reliever they can think of instead of matching the handful that decide the deal.

The danger is treating the blanks as fields with obvious answers. April Dunford, whose book on positioning is the closest thing the discipline has to a standard reference, is blunt about filling them in on autopilot:

The exercise assumes that there is only one answer for each of the blanks. You simply 'know' what it is.

— April Dunford, Author, Obviously Awesome

The point holds. Most products serve more than one job, fight more than one alternative, and relieve different pains for different segments, so the canvas is worth filling per audience, from evidence, not once from memory. The "for whom" comes straight from your ideal customer profile, and the finished value proposition is one part of a larger structure: it is the roof, and the pillars and proof live in your message map.

Once the canvas has its matched pairs, distill the strongest one into a line a buyer reads. Steve Blank's formula is the cleanest: We help [X] do [Y] by doing [Z], the buyer from your Customer Profile, the gain they want, and the pain reliever only you have. A compliance-automation tool might land on: we help lean security teams pass SOC 2 by automating evidence collection. If you want the alternative named too, Geoffrey Moore's fuller "For [target] who [need], our [product] is a [category] that [benefit], unlike [alternative]" spells out the differentiation. Either way, the canvas did the thinking. The sentence just reports it.

Further reading: Value Proposition Design: How to Create Products and Services Customers Want — Alexander Osterwalder et al. (Book)

The canonical book behind the Value Proposition Canvas, and the one title dedicated to value propositions specifically rather than positioning broadly.

What makes a good value proposition?

A good value proposition passes five tests, each traceable to a sourced framework rather than to opinion. Run your draft through them before it goes anywhere near a homepage.

Test	The question	Grounded in
1. Specific target	Is it for a named buyer, not "businesses" or "everyone"?	Moore's "for [target]"; Blank's X; your ICP
2. Specific outcome	Does it name the result the buyer gets, not the feature that produces it?	Moore's benefit; Blank's Y; Anderson's "value"
3. Point of difference	Does it say why you, over the next best alternative?	Moore's "unlike"; Anderson's favorable points of difference
4. Believable, not asserted	Is the claim substantiated, or is it benefit assertion?	Anderson, Narus & van Rossum's core warning
5. Clear in five seconds	Would a cold visitor understand what it is and for whom, fast?	Flint McGlaughlin's clarity principle, popularized by Peep Laja

Tests 3 and 4 are where all-benefits lists die, and they are the two most homepages skip. Anything can claim an outcome. The line that earns the click names a difference and gives the buyer a reason to believe it.

8 B2B value proposition examples, scored

Here are eight live B2B homepage lines run through the rubric. The point of grading real, admired companies is that even category-definers don't all pass, and the honest calls teach more than another list of tips.

Company	Hero line	Rubric read
Stripe	`Financial infrastructure to grow your revenue.`	Passes cleanly. Names the outcome (test 2), and the subhead's "first transaction to your billionth" does real work on believability (test 4). Outcome-led.
Linear	`The product development system for teams and agents`	Strong on target and category (tests 1, 3), but slightly mechanism-led: the subhead leans on "the AI era" rather than the result (test 2).
Gong	`#1 AI OS for Revenue Teams`	Bold category and a specific audience (tests 1, 3). The "#1" is unproven proof, the exact benefit assertion the rubric flags (test 4).
Vanta	`Trust is everything`	Memorable and brand-led, but fails test 5. A cold visitor can't tell what it is, which also starves tests 1 and 2.
Notion	`Where teams and agents Think together.`	Vision-led headline; the subhead carries the clarity (test 5) and the outcome (test 2). Strong once you read the second line.
Figma	`The intelligent canvas for infinite creativity`	The headline alone fails test 5. The subhead ("one workspace for your entire product development process") rescues it on target and outcome. A headline-vs-subhead lesson.
Datadog	`AI-Powered Observability and Security`	Mechanism-led. It leads with the "how," not the result, so it stumbles on test 2 and reads as benefit assertion (test 4). Graded honestly, the anti-pattern.
Intercom	`The only helpdesk designed for the AI Agent era`	The "only" is a genuine point of difference (test 3), and naming Fin as the differentiator supports test 4. Long, but it earns the length.

The pattern is consistent: outcome-led beats mechanism-led. Stripe tells you what you get; Datadog tells you what powers it, and asks you to work out the rest. Two of the eight stumble hardest, Vanta on clarity and Datadog on outcome, and both are companies with real products and real customers, which is the point. A famous logo does not exempt a line from the five-second test. For more annotated work, see B2B messaging examples, which goes deeper on the lines above and below the fold.

Keeping a value proposition sharp

A value proposition is only as good as the buyer language and proof underneath it, and it decays as the product and the competitors move. The line that resonated at launch describes a market that no longer exists eighteen months later. Keeping it sharp is ongoing work, not a one-time fill-in.

The harder part is aim. A value proposition earns its keep when it names the one pain a specific audience feels most, not the five that are merely real. Resonating focus is a targeting problem before it is a wording problem: find the single most important thing for the buyer you are writing to, then say it plainly. Get that wrong and no amount of polish saves the line.

Finding it is what Calven's Voice of Customer agent does. It reads where the pain actually shows up, across sales-call transcripts, community forums, review sites, and open online research, and surfaces which problems come up most and weigh heaviest for a given audience. The Messaging Agent grounds the value proposition in that evidence: it drafts and pressure-tests the line against the buyer's own language, and gives candid feedback against a framework instead of inventing benefits to fill a blank. The template gets you a draft. What decides whether the draft is any good is the same thing that always has: the most important pain for the audience you want, said in their words, with the proof behind it.

Value Proposition Canvas Template

David Kolinek — Tue, 14 Jul 2026 00:00:00 GMT

This is the fill-in canvas from our article on the value proposition template and 8 scored B2B examples. The article explains how to fill the canvas, why fit is the deciding rule, and grades eight live B2B homepage lines against a five-test rubric. Read it there for the full method.

What's inside

The Customer Profile side: the buyer's jobs, pains, and gains, with a prompt under each so you fill it from evidence, not memory.
The Value Map side: your products and services, pain relievers, and gain creators, mapped against that profile.
A fit check and a distill-to-one-line step, so the canvas ends in a value proposition a buyer can read, not just a full worksheet.

Free template: Value Proposition Canvas Template

A copy-ready value proposition canvas: the customer profile (jobs, pains, gains) and the value map (products, pain relievers, gain creators), with a prompt under every block and a fit check.

10 Competitive Advantage Examples: Real Moats vs Fake Ones

David Kolinek — Mon, 13 Jul 2026 00:00:00 GMT

A competitive advantage is what lets a company deliver the same benefit as its rivals at a lower cost, or a benefit they can't match at all. That's Michael Porter's definition, and it holds up. The word most companies get wrong is the next one: sustainable. An advantage only lasts if it's hard to copy, and most of what ends up on a positioning slide fails that part. Great support, a cleaner interface, an AI label on the box. Most of it holds for a quarter, then a competitor matches it. What follows separates the advantages that hold from the ones that only sound like they do.

What is a competitive advantage, and what is an economic moat?

Put plainly, the advantage is any edge that lets a firm earn better returns than its rivals. An economic moat is that same edge made durable, an advantage that lasts. The distinction is the whole game, because most advantages are temporary and only a structural few survive contact with a competitor who wants your customers.

Porter sorted the ways to win into three generic strategies: cost leadership (be the low-cost producer), differentiation (offer something rivals don't), and focus (own a narrow segment better than anyone). Pick one. Companies that try to be all three end up stuck in the middle, cost-competitive with no one and differentiated for no one.

The moat vocabulary came later, popularized by Warren Buffett and turned into a working framework by Morningstar, whose analysts still rate companies by it. Morningstar names five sources of a moat and grades each by how long it holds: wide (decades), narrow (about ten years), or none.

A truly great business must have an enduring "moat" that protects excellent returns on invested capital. The dynamics of capitalism guarantee that competitors will repeatedly assault any business "castle" that is earning high returns.

— Warren Buffett, 2007 Berkshire Hathaway shareholder letter

Moat source	What it is
Intangible assets	Brands, patents, and licenses rivals can't legally or credibly copy.
Switching costs	The pain, risk, and cost a customer takes on to leave you.
Network effect	Each new user makes the product more valuable to everyone already on it.
Cost advantage	A structural ability to produce cheaper, not a discount you're running this quarter.
Efficient scale	A market just big enough to profitably support the players already in it.

10 competitive advantages that actually hold up

Almost every real advantage traces back to one of those five sources, plus one more that Hamilton Helmer names and Morningstar doesn't: the cornered resource. The table classifies ten companies by their advantage's source and Porter's lens. Not one is "great execution" or "customer obsession." Each is structural.

Company	Moat source	Porter lens	Why it's durable
Visa / Mastercard	Network effect	Differentiation	More cardholders pull in more merchants and vice versa. A rival has to win both sides at once.
LinkedIn	Network effect, switching	Differentiation	The professional graph gets more useful with every member. Hard to displace even with a better product.
Salesforce	Switching costs	Differentiation	Deeply customized and integrated into daily operations. Ripping it out puts the business at risk.
SAP / Oracle ERP	Switching costs	Differentiation, focus	ERP is the financial and operational core. Replacement is a multi-year, high-risk project.
Microsoft	Switching, network, brand, scale	Differentiation	Office and Azure lock-in, ecosystem effects, brand, and scale. Four sources reinforcing each other.
Costco	Cost advantage, efficient scale	Cost leadership	Membership fees fund near-break-even pricing. Volume wins supplier discounts no one else gets.
ASML	Cornered resource	Focus	The only maker of the EUV machines advanced chips need. Three decades of know-how behind it.
Apple	Brand, ecosystem switching	Differentiation	Brand commands a premium. Hardware, software, and services lock in raise the cost of leaving.
Coca-Cola	Brand (intangible)	Differentiation	A century-old brand plus distribution reach Buffett named directly as a moat.
Amazon	Cost advantage, efficient scale	Cost leadership	Fulfillment scale economics, and AWS switching costs on top. The durable part is cost, not culture.

Three of these teach the idea best. LinkedIn's is the network effect at its purest, the kind Morningstar treats as a textbook moat: the graph gets more useful with every member, which is why a better-designed rival still can't pull people off it. Its own 2011 SEC S-1 filing named the network as the core asset. ASML sits on a cornered resource, the sole supplier of the extreme ultraviolet lithography machines advanced chips require, backed by three decades of know-how and a supplier ecosystem Works in Progress documented in detail that no rival can reconstitute.

Amazon is the honest one. The durable advantage is cost and efficient scale, in retail fulfillment and in AWS switching costs, not the "customer obsession" the annual letter leads with. Customer obsession is a culture. Culture is not a moat.

This is where the usual example lists go soft. The vendor listicles from Klue and Crayon reach for the same handful of consumer brands, Netflix and Red Bull and Warby Parker and Supreme, and stop at the story. None tests whether the advantage would survive a competitor who wanted it. Judging that is the actual work, and it starts with a clear-eyed competitor analysis.

The honesty test: is it a moat or a mirage?

There's a one-line test for whether a claimed advantage is real. Could a competitor credibly claim the same thing in a deal tomorrow? If yes, it's table stakes, not a moat. Three frameworks turn that instinct into a checklist.

Jay Barney's VRIO test is the cleanest. Run any advantage through four questions:

Valuable. Does the buyer actually pay more, or stay longer, because of it?
Rare. Do few or no competitors already offer the same thing?
Costly to imitate. Could a rival copy it inside a release cycle or a fiscal year? If they could, it isn't a moat.
Organized to exploit it. Is the company built to keep delivering it, not just claim it once?

Further reading: 7 Powers: The Foundations of Business Strategy — Hamilton Helmer (Book)

The clearest modern framework for telling a durable advantage from a temporary one, built entirely on the benefit-plus-barrier test this section runs on.

Helmer's 7 Powers frames the same idea as a benefit paired with a barrier: an advantage counts only if it helps the customer and rivals can't cross to match it on decent economics. A benefit with no barrier is just a feature, and it helps until someone copies it. April Dunford adds the qualifier worth keeping close: a feature is differentiated only relative to the competitive alternatives, so a capability you're proud of stops being a differentiator the moment two rivals ship it.

Run the test on a real B2B claim and it does visible work. "The easiest onboarding in the category" fails on costly-to-imitate: a rival redesigns its setup flow next release and the edge is gone. "Two years of the customer's workflow data and every integration already wired in" passes, because a competitor can match the feature but not the switching cost the buyer already paid.

This is the test worth running before anything reaches a positioning statement. Calven's Positioning Agent runs that differentiation analysis directly, scoring each claimed differentiator on how unique it is and how much buyers actually weight it.

The advantages that aren't: great support, ease of use, and an AI label

Great customer support is not a competitive advantage. It's a point of parity, the price of being considered rather than a reason to be chosen. Every serious competitor claims it, most deliver it adequately, and a rival can match it by hiring. It fails VRIO on rare and on costly-to-imitate in the same breath.

Ease of use fails the same way, just slower. A cleaner interface is real and worth having, and it's copyable in a release cycle. An AI label is the 2025 version of the mistake. Nearly every B2B product now claims it, which by definition makes it table stakes and not a differentiator. The market has noticed too. Morningstar downgraded six wide-moat software companies, Salesforce among them, citing the uncertainty AI injects into their future returns. When AI is the thing putting moats in question, an AI label is a strange thing to call one.

A single feature rivals will ship next quarter is not a moat either. Neither is first-mover advantage on its own. It's only durable if it converts into a network effect, switching costs, or scale before the fast followers arrive. Sponsorships and referral programs are tactics. They fill a pipeline. They don't defend one.

Most teams build their competitive advantage like a list, and the instinct is always to make it longer. Every strength goes on: support, uptime, ease of use, now AI. The trap is that your competitors can say all of those too, and something true of everyone defends nothing. The discipline isn't adding more claims. It's cutting the list to the two or three a rival genuinely can't match, and building the whole message on those.

— David Kolínek, Co-founder at Calven

The advantages that survive this test are the few worth building B2B messaging around and putting on a battle card. The rest are noise dressed as strategy.

Advantages decay, so keep checking whether yours is still real

A moat is a verb, not a noun. Buffett's own test isn't whether an advantage exists today but whether it's getting wider or narrower, and the answer moves as competitors do. Salesforce's switching costs are exactly what the AI question now puts in play. An advantage you verified once and stopped watching is one you're no longer sure you have.

So the honest read is a standing practice, not a one-time audit. It depends on knowing what rivals actually ship, price, and claim right now, not what they did last year. Calven's Competitive Intelligence Agent tracks every competitor's site, docs, pricing, and announcements and keeps the dossiers and win/loss evidence current, so the moment a rival crosses one of your moats, you know before the deal does.

Run the test on your own list this quarter. The advantages that survive it are the ones worth defending. The ones that don't were never moats, and the sooner you stop calling them that, the sooner you can go build one that is.

Direct vs Indirect Competitors: 10 B2B Examples, Classified

David Kolinek — Sun, 12 Jul 2026 00:00:00 GMT

A direct competitor sells the same kind of product to the same buyer. An indirect competitor solves the same customer need with a different kind of product. The difference is the product, not the need. Most product marketing teams stop at those two. A complete picture has four types, direct, indirect, replacement, and potential, plus the one almost every competitor list leaves off: the status quo, the buyer who keeps the spreadsheet, builds it in-house, or does nothing at all.

What is a direct competitor?

Direct competitors are the rivals you meet head-to-head: same product category, same target market, same customer need. They show up in most of your deals, matched feature for feature, and they are the names your buyer lists when weighing who else to evaluate.

In CRM, that's Salesforce, HubSpot, and Microsoft Dynamics 365. In cloud data warehouses, Snowflake, Databricks, and Google BigQuery. These are the rivals you meet in a live deal, so they're the ones that earn a battle card: a rep needs an answer in the room.

What is an indirect competitor, and how is it different from a substitute?

An indirect competitor solves the same customer need with a different kind of product. Same buyer, same job, different category, so the need gets met a way that never appears on a feature-for-feature grid. A team choosing Slack isn't only weighing messaging tools. Email competes for the same attention, and so does a project tool with chat built in. Neither is a messaging product; both are indirect competitors.

Where does a substitute fit? Sources split: some fold replacement into indirect, others break it out. The simpler read is to treat a replacement as indirect competition escalated. Both meet the same need with a different product; the difference is scope. Indirect competitors steal deals you knew about; replacement competitors eliminate deals you assumed would always exist.

What are replacement and substitute competitors?

A replacement or substitute competitor solves the need in a fundamentally different way and can make your whole category unnecessary for a segment of the market. This is Porter's threat of substitutes from the Five Forces (Michael Porter, HBR, 1979): a substitute meets the same need through a different mechanism, and it caps what your whole category can charge.

The clearest illustration is old. Netflix didn't out-store Blockbuster; streaming replaced the idea of driving somewhere to rent a disc, and the category went with it. The current B2B version is in spend management: SAP Concur and Expensify sell paid expense software, while card-first platforms like Ramp give expense management away free, funded by card interchange. A replacement competitor doesn't win the bake-off. It makes paying for the category look unnecessary. That points at the alternative no comparison grid contains at all.

If companies included nonconsumption as part of their competition, they would quickly find that it has the biggest share of many markets.

— Efosa Ojomo, Senior Research Fellow, Christensen Institute

Why the status quo is the competitor most teams never track

The status quo is the competitor most teams never put on the tracker: the buyer keeping a spreadsheet, building the thing in-house, or doing nothing at all. It has no website to monitor and no battle card to write, yet it wins more B2B deals than any named rival. Teams build the competitor list by scanning their own product category, so a non-product alternative never makes it on. A spreadsheet has no category. It just wins the deal.

April Dunford has made a version of this point for years, and it belongs at the front of any positioning work.

What would a customer do if your offering didn't exist? Sometimes the answer to that question is 'Do nothing.'

— April Dunford, Author, Obviously Awesome

Two adjacent traps distort the set from the other side. Phantom competitors are the rivals sales obsesses over in Slack and QBRs that barely register in real deal data. The trap is prioritizing competitors because they come up in meetings, not because they move pipeline. Potential competitors are the opposite: an adjacent, well-capitalized player not in your market yet, but one bundle away from being in it. Neither belongs in the same tier as a rival you lose deals to today.

The do-nothing loss rarely announces itself. It surfaces in win/loss analysis when you read why a deal stalled and find the buyer didn't pick a competitor. They picked staying put.

10 B2B examples of direct, indirect, and replacement competitors

Here are ten named B2B companies classified across direct, indirect, and replacement or do-nothing competition. Read the last column, not the first. The direct rivals are easy and mostly known. The replacement and do-nothing column is where deals actually leak, and it's the one most competitor docs leave blank.

Company	Category	Direct	Indirect	Replacement / substitute / do-nothing
Salesforce	CRM	HubSpot, Microsoft Dynamics 365, Zoho, Pipedrive	Airtable or Notion used as a light CRM; LinkedIn Sales Navigator	Spreadsheets; reps' inboxes and memory (the real status quo)
HubSpot	Marketing automation + CRM	Marketo (Adobe), ActiveCampaign, Salesforce Marketing Cloud	An agency doing the work; WordPress plus plugins	Sending campaigns by hand from Gmail and a list
Zoom	Video conferencing	Microsoft Teams, Google Meet, Cisco Webex	Slack or async chat, a phone call, meeting in person	Loom-style async video; "just send an email"
Slack	Team messaging	Microsoft Teams	Email; project tools with built-in chat (Asana, Basecamp)	Email threads, the status quo Slack displaced
Notion	Docs / knowledge / wiki	Confluence (Atlassian), Coda (now a Grammarly product), Microsoft Loop	Google Docs plus Drive, Evernote	Info living in people's heads and Slack; nothing written down
Figma	Product / UI design	Sketch, Framer, Penpot	Canva, Miro, Adobe Photoshop or Illustrator	Whiteboard or paper mockups; wireframes in Keynote
Snowflake	Cloud data warehouse	Databricks, Google BigQuery, Amazon Redshift, Microsoft Fabric	Analytics left inside an operational Postgres or MySQL	An in-house stack on open source; data left unqueried
Gong	Conversation intelligence	Chorus (ZoomInfo), Clari Copilot, Salesloft	CRM notes; a BI dashboard on pipeline	Reps typing notes by hand; nobody reviewing calls at all
DocuSign	E-signature	Adobe Acrobat Sign, Dropbox Sign, PandaDoc	Emailing a PDF back and forth	Print, wet-ink sign, scan; native in-app signing
Stripe	Payments infrastructure	Adyen, Braintree (PayPal), Checkout.com	A traditional bank merchant account	Building on raw bank APIs in-house; invoicing by hand

Two rows carry the nuance. Zoom's sharpest rival, Microsoft Teams, is a direct competitor that behaves like a replacement, because Teams ships bundled into a Microsoft 365 seat the buyer already pays for. The call isn't better; it's free and already there. Notion versus Google Docs is genuinely arguable: a direct rival for some buyers, an indirect one for others, and the status quo they never leave for plenty. That the call is contestable is the point. Classification is a judgment about the buyer, not a label on the product.

How do you classify your own competitors?

Ten examples are a pattern to copy, not a method. To classify your own market, run every candidate through four yes-or-no questions. The pattern of answers tells you the type, and the type tells you how much attention it's owed.

Type	Same product / category?	Same customer & market?	Same customer need / job?	Threatens the deal or the category?
Direct	Yes	Yes	Yes	The deal, head-to-head
Indirect	No	Yes	Yes	The deal, via a different product
Replacement / substitute	No (a different approach)	Often, sometimes a new segment	Yes	The category (can make you unnecessary)
Potential / future	Not yet	Could be	Could be	Future deals and the category
Status quo / do-nothing	No product at all	Yes, it's your buyer	Yes, met "well enough" already	The deal. Often the biggest loss column

The rule underneath the grid is one line: classify by the customer's need, not by your product category. If two offerings satisfy the same buyer job, they compete, however different they look. A spreadsheet and a CRM share no features and every buyer. Once the set is tiered, the direct rivals feed your battle cards and the whole set feeds a competitor analysis you actually maintain.

Keeping your competitor set current

A competitor set built once is usually wrong within a quarter. A new entrant appears, a replacement starts giving your category away free, a phantom rival everyone feared turns out not to take deals. The classification is the easy half. Keeping it true is the standing work, and it decides whether your competitive intelligence tools earn their place.

That is the job Calven's competitive intelligence agent carries. From a name and a URL it identifies your direct and indirect competitors, writes the per-competitor dossier, arms sales with battle cards, then watches competitor sites and docs and folds changes back into the set. Autonomy is a setting: human-in-the-loop by default, approvals you can turn off as trust builds. The do-nothing loss enters from the other side, where the win/loss agent reads lost-deal reasons, so a buyer who kept the spreadsheet joins the set like any named rival.

Classify by the need, tier by deal impact, and keep the do-nothing column honest. The competitor that beats you most often is rarely the one on the battle card. It's the one that was never on the list.

B2B Messaging Examples: 8 Taglines AI Is Quietly Erasing

David Kolinek — Sat, 11 Jul 2026 00:00:00 GMT

The textbook example of category creation in B2B is Gong. Every positioning talk cites it: the company that coined "Revenue Intelligence" and got buyers to judge an entire market by rules it wrote. Open gong.io in July 2026 and that phrase is nowhere on the homepage. The hero now reads "#1 AI OS for Revenue Teams." The line that built the category is gone, replaced by a claim four of its competitors could run tomorrow.

Good B2B messaging does five things at once. It names a category so a buyer knows what kind of thing you are. It argues against a status quo. It sells an outcome, not a feature. It says something a competitor couldn't paste onto their own site unchanged. And it claims something only you can. Most 2026 homepages manage one of the five: they say "AI." Only 6% of B2B SaaS marketing leaders rate their own brand "very distinctive," per Wynter's 2025 branding survey of 100 leaders at companies over $50M. The eight B2B messaging examples below show you why.

What separates good B2B messaging from generic messaging

Good B2B messaging is judged, not admired. Run any homepage line through a five-point lens and you get a score instead of a vibe: does it set a category, name an enemy, lead with an outcome, survive the paste test, and claim something only this company can? Clever is easy. Passing all five is rare, and rarer since every homepage started saying the same three-letter word.

The lens sits on top of positioning, not in place of it. Positioning is the strategic call about what market you're in and who you beat. Messaging is the words that carry it. April Dunford, who wrote the discipline's standard reference, keeps the two apart on purpose:

Positioning is not the same thing as messaging. It isn't a tag line.

— April Dunford, Author, Obviously Awesome

Five questions, asked of any line:

Clear category or frame of reference. Do you know instantly what kind of thing this is and who it's for? "Payments infrastructure" answers that in two words.
A named enemy. Is there a status quo it argues against? The strongest B2B messaging has a villain.
Outcome, not feature. Does it sell the promised land, or list the apparatus?
The paste test. Could a direct competitor put this exact line on their own homepage and have it still be true? If yes, it fails. This is the one most 2026 messaging flunks.
Only-ness. Different from the paste test, which asks whether the words are distinctive: this asks whether the claim underneath is. Does it assert something only true of you, that a rival couldn't honestly say even if they tried?

A line doesn't need a perfect five to work. But the great ones score four or five, the forgettable ones score one, and the gap is almost always the paste test. The same components run through Calven's message map and positioning statement guides, so this isn't a scorecard invented for a blog post. It's the same discipline, written down two ways.

Category creation, and the companies abandoning it

Category creation is the most celebrated move in B2B positioning, and in 2026 the companies famous for it are walking away from their own categories. Naming a category and a villain is how Gong, HubSpot, Salesforce, and Drift became case studies. Most of those lines are now diluted or dead.

Gong is the sharpest example, and the cleanest place to run the whole lens on one line instead of asserting a score. As Pitch Kitchen's teardown of B2B taglines documents, "Revenue Intelligence" named a category and set the rules buyers used to judge every rival: category, enemy, and only-ness, three clear passes. Now grade the live hero, "#1 AI OS for Revenue Teams." Category, yes. Named enemy, no. Outcome, implied at best. Paste test, failed, since any rival could run it tomorrow. Only-ness, gone, because "#1" is a ranking anyone claims and "AI OS" is a term everyone is racing toward. One pass out of five, where the retired line scored four. That was a deliberate, tested call, not a mistake. It just spends distinctiveness, which is the one thing the lens measures.

HubSpot did the same thing more slowly. "Inbound Marketing" was a movement, not a campaign. It named an enemy every marketer already resented, interruption, and pre-sold the buyer before they saw a price. As of mid-2026 the homepage leads with "HubSpot Agentic Customer Platform." It's accurate. It's also a sentence a dozen platforms could run without changing a word.

Here's the blunt part. The two most-cited named-enemy plays in the history of the craft are Salesforce's "No Software," which cast on-premise software as the villain and cloud as the escape, and Drift's "Conversational Marketing," which put web forms and five-day reply times on trial. One is a retired campaign. The other belonged to a company that was absorbed into Salesloft, so the category outlived the brand that coined it. When PMM teaches category creation, half its best examples are historical artifacts.

The sharp wedge versus the broad AI claim

The paste test is where most rebrands fail. A specific claim you can defend beats a broad promise every time, because the broad promise is available to everyone who can spell "AI." Superhuman and Slack both traded a sharp wedge for a soft one. Stripe changed its wording and kept its score, which is the tell that the problem was never change.

Superhuman is the cleanest before-and-after. Its wedge was "the fastest email experience ever made," a single, specific, believable claim that beat any list of features. The current hero reads "Superpowers, everywhere you work." Its Mail product still carries a sharp line, "The most productive email app ever made," but the front door now promises everything and locates nothing. Speed was a wedge. Superpowers is a mood.

Slack sold an outcome long before "outcome-led messaging" was a phrase people put on slides. "Be less busy" promised a calmer workday, not chat features, and Stewart Butterfield's internal "We Don't Sell Saddles Here" memo is still one of the most-shared artifacts in product marketing. The homepage now says "All your people and AI agents working together." True, timely, and interchangeable with half its market.

Every single website that I read, I have no idea what any of these companies do.

— Anthony Pierri, Co-founder, FletchPMM

Stripe is the counter-note that proves the thesis is anti-generic, not anti-change. Its classic line, "Payments infrastructure for the internet," was seven words that answered who it's for and what it does. The current stripe.com headline reads "Financial infrastructure to grow your revenue." Broader scope, same rigor: it still owns a category noun and now leads with an outcome. The wording moved and the score held. Changing your messaging isn't the risk. Changing it into everyone else's is.

Who kept their edge through the AI pivot

The sameness isn't inevitable. A company can lead with AI and still pass the lens, which means the ones dissolving into "AI platform" chose to. Two current examples make the AI pivot without erasing themselves.

Intercom leads with "The only helpdesk designed for the AI Agent era." The word doing the work is only. It's a straight only-ness claim, criterion five, and a rival can't copy it without the sentence simply being false for them. High score, current line, AI front and center. The pivot didn't cost the edge.

Linear reads "The product development system for teams and agents." The tell is system, not tool, a deliberate category signal that says this is opinionated infrastructure, not another app in the stack. It absorbed the AI era into a frame it already owned instead of replacing its frame with the era.

Both prove the point the erased taglines make in reverse. If Intercom and Linear can name AI and still say something only they could say, the generic homepage isn't a market condition. It's a choice, usually made by the company least able to see it.

The hard part isn't spotting good messaging. It's writing your own.

Scoring someone else's homepage takes five minutes. Scoring your own honestly is the hard part, because you're too close to it and you default to your feature list. Distinctiveness in 2026 doesn't come from a better AI claim. It comes from your buyers' exact words.

That reframe matters because it moves the problem to the right desk. As Anthony Pierri of FletchPMM puts it, "this is not a copywriting challenge; it's primarily a product marketing challenge." The fix isn't a wittier hero line. It's the discipline underneath the line.

Leading with "AI platform" feels like the safe move in 2026 precisely because everyone is making it, which is exactly what makes it the wrong one. A buyer reads "AI" on your homepage and learns nothing they haven't already read on ten other tabs that morning. The only line that holds up is the one a competitor can't copy, and that almost always turns out to be the one written in your customers' language instead of yours.

— David Kolínek, Co-founder at Calven

The language that passes the paste test already exists. It's in your win-loss interviews and your call transcripts, in the way customers describe the before and after in words no competitor would think to use. Pulling that evidence into something usable is exactly the kind of work AI is starting to change in product marketing: Calven's Voice of Customer Agent reads your call transcripts and interviews into an evidence base of how buyers actually talk, and its Messaging Agent turns that evidence into a messaging framework and gives you candid, honest feedback on your own lines, so a claim that fails the paste test gets caught before it reaches your homepage.

The homepages that survive the AI wave won't be the ones that shout "AI" loudest. They'll be the ones that still pass the paste test when everyone else's says the same thing.

Marketing as Code: Not a Move to GitHub. A Move to Agents.

David Kolinek — Fri, 10 Jul 2026 00:00:00 GMT

This article was drafted inside a Git repository, the same kind of shared, change-tracked folder that software teams keep their code in. An AI agent wrote the first draft, working from a set of plain text files. No content management system was involved, and the moment the draft was approved, it published itself. That way of working has a name, and it is becoming one of the most asked-about ideas in marketing. Almost every article written about it gets the reason wrong.

Marketing as code is the practice of running a product marketing team out of a shared repository. The team's knowledge, its brand and voice guidelines, its messaging, its campaign briefs, and even the instructions that tell its AI agents how to work, all live as plain text files that anyone can read and edit. Agents draft and operate against those files. It borrows a way of working software teams have used for over a decade, and adds the part that is new: agents that do the work. The repository is not the point. The agents are.

That last line is where most of the conversation loses the plot. The popular framing is that marketing teams are "moving to GitHub," as if the shared folder itself were the upgrade. It is not. Move a marketing team into a repository and give them nothing else, and all you have done is hand a group of writers a tool built for engineers and made their week harder.

So is this just docs as code for marketers?

Almost, and the "almost" is the whole story. Marketing as code takes a way of working software teams already trust, keeping their words in plain text, tracking every change, reviewing before publishing, and points it at marketing's material instead of a product's manuals. The genuinely new part is not the shared folder. It is the agents that read all that knowledge and produce and ship the work.

The workflow itself has a decade of history. Engineers have long kept their documentation as plain text files right alongside their code, reviewed the same way the code is. The Write the Docs community named this "docs as code" around 2015, and teams like GitLab run their own manuals this way, treating a first draft of the docs as part of finishing any new feature. What none of it had was agents, because until recently there were none worth the trouble.

That is the line marketing as code crosses. The brand guidelines are text. The messaging is text. The instructions for each agent, what it does, what it is allowed to touch, how it should sound, are text too. A little of it is genuine code, the small scripts that generate social images, but most of it is plain writing a marketer can read and change. This article is the proof: written by agents that are themselves described in plain text, reviewed and published by the same repository it is describing.

How do you run a martech tool without ever opening it?

You pay for the tool and let an agent operate it for you. Buy access to the service, not a seat for a person. The agent works the tool directly, pulls back exactly what it needs, and nobody on the team ever logs into that tool's dashboard again. The tool still does its job. The thing sitting in front of it is now an agent, not a person.

The future of martech is headless. People stop touching the tools at all. You pay for a service, an API key, and your agents operate it in the background, surfacing only what you asked them to surface. The dashboard you were supposed to open every morning becomes something no one on the team ever logs into again.

— David Kolínek, Co-founder at Calven

This is not science fiction, and it is not one company's private bet. The connections that let an agent operate an outside tool are already an industry standard. Anthropic, the maker of Claude, introduced one such standard, the Model Context Protocol, in late 2024, and within a year the biggest AI companies, OpenAI, Google, and Microsoft among them, had all adopted it. The plumbing exists. What is new is pointing it at a marketing team's tools.

The examples are ordinary. To size up search demand, an agent can pull the numbers straight from a data provider and drop the results into a file, for a tiny fraction of what a monthly per-seat subscription costs. To keep the customer database current, an agent can add, update, and look up records on its own rather than a person clicking through screens. In both cases an agent is doing the work a person used to do by hand, in the background, and reporting back.

There is a cost argument buried in here too. Marketers already use only about 42% of what their software can do, down from 58% a few years earlier, according to Gartner. A stack that agents operate changes what you are paying for: capability by the request, not a login you keep forgetting to use.

If it's this good, why isn't moving to GitHub the win?

Because the repository does nothing on its own. Move a marketing team into GitHub and change nothing else, and you have added process without adding any ability. Branches and reviews and approvals do not make a single piece of work better. The agents do. Most of the conversation has the cause and effect exactly backwards.

The solution is the agents. Fine-tuned skills with a stack of MCPs and CLIs they can actually operate. GitHub is just what makes that work: somewhere versioned, shared, and deployable for the knowledge to live. Moving a marketing team into a repo does not, on its own, make anything better. The repo is the floor you build on, not the reason you build.

— David Kolínek, Co-founder at Calven

The repository does earn its place, just not as the hero. It is where a whole team, and every agent, works from the same current source instead of a dozen slightly different copies. It keeps a history of every change, so you can see what moved and undo it if it was wrong. And it is what lets the approved work publish itself, with no one copying files around by hand. Those are real reasons to be there. They are also the ceiling of what the repository does by itself.

Where does marketing as code fit, and is this what a GTM engineer does?

It is the ground a GTM engineer stands on. The industry has been naming these pieces one at a time. GTM engineering is a new job title, the person who wires modern tools and AI into how a company goes to market. Marketing as code is the foundation that role works on.

The role is arriving fast. The company Clay coined the term "GTM engineer" in 2023, and it has spread through fast-growing software companies to roughly 100 job openings a month by early 2026. The pay is real and still being worked out: a 2026 study of 228 of these practitioners put the typical US salary around $135,000, while noting the market has not figured out how to price the role yet. Most of that energy has gone to sales operations so far. Product marketing is next, because the job is already getting more technical, and marketing as code is the ground it will grow on. You do not start with the whole thing. You start with one agent pointed at your own brand and messaging files, writing against them. The rest is what that grows into.

What does a marketing team running in a repo actually look like?

Like Calven's own marketing, which runs this way today. The website has no content management system and no database behind it. Every article is a plain text file kept in the repository, checked automatically for mistakes, and published the moment it is approved. Agents write the work and the repository ships it. Here is the honest version of what that looks like in practice.

The content is written by a chain of agents, and it starts with a person. Someone on the team sets the direction: the topic, the angle, what the piece needs to do. An agent takes that brief and turns it into clear instructions for the rest of the chain. From there, one agent researches the subject and produces a sourced brief, another shapes the outline, another writes the draft, and then a fact-checker, an editor, and a panel of agents role-playing the target buyer all go over it. Nothing publishes until a person approves it. This article went through exactly that chain. A person decided what to say and whether it was good enough. The agents did the heavy lifting in between, and they are themselves plain text files anyone on the team can open and change, not a black box.

Research and admin work run the same way. To size up search demand, an agent pulls the numbers straight from DataForSEO for a fraction of a cent per query and saves them to a file the team keeps and reuses, currently a list of 143 tracked keywords, instead of paying for a seat in a tool like Semrush. To keep the CRM current, an agent runs every change through Attio, our customer database, adding records, updating them, looking things up, so no one has to open the Attio screens at all.

Reporting works this way too, and this is where it gets genuinely useful. Instead of anyone opening an analytics dashboard, an agent reads the website's numbers from PostHog and posts a short weekly summary to the team's Slack: how many people visited, where they came from, how many signed up. The report comes to the team, rather than the team going to find it.

One thing is worth saying plainly, because it would be easy to misread. These are Calven's own internal agents, the way the team works day to day, not the product the company sells. The product is a different set of agents built for product marketers: competitive intelligence, positioning, and ideal customer profile agents. Everything above is proof that the approach works, not a tour of features.

When should a marketing team NOT do this?

When you are reaching for GitHub as a magic fix, which is the exact mistake this whole argument warns against. A repository adds real work to set up and learn, and it only pays that back when the agents, and the people who can direct them, are already in place. Without that, it is effort for its own sake.

There is a simpler starting point, and there is nothing wrong with it. If your team is not ready to build all of this, you can get a lot of the benefit from a tool like Claude's Cowork: you teach an AI a few reusable skills, point it at your brand and messaging, and let it help with the work, all without a repository or anything technical. For many teams that is the smart way to begin, and it may be all they ever need. The full move described here only makes sense when two things are true: you have someone technical on the team, and the people on it actually want to go through the change. And it is a real change, with new tools to learn and a genuine learning curve. If the appetite is not there, forcing it will cost more than it returns.

A couple of other cases argue against it. A small team that publishes rarely and runs a handful of channels will spend more time climbing the learning curve than it ever saves. And work that is not really words or settings, the live event, the customer dinner, the brand film, does not get better by living in a repository. Marketing as code is for the part of the job that is knowledge and repeatable work. That is a large part of it now. It is not all of it.

The shift worth taking in is not that marketing is moving to GitHub. It is that marketing is gaining agents that can build and run against its own knowledge, and the repository is simply the place that knowledge can be trusted, tracked, and shipped. Calven's own marketing works this way, and the product the team is launching is those product-marketing agents, not the internal ones shown here. Adopt the repository and never the agents, and you get all of the extra work and none of the payoff. The agents are the point.

Competitive Intelligence Tools: 7 Buyer Criteria the Listicles Skip

David Kolinek — Thu, 09 Jul 2026 00:00:00 GMT

Nearly every "best competitive intelligence tools" list scores the same five things: how many sources a tool monitors, how it alerts, where it hosts battle cards, what it integrates with, and what it charges. All five measure collection. Collection stopped being the hard part years ago. The criterion that actually separates one CI tool from another is what happens after a signal is detected, whether it reaches the battle card, the messaging, and the rep before the call, and almost no list scores that. One disclosure up front: Calven builds in this category, so treat what follows as a checklist to apply yourself, not a ranking with us at the top.

Why do "best competitive intelligence tools" lists rank the wrong things?

Because they all score the same commoditized surface: sources monitored, alerting, card hosting, integrations, price. Every one of those measures collection, the one thing every serious tool now does well. None of them measures what happens after a signal is detected, which is the only place tools meaningfully differ.

That after-detection work is where the value sits, and where every tool still struggles. A signal that a competitor repriced is worthless until it reaches the competitor analysis, the card, and the rep in the deal. Here is the checklist: seven criteria, the exact question to put to a vendor, and what a strong answer sounds like so a demo cannot talk past it.

Criterion	The question to put to the vendor	What a strong answer looks like, and the red flag
Collection vs. propagation	"When a competitor moves, does the tool push the change into my battle cards and messaging, or just alert me it happened?"	Ask them to show the signal landing inside an updated artifact, not an inbox. The red flag is a demo that ends at the alert, with "and your team takes it from there."
Relevance triage (the "so what")	"Of everything you detect in a week, how does the tool decide which few actually matter to me, and to which deals?"	Look for scoring tied to your competitors, your segments, and your open pipeline. If the answer is "you set up filters and keywords," the triage is still your job, just with extra setup.
Connection to the GTM surface	"Can the tool read my call recordings, cards, positioning, and personas, or does it only watch external sources?"	The best answer reaches into your win/loss calls and your live artifacts. A feed that sees only competitor websites is watching the least valuable source you have.
Agent / programmatic interface	"Can an AI agent read the data, and can it write a change back, and into what: your own cards, or my actual messaging?"	Read access through an MCP server or API is now the baseline. The differentiator is write-back that reaches the artifacts your reps use, not just the vendor's own dashboard.
Autonomy / human-in-the-loop	"Can I choose how much the agent does on its own, from approving every write to letting it propagate a change end to end?"	You want autonomy you can dial: approvals on while you build trust, off once you have it. Be wary of both extremes, a tool that can never act on its own and one that acts with no way to keep a human in the loop.
Time-to-propagate	"From the moment you detect a change to the moment my reps see it in the card, how long, and what has to happen in between?"	Count the manual hand-offs in the answer. Every step between detection and the field is where speed dies, and speed is what tracks to revenue.
True cost of ownership	"Beyond the subscription, who on my team authors and maintains the artifacts once you have detected the change?"	If the honest answer is "a dedicated owner, a day or two a week," add that salary to the quote. A tool that removes collection work but adds propagation work has not saved you anything.

Collection is solved. Propagation is the gap.

Collection has been automated for years. What no tool automates is moving a detected signal out of the CI silo and into the artifacts a seller actually uses: the messaging, the positioning, the personas, the live card. Those artifacts live outside the CI tool, so a person still carries the signal across by hand.

The cost of that manual carry is not small, and it compounds. Teams that share CI weekly or faster report revenue impact at 79%, against 41% for teams that share monthly or slower, per Crayon's 2026 State of Competitive Intelligence. Speed of propagation, not volume of collection, is what tracks to the number. And propagation is exactly the work that slips: 58% of CI pros say keeping battle cards and content current is a struggle, in Crayon's 2024 report. The subscription buys the detection. The struggle is everything after it.

That is why the subscription is usually the smaller line in the budget. The larger one is the person who reads the alerts, decides which few matter, and rewrites the card, the messaging, and the deck to match. Detection is cheap now, and the judgment and propagation around it are the standing cost that never shows up on the order form.

At Ataccama we paid north of $30,000 a year for a leading CI platform, and my team still spent one to two person-days every week doing the part we thought we were buying. When a competitor shipped a new feature, someone had to work out what it meant for us, brief the sales team, and update the battle card, again and again. The tool automated the collection. It never touched the propagation. And the people stuck with that, week after week, hated it.

— David Kolínek, Co-founder at Calven, former VP of Product Marketing at Ataccama

The best competitor signal now lives in your win/loss calls

The richest competitor signal is no longer on the competitor's website. It is in your own win/loss calls, where a buyer names the rival, repeats their pitch, and voices the objection your card should already answer. A CI tool that only watches external surfaces is instrumented on the wrong side of the deal.

Survey data backs that up. Internal knowledge and call recordings are now the top-cited CI source at 54%, ahead of competitor websites at 48%, in Crayon's 2026 State of Competitive Intelligence, and 46% of teams now run a Gong-style call tool for compete specifically. Win/loss insights rank third at 36%, above most of the external monitoring that CI tools were built to do. The signal has moved inside the building, and most tools are still pointed outward.

So one evaluation question cuts deep: does the tool see your call transcripts, or is it blind to the best source you have? A CI feed that cannot read the conversation where the competitor was actually named is watching press releases while the real intelligence sits in a Gong library it will never open.

Can an AI agent update your battle cards, not just read them?

Reading is the easy half, and it is largely solved. The incumbents now expose their competitive data to AI agents through Model Context Protocol servers, so an agent can pull your CI on demand. What separates tools is what the agent does next. Can it write the change back into the battle card and get it to the field, or does it hand you a summary and leave the update to you?

So the real thing to evaluate is control. The strongest setup lets you decide how much the agent does on its own: start with a human approving every write while you build trust, then hand over more of the last mile as the agent earns it, up to carrying a change end to end with no one in the loop. What you want is not an agent that reads, but an agent that can update your live assets, with the autonomy set wherever your confidence currently sits. That test belongs in the same conversation as the rest of your product marketing stack.

What does competitive intelligence software really cost?

An independent pricing tracker puts subscriptions commonly at $20,000 to $40,000 a year, scaling past $80,000 to $100,000 for mid-market and enterprise, per Vendr's crowd-sourced pricing data. That is the number on the order form. It is the smaller of the two costs you are signing up for.

The larger one does not appear on any invoice: the dedicated owner who authors and maintains every artifact the tool detected but did not update. That is the one-to-two person-days a week, every week, that the propagation work actually takes. Budget for it, or the tool becomes a very expensive alerting system whose output nobody has time to act on. Total cost of ownership for a CI tool is the license plus the labor, and the labor is usually the bigger half.

That gap is the reason Calven exists, and you should hold it to these seven criteria like any other tool. Its competitive intelligence agent reads the same win/loss transcript the win/loss agent reads, drafts the change into the battle card and messaging it should update, and has the persona agents validate that draft. One brain, not a feed bolted to a silo. The autonomy is yours to set: keep a human approving every change while you build confidence, then turn the approvals off and let the agent carry a signal from the call to the card on its own. Buy for propagation. Collection was never the hard part.

Win/Loss Analysis: 12 Questions, 2 About Features

David Kolinek — Wed, 08 Jul 2026 00:00:00 GMT

The reason your CRM gives for a lost deal, "price" or "missing feature", is usually the smallest and most convenient part of the truth. A rep picks it from a dropdown on the way to the next opportunity. Win/loss analysis is the discipline of going back to the buyer after a deal closes, won or lost, to learn why it really went the way it did, then acting on the patterns across deals. Done well, it covers the whole sales act: the competitive set, the sales experience, pricing and procurement, not only which features were missing. The proof is in the survey itself.

Why do B2B deals actually get lost?

Often not to a competitor, and often not to a feature. Between 40 and 60 percent of B2B deals end in no decision at all, and of those, roughly 56 percent are driven by buyer indecision, a fear of making the wrong call, rather than a genuine preference for the status quo. That finding comes from Matthew Dixon and Ted McKenna's analysis of more than 2.5 million recorded sales conversations in The JOLT Effect. The deal you think you lost to a missing integration was, more likely, a buyer who never got comfortable enough to sign anything.

Even among the deals that do close, the reason lives across the whole buying experience. Which vendors the buyer weighed, and how you actually compared. Whether the sales team knew the product cold and spoke the buyer's industry. How pricing landed against the runner-up, and whether procurement was smooth or a fight. The one-line reason the buyer would give a peer over coffee. A loss is usually about one of those five things, and only one of them is the product.

Read a loss as a feature request and you spend the next quarter building something that was never the reason, while the actual cause, a weak second call or a licensing model finance hated, goes unexamined.

Further reading: The JOLT Effect: How High Performers Overcome Customer Indecision — Matthew Dixon and Ted McKenna (Book)

The canonical, data-backed case that most losses are indecision and risk, not a feature gap.

What questions should a win/loss survey ask?

Five areas, in this order: the competitive set and how you compared, product fit, the sales team's performance, pricing and contract experience, and one open question on why. The bank below is the survey Calven's Win/Loss Agent actually runs, not a generic sample list. Count the questions and the argument makes itself.

Table A: the win survey (12 questions).

Section	Question	What it uncovers	Type
Competition	Besides [Your Company], which vendors did you evaluate?	The real competitive set, in the buyer's words	Vendor picker
Competition	How did [Your Company] compare to each vendor?	Where you won and lost against each rival, side by side	Comparison table
Product	Which features did you like the most?	The specific capabilities that pulled the deal in	Open text
Product	Which features were missing or did you like least?	Gaps that nearly cost you a deal you still won	Open text
Sales Team	How would you rate your experience with the sales team?	Whether the team, not the product, closed it	Rating 1-4
Sales Team	How would you rate the team's technical expertise?	Whether buyers trusted you on the hard questions	Rating 1-4
Sales Team	How would you rate the team's industry expertise?	Whether the team spoke the buyer's language	Rating 1-4
Sales Team	Overall, how would you rate the sales process?	The overall friction of buying from you	Rating 1-4
Pricing & Legal	How did pricing compare to the second-best vendor?	Whether you won or lost on price, not guesswork	Single-select
Pricing & Legal	What feedback do you have on the licensing model?	Where packaging helped or nearly blocked the deal	Open text
Pricing & Legal	How smooth was the contract negotiation?	Whether procurement was easy or a near-miss	Rating 1-4
Closing	In short, why did you pick [Your Company]?	The buyer's own one-line reason, unprompted	Open text

Now the math. Of those 12 questions, exactly two ask about product features: the open-text questions on features liked and features missing. The other ten interrogate the competitive set, the sales team, pricing, procurement, and the buyer's own summary. Build your survey to answer "which features were we missing" and you have designed it to miss almost everything that decides a deal.

The competitive set questions are the ones most teams skip and most regret skipping, because they name who you actually lose to and where. And when a loss traces back to a buyer who was never a fit in the first place, that belongs in your ideal customer profile, not your roadmap.

Same shape, trimmed: the loss survey turns the competition section into a direct question about who won.

Table B: the loss survey (also 12 questions, deltas from the win survey).

Section	Question	What it uncovers	Type
Competition	Besides [Your Company], which vendors were you evaluating?	The set you actually lost inside	Vendor picker
Competition	Which of those vendors did you ultimately choose?	Who beat you, named	Single-select
Competition	How did [Your Company] compare to each of those vendors?	Where the winner was stronger, item by item	Comparison table
Product	Which features did you like the most?	What still landed, even in a loss	Open text
Product	What features did [Your Company] lack versus the winner?	The real product gap, if there was one	Open text
Sales Team	The same four 1-4 ratings as the win survey	Whether the sell, not the product, lost it	Rating 1-4
Pricing & Legal	How did pricing compare to the vendor you chose?	Whether price was a factor or an excuse	Single-select
Pricing & Legal	What feedback do you have on the licensing model?	Whether packaging pushed them away	Open text
Closing	Why did you not select [Your Company]?	The buyer's own reason, in one line	Open text

Treat this as a starting point, not gospel. Every company is different, and some of these questions will matter less to you: maybe your buyers never test the sales team's technical depth, or licensing feedback is moot because you sell one standard license. Cut what does not apply and add what does. What carries across is the shape, the types of question worth asking, not the exact wording of each one.

The bank is also deliberately short. You could ask about onboarding expectations, champion strength, timeline, and budget authority, and each would tell you something. But completion falls as a survey grows: Survicate's 2025 benchmark shows median B2B completion dropping from roughly 87 percent on the shortest surveys toward the high 70s as length climbs. A survey nobody finishes teaches nothing, so this one stays trimmed to the load-bearing questions.

Free template: Win/Loss Survey Question Bank

The full win and loss surveys, five sections each, ready to copy and send.

Collection is the easy part. Analysis is the hard part.

Gathering responses is trivial. The value is created after, when you aggregate across deals, surface the themes that repeat, and push them to the teams that can act. One deal is an anecdote. The pattern across forty is a decision, and most programs never get to the pattern.

The evidence is blunt about it. Only about 39 percent of companies run an ongoing, cross-functional win-loss program, per Clozd's 2025 State of Win-Loss report; the rest do it once and file the deck. That same report finds 63 percent of companies see a win-rate increase from win-loss, rising to 84 percent for programs that have run two or more years. On average, only about 40 percent of deals get analyzed at all, per Klue's 2025 survey of 313 leaders. The insight exists. It is sitting in a folder nobody opens.

Treat completed surveys as a live repository instead, and put analysis on top of it. Calven's Win/Loss Agent is one way to do that: it runs the survey as a standing discipline and keeps a single cumulative read on why you win, why you lose, and the themes that recur, updated as each new response lands. It ties every claim back to the actual response counts, so a theme it reports maps to a real number of deals rather than a hunch, and thin evidence gets flagged instead of dressed up as a trend. From there the finding goes where it can act: a repeated objection sharpens your message map, a pattern of losing to one rival feeds the battle card, a win reason that keeps recurring belongs in your positioning.

Can AI run win/loss interviews?

Yes, and the economics changed recently enough that most teams have not caught up. Conversational voice AI tools now let a company stand up its own interviewer for cents per minute, work that used to require a research vendor. A buyer who opts in speaks their answer instead of typing it, and spoken answers to open questions run roughly twice as long and cover more ground than typed ones, per Höhne and colleagues' 2024 study. Buyers are also more forthcoming with a neutral interviewer than with the vendor who just lost or won the deal.

That neutrality matters most on the deals you least control. If you run win/loss yourself, expect the loss interviews to be the hard part: a buyer who chose someone else rarely wants a call with the vendor they passed on. Win interviews are easier, since a happy buyer will usually talk. For the losses, a neutral party is often the only way to get the conversation at all, whether that is dedicated win/loss software, an outsourced program, or an agent like Calven's Win/Loss Agent, which runs both the survey and the interview and reads as neutral because it is not the rep who just lost the deal.

There is an honest trade-off worth naming. Typed answers are more precise, more comparable across deals, and easier to count. Spoken answers win on depth. The best programs offer both and let the respondent choose, because a buyer who would click through a short form may not want a call, and coverage matters more than richness on any single deal.

Approach	Typical cost	Deal coverage	Answer depth
Managed / outsourced program	~$25K to $80K per year	A small share of deals	Deep, human-led
Win-loss software	~$15K to $50K+ per year	Broader, survey-based	Structured, comparable
AI-moderated interview	From ~$20 to $25 per completed interview	Can scale to most deals	Spoken, richer on the "why"
Calven Win/Loss Agent	No extra cost, part of your subscription	Unlimited	Structured and spoken, both

Cost and coverage figures for the first three rows are vendor-published ranges from User Intuition, Won.Studio, and Koji; treat them as directional, not fixed.

Win/loss stops being a quarterly deck and becomes a standing instrument the moment two things are true: the whole sales act is on the survey, not just the product, and something is reading every response as it lands. Get those right and the CRM dropdown stops being the story you tell yourself about why you lost.

Win/Loss Survey Question Bank

David Kolinek — Wed, 08 Jul 2026 00:00:00 GMT

This is the copy-ready survey from our playbook on win/loss analysis. The playbook explains why a good win/loss survey covers the whole sales act, not just which features were missing, and what to do with the responses once they land. Read it there for the full method.

What's inside

The win survey (12 questions) and the loss survey, grouped into the same five sections: Competition, Product, Sales Team, Pricing & Legal, and Closing.
The exact wording and question type for every question, ready to drop into a form and send.
A build that puts only two of the twelve questions on product features. The rest surface the competitive set, the sales experience, pricing, and procurement, the things that usually decide a deal.

Free template: Win/Loss Survey Question Bank

The full win and loss surveys, five sections each, ready to copy and send.

Battle Cards: What Goes in One, and Why They Rot (+Template)

David Kolinek — Tue, 07 Jul 2026 00:00:00 GMT

Accuracy is the first thing a battle card loses. One that was right the day it shipped can be wrong by the time a rep opens it in a live deal. The competitor repriced. The feature the card listed as a gap shipped last month. The rep reads the line out loud, the buyer corrects them, and the card is finished. Not just for that call, for good. A rep who gets burned once stops opening it, and they tell the other reps. That is the real story of battle cards: the template is trivial, and the discipline that keeps it true is the entire job. A battle card is the sales-facing distillation of your competitor analysis, and a distillation is only worth anything while it holds up.

What is a battle card?

A battle card is a one-page, sales-facing reference for selling against a specific competitor: the differentiators, objection responses, traps, and proof a rep can use live in a deal. Some vendors write it as one word, "battlecard," most write two. Either way it's the distilled, speakable output of your competitor analysis, not a fresh research document.

The test is whether a rep can skim it in about thirty seconds between a discovery call and a demo. One screen, scannable, built to be glanced at under pressure. If a rep has to sit and read it, it stopped being a battle card and became a document.

That distinction matters because a battle card is a product marketing deliverable, and PMMs are the people most tempted to overload it. The instinct is to put everything you learned about the competitor onto the card. The discipline is to leave almost all of it off.

What goes in a battle card?

A battle card carries the handful of things a rep can actually say when a competitor is in the deal: a one-line snapshot, why you win, where the rival is genuinely strong, objection responses, landmines that expose their weakness, discovery questions, a positioning one-liner, a pricing contrast, and proof. Nine elements at most, each written as a talk track rather than marketing language.

Element	What it is	What good looks like
Competitor snapshot	One or two lines on who the rival is and who they sell to	A sentence a rep can drop in passing, not a company bio
Why you win	The two or three differentiators that hold against this specific competitor	A buyer outcome, not a feature: "you keep it current without a person babysitting it"
Where they're strong	An honest read of where the rival genuinely beats you	Stated plainly. This is the row reps trust the card for
Objection responses	The objection you hear when this rival is in, and the approved reply	A line the rep can say out loud, tested, not a paragraph of positioning
Landmines	Questions that expose the competitor's known weakness	"Ask them how long setup actually takes." A trap, not a claim
Discovery questions	What to ask to learn whether that weakness even applies to this buyer	Two or three, tied to the landmine above
Positioning one-liner	The single line a rep says when the buyer asks why you over them	One sentence, drawn from your positioning, short enough to repeat
Pricing contrast	How your pricing and packaging compare, and where theirs is gated	The gap named plainly, not a full price teardown
Proof	A reference or switch story from a customer who chose you over this rival	A named, sayable proof point, ideally a short quote

One framing keeps every row honest: Fact, Impact, Act, a battlecard framework Klue popularized. Each field should carry a fact, why it matters to this buyer, and the thing the rep does with it, a question to ask or a line to say. A row that stops at the fact is trivia. The "where they're strong" row is the one reps check the card against, because it's the one that gets them ambushed if it lies. And landmines are what separate a battle card from a fact sheet. A fact sheet describes the competitor. A battle card gives the rep a move.

If your sales reps can't communicate it to a prospective customer, it means nothing.

— Patrick Wall, Competitive Enablement Leader, Palo Alto Networks

Free template: Battle Card Template

A one-page competitive battle card: competitor snapshot, why-we-win, landmines, objection responses, and proof, with a prompt and a quality check under every field.

Battle card vs. competitor analysis: what's the difference?

The competitor analysis is the research system: a per-competitor profile of a dozen fields that lives internally and gets maintained. The battle card is its sales-facing distillation, the four or five things from that profile a rep can actually say in a deal. The analysis is where you think. The card is what you hand the field.

Keep the two jobs separate and both get better. The full SWOT, the funding history, the GTM motion, the source links, all of that belongs in your competitor analysis, where a PMM reasons about the competitor over time. The card carries only the distilled subset a seller speaks under pressure. Collapse them and you get the two failures everyone recognizes: a battle card so long no rep reads it, or an analysis so thin it's just the card with more whitespace.

The dependency runs one way, and it's the reason cards go wrong. A card inherits every error upstream of it. Tag the wrong competitor in the analysis and the card becomes a confident, well-designed lie in a rep's hands. The card can't be more honest than the research it distills.

What are the types of battlecards?

Most battle cards are competitive head-to-head cards, one per priority competitor, and that's the type worth building first. The other common types reorganize the same intelligence around a different axis: persona cards around a buyer's pains, objection cards around specific objections, and win/loss cards built from why deals were actually won and lost.

The competitive head-to-head card is the default, one per named rival you lose deals to. Build these first.
A persona card reorganizes the same intelligence around a buyer type, for when the competitor looks different to a CFO than to an engineer.
Organize by objection rather than by rival and you have an objection-handling card, which earns its place when one objection (price, security, migration) cuts across several competitors.
Win/loss cards are built from what buyers actually said. That is the most trustworthy source, and the hardest to keep fresh.

Layer role on top when it earns its keep: BDRs want the discovery questions, AEs want the late-cycle objection handling, SEs want the feature and roadmap depth, execs want the snapshot. Same intelligence, four depths. What you don't want is twelve variants of a card nobody has time to maintain, which is the next problem.

Why do battlecards go out of date?

Battle cards rot because the card is static and the competitor isn't. Rivals reprice, relaunch, and reposition on their own cadence, continuously, while the card sits frozen at the moment it was authored. And the failure isn't gradual. Trust is binary: a rep who gets burned by one wrong line in front of a buyer stops opening the card at all.

The pressure behind that is rising. 57.5% of teams report more competitive deals than a year ago, per Crayon's 2026 State of Competitive Intelligence, which means the cost of shipping a stale card keeps going up. Meanwhile the maintenance itself is the acknowledged bottleneck. 58% of CI pros say keeping battlecards and content updated is a struggle, per Crayon's 2024 report. Every week that struggle loses, the gap between the card and reality widens.

Most of the ways a card fails trace back to that decay, or to skipping the distillation discipline in the first place.

Mistake	Why it fails
Too long to skim	A rep can't read a page mid-call. If it isn't scannable in thirty seconds, it doesn't get used
Out of date	Built once, never refreshed. One wrong line in a live deal and the rep abandons the card
Jargon, not talk tracks	Written in positioning language the rep can't say out loud
No landmines	A fact sheet with no offensive move. Describes the competitor, arms the rep with nothing
Dishonest "why we lose"	Omitting where the rival genuinely wins. The rep finds the gap in front of the buyer
Built off the wrong competitor	Inherits a bad CRM tag or the rival's own marketing claims from upstream

The card that's out of date is the expensive one, because it doesn't fail loudly. It fails the first time a rep trusts it in a deal, and you find out from the loss review.

How do you keep battlecards current?

You keep a battle card current by treating it as a living distillation, not a document, and refreshing it whenever the underlying competitor signal changes rather than on a calendar. Tie the cadence to how fast each rival moves: tight on the two or three you actually lose to, light on the rest. A quarterly-refresh policy is really just a decision about which quarter the card will be wrong in.

"Update harder" was never the fix, because the work doesn't fit a human's week. Watching every competitor's pricing page, docs, changelog, and reviews, catching each change, and folding it back into the right line of the right card is continuous, low-glory work that gets dropped first. This is where AI takes the grind while the judgment stays human: the machine watches and drafts, the PMM decides what's true and what's sayable.

That standing cost, the watching no one has time for and the card that goes wrong in silence, is the job the competitive intelligence agent Calven builds exists to carry. Today it identifies your direct and indirect competitors, writes the per-competitor dossier, generates sourced battle cards, and watches competitor websites and docs, folding each change back into the right section instead of letting it age.

The mistake I see most often is treating the battle card as a deliverable you finish. It isn't something you complete, it's a distillation you maintain. The template takes an afternoon. Keeping it true takes forever, and the day it lies to your own reps is the day it's dead.

— David Kolínek, Co-founder at Calven

The template was never the hard part. Any team can fill in nine fields once. The card that earns its place is the one still true on the day a rep pulls it up in a live deal, and still short enough to say out loud.

Battle Card Template

David Kolinek — Tue, 07 Jul 2026 00:00:00 GMT

This is the fill-in template from our guide on battle cards: what goes in one, and why they rot. The guide walks through the anatomy, draws the line between a battle card and the competitor analysis it distills, and covers why cards go stale and how to keep them current. Read it there for the full picture.

What's inside

A one-page competitive card with all nine fields, from the competitor snapshot to the proof points a rep can actually say.
A prompt and a quality check under every field, so you fill each one from real research instead of guessing.
The "where they're strong" and landmine fields that separate a battle card from a fact sheet.
A short note on keeping the card current, since the fields that go wrong fastest are the ones that cost you the most.

Free template: Battle Card Template

A one-page competitive battle card: competitor snapshot, why-we-win, landmines, objection responses, and proof, with a prompt and a quality check under every field.

Competitor Analysis Template

David Kolinek — Mon, 06 Jul 2026 00:00:00 GMT

This is the fill-in template from our guide on how to fill in a competitor analysis. The guide walks through the method, shows where to source every field, and covers the mistakes that break a competitor doc. Read it there for the full picture.

What's inside

A per-competitor profile with all twelve fields, from company snapshot to the mandatory "so what to action" column.
A you-vs-rivals comparison matrix that reads "fully / partially / not supported," never a bare yes or no.
A prompt, a source, and a quality check under every field, so you fill each one from the right place instead of guessing.
The tiering step and the sourcing checklist that decide where your effort actually goes.

Free template: Competitor Analysis Template

A per-competitor profile and a you-vs-rivals matrix, with a prompt, a source, and a quality check under every field.

Message Map Template

David Kolinek — Sun, 05 Jul 2026 00:00:00 GMT

This is the fill-in template from our guide on how to build a message map. The guide walks through the method, resolves the map-versus-framework confusion, shows a worked example, and covers the mistakes that break a message map. Read it there for the full picture.

What's inside

The full hierarchy laid out blank by blank: core message, three pillars, proof points, and a persona cut.
A prompt and a quality check under every blank, so you fill each one with a real decision instead of the obvious guess.
A "capture the buyer's language" step before the map, because a map built on internal jargon is a well-formatted document that doesn't land.
A worked example reconstructed from public messaging, and the quality checks that catch a soft pillar or an unsourced proof point.

Free template: Message Map Template

A blank message map skeleton: the core message, three pillars with a supporting line each, proof points under every pillar, a persona cut, and a prompt plus a quality check under each blank.

The Product Marketing Tech Stack in 2026: Sprawl and Consolidation

David Kolinek — Sat, 04 Jul 2026 00:00:00 GMT

A two-person product marketing team today is usually paying for a competitive intelligence subscription, a conversation intelligence seat, an enablement platform, two or three doc tools, and a couple of AI assistants. They use maybe a third of it. The modern PMM stack is a pile of point tools nobody has the headcount to run, it's already consolidating, and the place it's heading isn't a bigger dashboard. It's one always-current knowledge base you reach inside your own AI tool.

What's actually in a product marketing tech stack?

A 2026 PMM stack breaks into roughly six categories of point tools, each with a market leader or two and its own login, and most of them are owned or leaned on by product marketing even when another team pays the bill. Here's the honest map.

Category	Example tools	What it does	Typical cost band
Competitive intelligence	Klue, Crayon, Kompyte	Monitors competitors, builds battle cards	~$15–40k/yr; Kompyte the budget option
Conversation intelligence	Gong, Chorus (by ZoomInfo)	Records, transcribes, and analyzes sales calls	~$160–250/user/mo
Sales enablement	Highspot, Seismic	Houses and governs sales content	Enterprise, five to six figures
Docs / knowledge	Notion, Confluence, Google Workspace	Where positioning and messaging actually live	Per-seat, low
AI assistants	ChatGPT, Claude, Copilot	Drafting, research, analysis	Per-seat
Analytics (adjacent)	GA4, Amplitude	Product and marketing measurement	Varies

The most underused fuel on that list is conversation intelligence. Product marketing usually buys the CI tool and treats Gong as the sales team's toy, but the recorded calls are where the real objection language lives, where a competitor's name comes up unprompted, and where the actual reason a deal died gets said out loud. That's win/loss signal and messaging research, sitting in a tool most PMMs never open.

Why do small teams use only a third of their martech?

Because you can't staff what you buy. The martech landscape has plateaued at roughly 15,000 tools (CMSWire, "Peak Martech"), and utilization keeps falling: marketers report using about 42% of their stack, down from 58% in 2020 (Chief Marketer on the Gartner martech survey). The tools keep getting bought. The hours to run them don't appear.

Product marketing feels this harder than most of marketing, because the teams are so small. 44.3% of PMM teams are one or two people (PMA State of Product Marketing 2025), and 41% of PMMs hired in 2026 said the role's scope wasn't realistic for one person (Fluvio 2026 Hiring Trends Report). A CI platform needs someone to tune the alerts and write the battle cards. A conversation intelligence tool needs someone to actually listen. Buy both, staff neither, and you own two subscriptions that generate guilt.

Anyone who has run the function has lived the pattern: the stack grows faster than the team, and at least one tool in it is paid for and barely opened. The subscription outlives the person who championed it. That is not a discipline problem you can buy your way out of. It's a sign the stack is overdue for the shift AI is already bringing to the PMM role: fewer tools you operate by hand, more of the work the tools do themselves.

Where does a bare AI chatbot fall short for PMM work?

AI is already in the stack, and deeply. PMM adoption of ChatGPT and Claude is near-universal for market scanning, messaging, and competitive analysis, and most teams reach for a chatbot before they open a dedicated tool (Fluvio 2025 AI Trends Report). The problem isn't AI. It's ungrounded AI.

A bare chatbot has no memory of your positioning, no live feed of competitor moves, and a habit of filling gaps with confident invention. Ask it about a competitor's pricing and it will answer, fluently, from whatever it absorbed months ago or simply made up. Anyone who builds AI agents daily learns the same lesson early: the model isn't the constraint, the grounding is. A chatbot that can't read your current knowledge base will invent a competitor fact rather than admit it doesn't have one, and it says it with total confidence.

The fix isn't a better chatbot. It's giving the one you have something true to stand on.

What is an AI product marketing platform?

The endpoint of consolidation isn't a bigger dashboard you check. It's a single source of truth that holds your definitions, positioning, competitive signals, personas, and messaging, and stays current, unlike the doc tools where that knowledge lives today and goes stale. Notion doesn't know your competitor repriced. A shared knowledge base that agents maintain does.

That's the shape of the emerging category. A competitive intelligence agent watches the market, a win/loss agent reads the closed deals, a messaging agent keeps the matrix aligned, and a persona agent pressure-tests every draft, all writing into and reading from one live base. Competitive signal feeds messaging. A market shift flags your positioning for review. Nothing goes stale because keeping it current is the agents' job, not a quarterly scramble. The concrete version looks mundane: you ask "what's changed on our top competitor this month, and what does it mean for the battle card," and the answer comes back from a base an agent has been maintaining, not from a doc someone last touched in Q1. Calven is one early entrant building in this direction; it won't be the only one.

Positioning is exactly the knowledge that most rewards being kept live, because AI raises the stakes on it. As the cost of building software collapses and products start to look interchangeable, the judgment about what actually sets you apart matters more, not less.

I think the impact of AI on positioning is potentially much more potent.

— April Dunford, Author, Obviously Awesome

What is MCP, and why does it matter for product marketing?

MCP, the Model Context Protocol, is an open standard for secure, two-way connections between AI tools and the systems where your data lives. Instead of a fragmented set of one-off integrations, it gives any AI assistant a common way to read from and write to your sources. Anthropic introduced it in November 2024 (Anthropic, Introducing the Model Context Protocol).

It's not one vendor's bet. OpenAI adopted MCP in March 2025, Google DeepMind confirmed support in April 2025, and Microsoft and GitHub joined the steering committee in May 2025. In December 2025, Anthropic donated the protocol to the Agentic AI Foundation under the Linux Foundation, making it vendor-neutral and industry-governed (Anthropic, Donating the Model Context Protocol). When the three largest AI labs and the Linux Foundation all back the same standard, it's infrastructure, not a trend.

Here's why that reshapes the stack. Your knowledge base stops being locked in a vendor UI. You query your positioning, your competitive signals, and your personas from inside Claude, ChatGPT, or Cursor, the tool you already work in, instead of logging into yet another dashboard to go find them. This is the direction the stack is moving, not how every PMM already works, but it's close enough to build toward. Calven's content is MCP-accessible by design, which is less a feature than a stance on where the knowledge should live.

The knowledge a product marketer needs, the current positioning, the live competitive read, has to answer the moment you ask for it, right inside the tool you're already working in. And it has to be current every single time, not a snapshot someone refreshed last quarter.

— David Kolinek, Co-founder, Calven

The stack is contracting toward one source of truth

Fewer tools, one live knowledge base, reachable where the work already happens. That's the direction Calven's platform is built around, and the numbers already lean that way: martech has stopped growing, and most of a stack goes unused.

Consolidation is only worth it on one condition. The single source has to be genuinely kept current, by agents that maintain it, not by a person who was already too busy. Trading three tools you underused for one platform you underuse is not progress; it's a worse UI with a bigger invoice. The bet worth making is the one where the knowledge stays live on its own and answers from inside the tool you already use. Judge any platform by that, and most of the stack's sprawl answers itself.

Ideal Customer Profile (ICP): How to Build One (Examples + Template)

David Kolinek — Fri, 03 Jul 2026 00:00:00 GMT

Most ICPs describe a market a team would like to sell to, not the customers who actually succeed with the product. The document reads well in a slide, names an industry and a company size, and commits the go-to-market motion to the wrong accounts. An ideal customer profile is only worth having if it's built from evidence about who already wins with you. Below is the definition, the distinctions worth getting straight, a method grounded in real signal, and a template you can copy.

What is an ideal customer profile?

An ideal customer profile is a description of the company, at the account level, most likely to become one of your best customers: to buy, get value fastest, and retain and expand. It is defined by the attributes that correlate with real value, and it is built from evidence about who already succeeds with you rather than an aspirational guess.

Gartner frames it as the firmographic, environmental, and behavioral attributes of the accounts a company expects to become its most valuable customers, developed through both qualitative and quantitative analysis. Two things in that definition carry the weight: it describes accounts, not people, and it comes from analysis, not opinion.

A definition on paper is only a hypothesis. It holds for exactly as long as it matches reality, and reality moves. Your product changes, your buyers change, your sales team changes, and the profile that fit last year stops fitting. A useful ICP is one you check against real results and adjust, not a slide you write once. Defining the right audience is core product marketing work, and the ICP is where that work either grounds itself in data or drifts into wishful thinking.

The ICP gets confused with four adjacent concepts, and the confusion is expensive because each one operates at a different level. An ICP describes a company. A persona describes a person. A target market describes who could buy. TAM/SAM/SOM sizes the market. The ICP is a fit definition that cuts across all of them.

Concept	What it describes	Level
Ideal customer profile	The companies most likely to become your best customers	Account
Buyer persona	A specific person inside the account you sell to	Person
Target market	Every company that could plausibly buy	Segment
TAM / SAM / SOM	The size of the opportunity (nested: TAM ⊃ SAM ⊃ SOM)	Market

ICP and persona are the pair people mix up most, and they don't compete. They compound. Personas tell you who you're speaking to. The ICP tells you which companies are worth speaking to in the first place: it narrows you to the right accounts, and the personas tell you how to talk to the humans once you're in the room.

Target market and TAM sit on the other side. A target market is everyone who might buy; the ICP is the subset you most want to win because they actually succeed. TAM, SAM, and SOM are sizing estimates, useful for a board deck and useless for a rep deciding who to call. The ICP is the bridge from an abstract market to a concrete list of accounts worth pursuing this quarter.

What goes in an ICP?

An ICP has six layers, and the first two are the ones every template covers: firmographics and technographics. The layers that separate a real ICP from a data-vendor filter are the last four, where the profile stops describing a market and starts explaining why specific accounts buy, succeed, and stay.

Layer	What it captures	Examples
Firmographics (who)	Static company shape	Industry, employee count, revenue, geography, growth stage, funding
Technographics	The stack they run	Tools you integrate with, tools you replace, the core platform they've standardized on
Behavioral / intent (when)	Proxies for urgency	Funding rounds, new leadership, hiring patterns, product usage, active category evaluation
Fit signals (why)	The job, pain, and trigger	The specific pain your product solves and the event that starts the search
Value / retention lens	Who succeeds and stays	Accounts that get value fastest, close fastest, retain and expand
Negative filters	Who to exclude	Attributes that predict churn or poor fit

A firmographics-only ICP filters on size and industry with no account of why anyone buys, which is the most common way teams get it wrong. Two companies can look identical on paper: same industry, same headcount, same stack. One renews and expands; the other churns in eight months. The difference lives in the bottom three rows: the trigger that made them look, the pain the product solved, and whether that kind of account succeeds. Those rows are where an ICP earns its keep.

How to build an ICP, step by step

Build the ICP from evidence about who already wins with you, then validate it against live pipeline. The order matters: start from your real best customers and work backward to the attributes they share, rather than starting from the market you wish you owned and hunting for accounts that fit the story.

Start from your best current customers. List the accounts that bought fast, got value fast, retained and expanded, and referred you. This is the account-level sibling of the best-fit customers discipline that underpins good positioning, and it's the same move April Dunford's best-fit-customers method makes: study who already succeeds, don't guess.
Find the common attributes. Pull firmographics, technographics, and product usage across that set from the CRM and product data. Sort the list by ACV and retention, then look for the attributes that concentrate in the top quartile and thin out below it.
Add the fit layer. Firmographics tell you who; use case, pain, and trigger tell you why now. A firmographic twin that doesn't share the pain is not in your ICP.
Isolate the attributes that actually predict fit. Keep the attributes that show up far more often in your wins and renewals than in your losses and churn; drop the ones that are true of everyone regardless of outcome. This is the hard, data-grounded part, and the one that most teams never finish.
Write the negative profile. Name who to exclude. A useful ICP excludes most of the market on purpose.
Validate against live pipeline. Check whether the accounts actually moving and closing match the definition. If they don't, the profile is wrong, not the pipeline.

Free template: Ideal Customer Profile Template

The six ICP layers, a prompt and a quality check under every field, the best-customer evidence step, and the pipeline-validation check.

An ICP example

Real ICPs are rarely public, because a company's ICP is an internal strategic document, not marketing copy. One useful exception: Cognism published its own worked ICP structure inside its guide, walking through the firmographics, the roles, and the qualifying signals it uses. A company showing its actual profile is worth studying precisely because so few do.

To make the finished shape concrete, here is a six-layer ICP for a fictional B2B SaaS analytics product. This is a constructed illustration, shown to demonstrate the shape, not a real company's profile:

Firmographics: Series B software companies, 200 to 800 employees, North America and UK.
Technographics: running a modern data warehouse, already paying for a BI tool they've outgrown.
Behavioral / intent: hired a first head of data in the last year.
Fit signal: analysts spending days building reports the business needs in hours.
Value / retention lens: teams that connect a live data source in week one retain at twice the rate of those that don't.
Negative filter: pre-Series-A companies with no dedicated data owner.

The retention line is the one that does the real work, and it's the one a firmographic-only profile never captures. It also settles the most common internal argument about an ICP. If the CEO dreams of Fortune 500 logos but the accounts that actually buy fast and stay are Series B SaaS companies, the ICP is Series B SaaS companies. The evidence decides, not the ambition.

Common ICP mistakes

Most broken ICPs fail in a handful of predictable ways, and they all share a root cause: the profile was written from what a team believes about its market instead of what the data shows about its customers.

Mistake	Why it fails
Too broad ("mid-market B2B SaaS")	Describes a market segment, not which companies to call. A real ICP excludes most of the market.
Aspirational, not real	Built around the logos leadership wants, not the customers who succeed and retain.
Static, never updated	Product, market, and pricing move; an 18-month-old ICP targets a customer you no longer sell to.
Confusing ICP with persona	Person-level attributes where account-level belong; targets people before qualifying companies.
Firmographics only, no pain or trigger	Filters on size and industry with no account of why anyone buys.
Built from opinion, not evidence	A whiteboard guess instead of closed-won patterns, product usage, and buyer evidence.

The last one is the costliest, and it's the one that produces all the others. An ICP built from opinion can be too broad, aspirational, and stale all at once, and no one will notice until the pipeline stops converting. Build it from evidence and most of the rest of the table takes care of itself.

What a good ICP is really grounded in

The template is the easy part. A good ICP is grounded in evidence, and the most decisive evidence is your own pipeline. Are the accounts that match your ICP actually converting, closing, and retaining at higher rates than the accounts that don't? Some off-profile deals always close. The real question is whether ICP-fit accounts win at a materially better rate. If they don't, the profile is wrong, not the pipeline, and it needs to change.

A good ICP also stays current. Refresh it as often as your product, market, pricing, and best-customer set move, which for most companies is continuously. A profile written 18 months ago and left in a Google Doc describes a customer you may no longer sell to, and the failure is silent: a stale ICP keeps sending reps after the accounts you used to win and away from the ones you win now.

Understanding your buyers in their own terms, not yours, is the other half. The pain and trigger that define fit come from what customers actually say in sales calls and win/loss interviews, and that language rarely matches the vocabulary a team uses internally. Bob Moesta, who co-created Jobs-to-be-Done and built a career interviewing buyers about why they bought, puts the trap plainly:

The English language kind of sucks. The same word can mean five different things, and five different words can mean the same thing.

— Bob Moesta, Co-creator of Jobs-to-be-Done

Grounding an ICP in real results and keeping it current as your product, market, and customers move is judgment work. AI can draft and monitor faster than any small team, but deciding which accounts actually succeed with you stays yours.

Further reading: Demand-Side Sales 101 — Bob Moesta with Greg Engle (Book)

The practitioner text on why buyers actually buy and how to capture their real language, the exact evidence a good ICP is built on.

Keeping that profile honest against live pipeline is the job Calven's ICP Agent, still in progress, is meant to take on: validating the ICP against the accounts actually moving and closing in your CRM, and isolating the attributes that genuinely predict a win rather than the ones that just describe your customers. It isn't fully shipped yet.

The template you can finish in an afternoon. The ICP underneath it is only as good as the evidence you ground it in, and only as useful as it is current. Copy the template above, start from the accounts that already succeed with you, and adjust it as reality does.

Ideal Customer Profile (ICP) Template

David Kolinek — Fri, 03 Jul 2026 00:00:00 GMT

This is the fill-in-the-blank template from our guide on how to build an ideal customer profile. The guide walks through the method, shows a real published example and a labeled illustration, and covers the mistakes that break an ICP. Read it there for the full picture.

What's inside

The six layers of an ICP, laid out field by field: firmographics, technographics, behavioral and trigger signals, use case and pain, the value and retention lens, and the negative profile.
A prompt and a quality check under every field, so each one gets filled with a real decision instead of the obvious guess.
The best-customer evidence step to run before you touch the attributes.
A pipeline-validation check to keep the profile honest against the accounts actually closing.

Free template: Ideal Customer Profile Template

The six ICP layers, a prompt and a quality check under every field, the best-customer evidence step, and the pipeline-validation check.

How to Write a Positioning Statement (Examples + Template)

David Kolinek — Thu, 02 Jul 2026 00:00:00 GMT

The fill-in-the-blank positioning formula every listicle hands you is the easiest thing in product marketing to complete, and the easiest to complete wrong. Each blank hides one of the hardest strategic decisions a company makes, dressed up as a form field. Fill them in on autopilot and you haven't written positioning. You've written a sentence that sounds like it and commits you to nothing. Below is the method that gets the order right, real B2B SaaS examples with their sources labeled, and a template you can copy.

What is a positioning statement?

A positioning statement is an internal strategic decision written down in one paragraph: who a product is for, what category it belongs to, and why it wins against the alternative a buyer would otherwise choose. It is the source document every piece of external messaging draws from, not the homepage headline, the tagline, or customer-facing copy. Positioning and messaging is the core judgment call of the discipline.

Most people meet it through one canonical format, from Geoffrey Moore's Crossing the Chasm:

For (target customer) who (need or opportunity), the (product) is a (category) that (key benefit). Unlike (primary alternative), our product (differentiator).

Writers call this the positioning "madlib." The structure is useful: it forces a single target, a single category, a single differentiator, more discipline than most teams apply on their own. What it does not do is help you make the judgment about what goes in each blank.

The positioning statement gets confused with three cousins: the value proposition, the tagline, and the mission statement. All four get used interchangeably, and several of the pages ranking for this query get them wrong, calling a mission statement a positioning statement. The distinction worth drawing is audience and job.

Artifact	Audience	Job	Example shape
Positioning statement	Internal	The strategic decision: who it's for, the category, why you win vs. the alternative	"For [X] who [need], [product] is the [category] that [benefit]. Unlike [alternative], it [differentiator]."
Value proposition	External	The customer-facing promise of value; a headline claim	"Close more deals with the #1 revenue AI."
Tagline	External	A short, memorable brand phrase	"Just Do It."
Mission statement	Internal / external	The company's reason for existing, beyond any product	"To make the world more inbound."

A mission statement sitting in the positioning slot means no positioning decision was ever made.

Why the fill-in-the-blank formula fails as a discovery tool

The formula is a fine way to record a decision and a dangerous way to reach one. April Dunford, whose book on positioning is the closest thing the discipline has to a standard reference, is blunt about the exercise: she thinks it is "not only pointless but potentially dangerous."

The exercise assumes that there is only one answer for each of the blanks. You simply "know" what it is.

— April Dunford, Author, Obviously Awesome

You usually don't. Most products could sit in more than one category, compete against more than one alternative, and deliver different value to different segments. The blanks aren't fields to fill with the obvious answer. They are the strategy. Dunford's method replaces the guessing with an ordered set of components: the competitive alternatives, the unique attributes the alternatives lack, the value those create, the segment that cares most, and last the market category that makes all of it obvious.

Further reading: Obviously Awesome: How to Nail Product Positioning — April Dunford (Book)

The definitive playbook on B2B positioning, and the source of the five-component method this piece is built on.

How to write a positioning statement, step by step

Write the statement last. Run Dunford's discovery in order, then pour the answers into Moore's structure.

List the real competitive alternatives. What the buyer would genuinely use instead, which for most B2B products is a spreadsheet, a manual process, or doing nothing.
List the attributes those alternatives lack. The specific capabilities only your product has.
Translate each attribute into value. The outcome it produces. An attribute nobody values is trivia, not a differentiator.
Identify who cares most. The segment that gets the most value fastest: your fastest-closing, best-renewing accounts, not the hardest sells you wish you could win.
Choose the category that makes your strengths obvious without handing you an unwinnable incumbent fight. Call yourself a database and you're compared to Oracle. Call yourself an embeddable database for mobile and the field clears.
Then fill in the formula. Now the blanks have earned answers. The fill-in template below has a prompt and a quality check under each one.

Discovery is the part no tool does for you. AI drafts and monitors faster than any small team, but the judgment of what a product uniquely delivers is yours to earn.

Free template: Positioning Statement Template

The Moore skeleton, a prompt and a quality check under every blank, the discovery steps, and the mistakes to avoid.

Real positioning statement examples (with sources)

No company publishes its internal positioning statement. It's a strategic document, not marketing copy. What you can find, and learn far more from, are documented cases where a named authority walks through the positioning decision itself: the move from a weak position to one that wins. Here are three, each sourced, each demonstrating one of the five decisions from the method above.

Company	The positioning move	What it teaches
Help Scout	Reframing customer support from a cost center to a growth driver	Choosing your frame of reference
Janna Systems	From "enterprise CRM," up against Siebel, to "CRM for investment banks" (later acquired by Siebel)	Narrowing the segment to win
Dunford's embeddable database	From "database," measured against Oracle, to "embeddable database for mobile"	Your category picks your competitors

All three come from April Dunford, either her own case or one she has documented. Notice what they are: not clever sentences but decisions. A frame, a segment, a category, which the Moore formula records only after the work is done. Poured into that formula, the Help Scout decision reads:

For support teams who are treated as a cost center, Help Scout is the help desk that turns support into a driver of growth and retention. Unlike ticket-first tools built for deflection, it is built around human conversations.

That's a reconstruction, shown to make the finished shape concrete. Help Scout has never published it verbatim, and neither has anyone else on this list. Which is the point. The statement is the last, easy step. The decision underneath it is the work.

Common positioning statement mistakes

Most bad positioning fails in predictable ways: putting the wrong artifact in the positioning slot, a tagline or a mission statement; targeting everyone, so the value lands with no one; stacking five differentiators where the formula allows one, which ranks none; and picking a category with a fatal incumbent (the Oracle trap). But the costliest mistake is the first move, not the last: starting from the blanks and treating the formula as discovery.

"We save time" isn't a position; it's a placeholder.

— April Dunford, Author, Obviously Awesome

How often should you update your positioning? (positioning is never "one and done")

As often as your product, your competitors, and your market change, which is to say continuously. You can't see a company's internal statement, but you can watch its positioning move in public, on the homepages of companies everyone knows. Six, captured mid-2026:

Company	Live positioning line (mid-2026)	What moved
Gong	"#1 AI OS for Revenue Teams"	Category creation, from conversation intelligence to "Revenue AI OS"
HubSpot	"HubSpot Agentic Customer Platform"	Inbound marketing → CRM → customer platform → agentic
Jira	"Project Management for the AI Era"	Same category, new "why now"
Notion	"Where teams and agents think together"	Re-anchoring the workspace around agents
Slack	"All your people and AI agents working together"	Drift: 2019's "collaboration hub" still gets quoted
Figma	"The intelligent canvas for infinite creativity"	Expanding the category beyond "design tool"

These are taglines, not positioning statements. They're the public expression of a positioning decision, not the decision itself. But read each against what the company used to say and the drift is unmistakable. If their positioning moved this much, yours did too. Dunford makes the point directly: positioning is never "one and done".

Keeping it current is the job the Positioning Agent Calven is building is meant to take on: maintaining positioning as a living document, assembled from your ICP, personas, competitive intelligence, and market research, and flagging itself for review when those inputs shift. That agent isn't shipped yet. But positioning that isn't maintained doesn't stay neutral. It turns into a statement that actively misleads your own sellers.

The formula was never the hard part. The hard part is the five decisions underneath it: the alternative, the attributes, the value, the segment, and the category. When you're ready to write yours down, copy the template above and work through it, one earned answer at a time.

Positioning Statement Template

David Kolinek — Thu, 02 Jul 2026 00:00:00 GMT

This is the fill-in-the-blank template from our guide on how to write a positioning statement. The guide walks through the method, shows real B2B SaaS examples with their sources, and covers the mistakes that break a positioning statement. Read it there for the full picture.

What's inside

The Geoffrey Moore skeleton, laid out blank by blank.
A prompt and a quality check under every blank, so you fill each one with a real decision instead of the obvious guess.
The five discovery steps to run before you touch the formula.
The common mistakes that turn a positioning statement into a mission statement.

Free template: Positioning Statement Template

The Moore skeleton, a prompt and a quality check under every blank, the discovery steps, and the mistakes to avoid.

How AI Is Changing Product Marketing (and Why It Won't Replace PMMs)

David Kolinek — Wed, 01 Jul 2026 00:00:00 GMT

No, AI is not going to replace product marketers. It is going to do close to the opposite of what the layoff headlines suggest, because product marketing was always understaffed relative to what it was accountable for, and agents finally close that gap. The stronger claim, the one this piece sets out to earn, is that AI may be the best thing that has happened to the discipline.

Will AI replace product marketers, or just the grind?

AI replaces tasks, not the function, and the tasks it takes are the ones a chronically understaffed team never had time to do well anyway. The bottleneck was never judgment. It was hours. In Fluvio's 2026 hiring survey, 41% of PMMs hired that year said the scope of the role wasn't realistic for one person (Fluvio 2026 Product Marketing Hiring Trends Report). This is a structurally understaffed function where 44.3% of teams are one or two people. Agents change the arithmetic on the parts that were pure hours. What they cannot touch is the part that was always the actual job.

Generative AI is an assistive technology... It's terrible at critical thinking and creative problem-solving. That is the key skill set of a good PMM.

— Claudia Michon, Former Head of Product & Solutions Marketing, Airtable

Shift one: the stack moves from point tools to platforms

A PMM's stack has been a stitched-together set of point tools, each one needing a person to run it. Klue and Crayon start around $15,000 to $20,000 a year, and those are starting prices, not typical ones. In enterprise organizations the seat count, competitors tracked, and integration scope push a deployment to $40,000 or more, and even then the license is rarely the real cost (Parano.ai, Klue vs Crayon, 2026). A tool a two-person team cannot staff is a subscription that generates guilt. Gartner has repeatedly found marketers use just 42% of their martech stack, down from 58% in 2020 (Chief Marketer on the Gartner martech survey). The monitoring and triage that justified a dedicated tool is exactly the first-level work an agent does without a seat to fill, which is why an AI product marketing platform is emerging as a category rather than one more point tool.

Shift two: the skills move from doing the work to orchestrating it

The core skill is shifting from producing the work to directing the AI that produces it. AI fluency is now a baseline hiring requirement, not a differentiator (3Search, most in-demand marketing skills for 2026; National University, top marketing skills for employers). The frontier skill is not prompting a chatbot well. It is connecting AI to your actual tools and data. The concrete version has a name: the Model Context Protocol, introduced by Anthropic in November 2024 and adopted by OpenAI and Google through 2025 (Anthropic, Introducing the Model Context Protocol; Anthropic, Donating MCP). It lets an assistant reach into the CRM, the win/loss transcripts, and the competitor pages instead of sitting in a chat window. To stay ahead you don't need to learn to code. You need to learn to orchestrate.

Shift three: perception flips from "the no team" to visible enabler

Product marketing is often treated as a speed bump, and under-credited because it can rarely measure its own impact cleanly. In the Fluvio survey, 0% of PMMs hired in 2026 called their success metrics "very clear" (Fluvio 2026 Hiring Trends Report). Proactive output changes that. When the competitive digest lands in the sales channel before anyone asks, and the battle card already reflects a pricing change a rep is about to hit, the team stops looking like the place requests go to die. A connected set of agents makes that routine: a competitive finding feeds the messaging, the messaging feeds the card, and the chain stays current without a person re-running it every quarter.

Shift four: what a PMM's week actually looks like now

The change is in where the hours go. The 2022 week was consumed by gathering, drafting, and monitoring done by hand. The 2026 week hands that to agents, and the PMM spends the reclaimed hours on judgment.

Task	2022, by hand	2026, with agents
Market research	A slow desk-research effort done a few times a year, out of date by the time it was written up	Agents track the macroeconomic, technology, regulatory, and vertical shifts that move your market and spell out what each one means for your GTM, then synthesize those trends, competitive signals, and analyst reports through your ICP and positioning into specific opportunities, each pressure-tested against your own site before it surfaces.<br /><br />The PMM decides which of those openings is worth a bet.
Competitive monitoring	Checked in bursts between other work, stale the moment a rival shipped or repriced	Agents watch each competitor's site and news, mine your call transcripts and win/loss notes for competitor mentions, and read the reviews on public review sites and practitioner communities; a confirmed move updates that competitor's dossier on its own and lands as a verified signal that flows straight into the battle cards.<br /><br />The PMM spends the time on how to counter it, not on finding it.
Win/loss analysis	A quarterly scramble to read call notes, when it happened at all	Wired into your CRM so every closed deal triggers outreach, with agents running the survey or an AI-assisted interview; the responses come back as categorized quotes and a per-deal summary, rolled up into a current read on why you win and why you lose.<br /><br />The PMM turns the patterns into positioning and enablement moves.
Audience and persona research	A once-a-year project, when the calendar ever allowed it	Your ICP is kept current and validated against the live pipeline, flagging which open deals actually fit it and which do not; personas stay up to date from call transcripts, online research, and community forums, and are on hand to give real-time feedback on any positioning or messaging draft.<br /><br />The PMM makes the segment calls the data only hints at.
Positioning and messaging	Squeezed in between deliverables, rarely revisited once it shipped	Built on everything the other agents already maintain: a current ICP and personas with their real pain points, the live market trends, and the documented weaknesses of each competitor, so the positioning and messaging actually resonate with the audience instead of guessing at it.<br /><br />The PMM owns the decisive call: what actually sets you apart.
Preparing for a product launch	Assembled from scratch under deadline, the buyer only half understood	Grounded in a real read on the buyer, with agents reviewing the launch narrative through each persona and tying the relevant trends and opportunities to the concrete use cases the launch can lead with.<br /><br />The PMM sets the narrative and decides what to lead with.
Sales enablement and battle cards	Rebuilt by hand each quarter, usually out of date before a rep opened them	Battle cards stay current on their own and reach the deck through MCP, so the slides a rep pulls up are never stale; the same knowledge of your audience, personas, positioning, and messaging can power coaching and training agents that ramp the team.<br /><br />The PMM coaches the judgment the material can't, in the room.

The reclaimed hours are review time plus strategy time, not free time. AI is confidently wrong often enough that a team treating the draft as finished will ship that 72% inconsistency straight to sales.

Why positioning matters more in the AI era, not less

As AI collapses the cost of building software, products proliferate and start to look interchangeable, and buyers weigh "just use a chatbot versus buy the dedicated tool" constantly. That makes the PMM's core judgment, deciding what actually differentiates a product, more load-bearing than ever. When it costs almost nothing to build a competent version of a thing, "it works" stops being a differentiator, and the moat moves to what is hard to copy: the specific judgment about what a defined set of customers actually cares about. April Dunford has written that the last time she saw this many companies abruptly struggling with their positioning was March 2020, at the dawn of the global COVID outbreak. This time, she thinks, the disruption could run deeper.

I think the impact of AI on positioning is potentially much more potent.

— April Dunford, Author, Obviously Awesome

When an AI agent shortlists the vendors, the product with sharp, defensible positioning is the one that makes the list, a shift Wharton's Stefano Puntoni frames as a second revolution in who makes the purchasing decision (Stefano Puntoni in HBR, 2026). Vague positioning does not just lose slowly anymore. It gets filtered out before a human ever sees it.

This is where the tools are heading, too. A category of AI product marketing platforms is forming, where connected agents run competitive intelligence, audience work, and messaging as one system instead of separate tools with separate owners. No vendor owns it yet. Calven is one of the early entrants building there, with an Intelligence Core that connects agents for competitive intelligence, audience, and messaging, the first-level work the old point-tool stack forced a team to stitch together by hand.

Every shift here points the same direction, all because AI took the grind off a function that never had enough hands. And as products blur into each other, the judgment about what actually differentiates one gets harder and worth more. The output was never the value. The decision was, and AI just made it the scarcest thing in the room. That is why the importance of the role goes up, not down.

What Is Product Marketing? The 2026 Guide

David Kolinek — Tue, 30 Jun 2026 00:00:00 GMT

Product marketing is the discipline of deciding what a market needs to hear about a product, and making sure it actually lands: who it's for, why it's different, and how it wins. It is not the team that makes the slides. It is the team that decides what the slides are allowed to say.

Surprisingly little of the job is writing. Most of it is deciding: what is true about a product that a competitor cannot say, which of the ten things it does actually matters to the buyer in the room, what gets said in month one versus month six. Everything downstream of that decision, from a battle card to a launch email, is execution. The decision is the job. That matters more now than it did a few years ago, because AI has gotten good at the execution. It gathers, drafts, and monitors faster than any two-person team ever could. What it still cannot do is the judgment call about what actually differentiates a product, which was always the hardest part of the work.

What is product marketing?

Product marketing is the discipline of bringing a product to market and driving its ongoing success. It sits at the intersection of product, sales, and marketing, and it owns three questions no other function is accountable for: who the product is for, why it's different, and how it wins.

That framing holds up because it is built around ownership, not tasks. A demand-gen team can run campaigns without ever answering "why is this different." A product team can ship a great feature without deciding who it's for. Product marketing has to answer both, every time, or nothing downstream in the go-to-market has a foundation to stand on.

April Dunford, whose book on positioning is the closest thing PMM has to scripture, puts the stakes plainly:

Positioning defines how your product is a leader at delivering something that a well-defined set of customers cares a lot about.

— April Dunford, Author, Obviously Awesome

The emphasis matters. It is not how a product is described. It is how the product is a leader at something a specific group cares a lot about. Vague products with vague positioning do not lose to better products. They lose to products whose PMM did the harder work of choosing a hill to stand on.

Further reading: Obviously Awesome: How to Nail Product Positioning — April Dunford (Book)

The definitive playbook on positioning, referenced throughout this guide.

What does a product marketing manager do? The core deliverables

A PMM owns a connected chain of work: the customer and market research that tells the business what's actually true, the ICP and personas that define who the product is for, the positioning and messaging that decide how it wins, the launches that take it to market, the campaigns that carry it, and the sales enablement sellers use in-deal. Positioning and messaging is the most universal of these: 91% of PMMs own it, more than any other responsibility (PMA 2025 State of Product Marketing Report).

Deliverable	What it is
Customer & market research	Tracking market trends and opportunities, watching competitors, and learning from closed deals. This bucket folds in competitive intelligence, owned by 65.2% of PMMs, and win/loss analysis: understanding why the company actually wins and loses, then feeding it back into positioning (PMA 2025).
ICP & personas	Defining who the product is actually for and what they care about.
Positioning & messaging	Deciding how the product is a leader at something a defined set of customers cares about, then turning that into the words that land across audiences and channels. Owned by 91% of PMMs, more than any other responsibility (PMA 2025).
Product launches	Taking a product or feature to market, from big-bang launches to the continuous drip of feature releases.
Campaign strategy	Shaping the message and target audience for go-to-market campaigns, so the positioning shows up where buyers are. Demand gen usually owns the channels and budget; PMM owns the story that drives them.
Sales enablement	Battle cards, one-pagers, and the assets sellers use inside a live deal. Ownership rose from 64% to nearly 79% of PMMs in a year (PMA 2025).

These are not six independent projects. They are one connected system, and treating them as separate checkboxes is how PMM teams end up busy and irrelevant at once. Research surfaces what a rival just claimed, which tells positioning what hill is still open, which tells messaging what words to use, which is what shows up on the battle card a rep pulls up mid-call and in the campaign that carries the launch. Break one link and everything downstream goes stale even if the people doing the work are working hard. A battle card built on six-month-old positioning is not a small problem. It is a rep repeating a claim the competitor already neutralized. That same chain, run continuously instead of once a quarter, is what a connected set of AI agents does well, a point this guide returns to below.

It is worth being blunt about what matters versus what is theater. The endless stream of one-off requests from sales for "just one more" one-pager or slide, each pitched as the thing that will finally close the deal, mostly does nothing. Often no one on the buying committee even opens it, and a slide does not close a B2B deal anyway. What actually moves the business is slower, and nobody asks for it loudly: deep enablement, real training on how to pitch the company, genuine depth on both the buyer and the user persona, and research that goes past the first page of a competitor's website. Once a seller has internalized the story, they can tell it unscripted and win the deal with no slide in sight. Nobody requests that the way they request a one-pager, because it asks something of the sales side too. But B2B deals were never won by the right slide arriving five minutes before the call.

Product marketing vs product management: what's the difference?

Product management decides what gets built and why. Product marketing decides who it's for, why it's different, and how it wins in the market. PM is internally focused, working the roadmap with engineering. PMM is externally focused, working positioning and go-to-market with the market. Both do research; they just do it for different decisions.

	Product management	Product marketing
Owns	What gets built and why	Who it's for and how it wins
Focus	Internal: roadmap, requirements, engineering	External: market, positioning, GTM
Core question	"Is this worth building?"	"How does the market understand this?"
Shared ground	Market research, customer interviews, competitive analysis, market sizing	Same

That is the textbook version, and it is where most definitions of the role stop, usually with some version of "PM builds, PMM sells," which flattens a messier relationship into a slogan. It gets more useful when extended one step. Product strategist Ant Murphy frames PM as deciding what to build, PMM as deciding how it wins, and growth PM as optimizing the funnel between them (Ant Murphy, PM vs PMM vs Growth PM). That third role is where the line blurs in practice, and a lot of the friction in this triangle is really a fight over who owns that middle ground.

One caveat sits under all of this: product marketing is a young function, and a still-standardizing one. It is newer and less codified than product management or demand generation, and its remit swings hard from company to company. What a PMM owns at a fifty-person startup barely resembles what one owns at a five-thousand-person enterprise. So treat any "here is what product marketing does" description, including this one, as the common core of the role, not a fixed job spec. Where the line falls between PM and PMM, and how much of that contested middle ground each one owns, is something every company settles differently.

Where does product marketing sit? Team size, reporting line, and salary

Product marketing most often reports into marketing, and Forrester's recommendation is explicit: it should sit in marketing, under the CMO (Forrester: Where Should Product Marketing Report?). In practice the pattern is looser, and that looseness is the point. Because the function is young and its maturity varies so widely between companies, where it sits is one more thing every org decides differently. Startups frequently place it under product, and at early-stage companies it isn't unusual for PMM to report straight to the CEO, because positioning is an existential question before there's a marketing org big enough to house it.

Teams are usually small enough that "team" is generous: 44.3% of PMM functions are one or two people (PMA 2025). That is not a rounding error, it is the median experience of doing this job. The pay reflects a wide seniority band more than that small headcount: in the US, average pay runs from about $127K (Salary.com) to $141K (Glassdoor) as of 2026, with entry-level closer to $100K and senior roles well above that. Well-paid, and chronically stretched thin.

On a small team, that scarcity has a specific texture, and it is not the one most people assume:

When I ran a three-person PMM team, nothing ever got dropped for good. It just never got done at the same time. One quarter I would chase every competitor move and rebuild the battle cards from scratch. The next I would pull positioning apart because the old version had stopped matching how we actually won deals. Whatever was not that quarter's fire went quiet and aged. It was never a gap in the org chart. It was a permanent backlog that kept getting older no matter how hard we worked, because there was always exactly one fire at a time and everything else waited its turn.

— David Kolínek, Ex VP of Product Marketing, currently co-founder at Calven

None of this is about bad management. It is a structural fact about a function that is understaffed relative to what it is accountable for, which is exactly the condition that makes the next section matter.

How is AI changing product marketing?

AI now does a meaningful share of the gathering, drafting, and monitoring that used to eat a PMM's week. Adoption is close to universal: 93% of PMMs use AI to improve productivity and 92% say it helps them create content faster (Fluvio 2025 Product Marketing AI Trends Report, n=40). But 72% of frequent users name inconsistent or low-quality output as their main concern (same report). That gap between adoption and trust is the real story, not the adoption number by itself.

The rest of the Fluvio numbers describe a discipline that adopted AI everywhere at once rather than in one careful pilot: 100% use it for market research, 92% for messaging, 82% for competitive analysis. Yet only 15% call their usage deeply embedded, and just 28% provide any training on it. A discipline racing ahead of its own guardrails is exactly what produces that 72% inconsistency number.

That inconsistency has a shape, and the shape is context. A generic assistant answers the prompt in front of it and forgets everything the moment the conversation ends. Ask it for a competitor update on Monday and a positioning refresh on Wednesday, and it treats them as unrelated jobs, even though the second should have been shaped by the first. That is the actual mechanism behind "inconsistent output": not that the model writes badly, but that it has no memory of the decision it made yesterday.

Purpose-built PMM agents are a different category, not a better prompt. The distinction is architectural: they carry context between tasks, so a competitive-intelligence agent's finding informs the messaging agent's next draft, and a competitor's pricing change flags exactly which battle cards need a look instead of sitting undiscovered until a rep gets burned by it in a call. That is the shape Calven's Intelligence Core runs on: research feeds audience work, audience work feeds messaging.

Will AI replace product marketers?

No. AI is taking over the execution: the first-pass research, the first draft, the daily competitive scan nobody had time to run consistently before. What it cannot take over is the judgment call about what is actually true and differentiated about a product, because that requires understanding a market well enough to know what it will and will not believe. That is not a gathering problem or a drafting problem. It is a taste problem, and taste does not get automated.

Where the role goes from here is its own piece, and we wrote it: how AI is changing product marketing walks through the four shifts reshaping the job and why none of them point to replacement. The short version is straightforward: AI is likely to be one of the best things that has happened to product marketing, not a threat to it. This has always been an understaffed function forced into phased triage, doing everything eventually but never all at once. Agents that take the grind off a two-person team do not shrink the job. They let that team finally produce at the volume the rest of the company always assumed it could. And when a PMM team's output visibly jumps, so does the credit, which is the recognition this discipline has been owed for years. Bigger output, better judgment work, more respect. That is a good trade.

AlphaSense vs Kompyte (2026): Two Different Jobs, One PMM Gap

David Kolinek — Mon, 29 Jun 2026 00:00:00 GMT

AlphaSense is a market intelligence tool, not a competitive intelligence tool. Kompyte is the CI tool here: a lightweight one built to keep a rep's battle card current. They land on the same shortlists because both get filed under competitive intelligence, a label that stretched as the category grew, and Gartner folded competitive and market intelligence into a single 2026 Magic Quadrant, though only AlphaSense is named in it. That label is what makes an analyst-grade research platform, priced for a finance desk, look like a Kompyte alternative. It is not, and a product marketer weighing one against the other is usually answering the wrong question. This piece is clear about what AlphaSense does and does not do, then turns to the step neither tool takes.

What's the actual difference between AlphaSense and Kompyte?

The difference is the job, not the feature set. AlphaSense is market and financial research intelligence: 500 million-plus premium documents, generative search, built for an analyst sizing a market or dissecting a company. Kompyte is automation-first sales-enablement CI: it tracks rivals and keeps battle cards fresh for reps, cheaper and narrower. Different buyers, different work, and only the thinnest shared ground.

Analyst recognition draws the same line. AlphaSense was named a Leader in the inaugural 2026 Gartner Magic Quadrant for Competitive and Market Intelligence Platforms. Kompyte has no publicized placement: Gartner referenced several dedicated CI vendors, but not all disclosed their positions, and Kompyte and its owner have announced none. Read it as a scope signal, not a scoreboard.

Dimension	AlphaSense	Kompyte
Category	Market and financial research intelligence	Sales-enablement competitive intelligence
Primary buyer	Analysts, corporate strategy, finance, consultants	Product marketing, sales enablement, growth marketing
Core job	Search and synthesize premium research at analyst grade	Track rivals and keep battle cards current for reps
Analyst standing (2026)	Gartner Leader, C&MI Magic Quadrant	No publicized MQ placement
Ownership	Independent, ~$7.5B valuation	Adobe, via Semrush
Price altitude	Five figures per seat	Gated, lighter than the dedicated CI platforms

For a PMM, that resolves into a clean call about which job the budget is actually for.

If this is you	Lean
You need deep market and financial research: filings, earnings transcripts, expert calls, broker research	AlphaSense
You need battle cards kept fresh and reps prepped against named rivals	Kompyte
You are sizing a market or taking a company apart, not arming a rep for a call next week	AlphaSense
You want automated competitor monitoring that a marketing team can run on its own budget	Kompyte

What does each one actually do?

They do almost non-overlapping work, so the honest picture is two separate columns, not a scorecard. AlphaSense searches a research library that dwarfs anything a CI tool holds; Kompyte watches competitors and auto-updates battle cards for sellers. Where one has depth, the other has a blank.

AlphaSense points at the research desk. Its library spans more than 500 million documents: SEC filings, earnings transcripts, broker and equity research from 1,700-plus providers, plus the expert-call transcript library it gained by acquiring Tegus in 2024, now more than 250,000 interviews. Generative search, Deep Research, and its Generative Grid run natural-language queries across all of it and return citation-linked answers. This is research infrastructure for a hedge-fund analyst or a corporate strategy team. The telling part for a PMM: AlphaSense's own product materials describe no sales battle cards, no win-loss, and no rep delivery. That is a verified absence, not an oversight.

Kompyte points at the rep. It runs automated tracking of competitor websites, pricing pages, ads, review sites, and job posts, then refreshes cards as it detects new data and pushes them into Slack, Microsoft Teams, Salesforce, and HubSpot, leaning on Semrush's web and ad data since the 2022 acquisition. Two caveats. The "auto-updates, no manual work" framing is Kompyte's own marketing, and independent reviews report the cards still need real editing to be usable. And it is the mirror of AlphaSense's blank: reviewers flag that it lacks market-intelligence depth, so where AlphaSense reads a company's filings, Kompyte reads its pricing page.

Job	AlphaSense	Kompyte
Search over filings, earnings transcripts, broker research	Core strength	Not offered
Expert-call transcript library	Core strength	Not offered
Generative search across premium documents	Core strength	Narrower, CI-scoped
Automated competitor monitoring: sites, pricing, ads	Not the focus	Core design
Auto-updating battle cards delivered to reps	Not offered	Core design
Win-loss as a discipline	Not offered	Limited, partner-dependent

Blanks on both sides. That is what different jobs look like on a page.

How much do AlphaSense and Kompyte cost?

The price gap is wide, and both vendors are gated, so neither publishes a list price. AlphaSense is the trickier read. Its reported median contract on Vendr sits near $17.5K, which looks modest until you notice it is priced per seat, commonly around $10K to $20K each. A five-seat research team therefore runs roughly $50K to $100K+ per year, and enterprise deployments reach well past that, with the largest deals cited above $1M. Treat those as third-party, negotiated figures, not an official rate.

Kompyte has thinner public data. It is consistently described as the more affordable option, priced below Klue and Crayon and often absorbed into existing marketing-ops or PMM budgets rather than a dedicated CI line. The third-party estimates that circulate conflict, and none are Vendr-verified, so no specific Kompyte figure belongs in a buying decision. What is safe to say: it is the cheaper, lighter tool.

Dimension	AlphaSense	Kompyte
Pricing model	Quote-only, priced per seat	Quote-only, sold through Semrush
Reported median (Vendr)	~$17.5K/yr per contract	No public median found
What a real team pays	~$50K-$100K+/yr, enterprise past $1M	Reported below the CI platforms; figures gated
Priced for	A research desk	A marketing budget

Source: Vendr marketplace data, AlphaSense, 2026. These two are built, and priced, for different departments. When you evaluate a competitive intelligence tool, the altitude alone tells you which job the vendor built for.

What do customers say about AlphaSense and Kompyte?

Both review well, and the complaints track their designs. AlphaSense sits around 4.6 on G2 from roughly 308 reviews (worth confirming against the live listing), praised for research-search quality and content breadth; its number-one complaint by a wide margin is cost, trailed by a steep learning curve and the recurring verdict that it is over-scoped for sales. That last one reads less like a defect than the market confirming what AlphaSense is for.

Kompyte reviews solidly too, a notch below the enterprise research platforms, praised for fast setup and automation that cuts manual research time. Its complaints are more useful than its rating: reviewers report it sometimes pulls parent-company data instead of the direct competitor, misses smaller rivals, and, the one a PMM should weigh hardest, produces cards that still need manual editing to be usable. That last complaint is the same gap the automation marketing papers over.

What do AlphaSense and Kompyte both leave undone?

Neither one turns its output into positioning, personas, and persona-tested messaging. That is the shared gap, a scope point rather than a knock on either tool doing its own job. AlphaSense stops at the research answer, Kompyte one step downstream at the battle card. A competitor cuts prices, and both deliver the input; then a person decides what the cut means for the positioning, rewrites the value props it weakened, checks whether it shifts a buyer persona, and refreshes the messaging sales repeats in every call. The assets that go to market still get assembled by hand after both tools have done their part.

When a team puts these two in the same evaluation, the real problem is usually the question, not the tools. One is built for a research desk sizing a market. The other keeps a rep's battle card current. Neither one decides what a competitor's move means for your positioning, and that decision is the work a product marketer actually owns.

— David Kolínek, Co-founder at Calven

The scale of that gap is measured, not theoretical. Roughly 91% of PMMs own positioning and messaging, versus about two-thirds who own competitive intelligence, per the Product Marketing Alliance 2025 State of Product Marketing, and 58% of CI pros say keeping battle cards and content updated is a struggle, per Crayon's 2024 State of Competitive Intelligence. The downstream work is the bigger share of the job, and the share both tools hand back.

For a PMM, the two miss in opposite directions. AlphaSense is over- and under-scoped at once: analyst-grade research depth priced for a finance desk, far more than a competitive-intelligence job needs, and no battle cards, win-loss, or messaging output at the other end. Kompyte is narrower still, a battle-card point tool reviewers flag as thin on market intelligence, now one component inside a large marketing suite (Kompyte to Semrush to Adobe). Competitive intelligence is one slice of what product marketing owns, and a signal that should reach the positioning, value props, and personas instead reaches a research answer or a feed, and someone copies the rest into a doc that is stale by the next release.

Where does Calven fit?

Calven is an AI product marketing platform that runs the whole product marketing surface, not a CI tool or a research library bolted onto the rest of the stack. It overlaps more with Kompyte than with AlphaSense. Its Competitive Intelligence Agent tracks competitor sites, docs, and pricing and keeps battle cards current, its Win/Loss Agent runs win-loss as a discipline, and its Market Research Agent reads market and analyst signals with less depth than AlphaSense brings to the same question.

What neither of them does is carry the signal forward. Take the same price cut. Instead of stopping at a card or a citation, it flows into the assets a PMM owns: the Positioning Agent and Messaging Agent revisit the value props the cut weakened, pressure-tested against the buyer personas it would affect. Whether that update lands as a draft for your review or propagates straight to the live assets is a setting you control. The practical difference is timing: the messaging sales repeats is current for the next call, not three weeks and one stale doc later.

AlphaSense wins research breadth, Kompyte wins cheap automated battle cards, and each is the right buy when that narrow thing is the job you are funding. Calven's claim is different: it finishes the last mile both leave undone. The real question is whether the budget is going to a research desk, a battle-card tool, or the product marketing work that turns either signal into messaging that is live in the deal. For the build-it-yourself version of that last mile, ChatGPT and Claude for competitive intelligence is the next read.

Crayon vs AlphaSense (2026): Two Different Jobs, One PMM Gap

David Kolinek — Mon, 29 Jun 2026 00:00:00 GMT

AlphaSense is a market intelligence tool, not a competitive intelligence tool. Crayon is the CI tool here: it tracks your named rivals and keeps your reps armed with battle cards. They land on the same shortlists for one reason: Gartner's inaugural 2026 Magic Quadrant is titled Competitive and Market Intelligence Platforms, and grouping those two categories under one banner makes AlphaSense look like a Crayon alternative. It is not. AlphaSense is an analyst-grade research platform built for hedge funds and corporate strategy desks, sitting on more than 500 million premium documents and a quarter-million expert-call transcripts. Buy one when you needed the other and the mistake shows up on the invoice. This piece is clear about what AlphaSense provides and what it does not, then gets to the job both tools stop short of.

What's the actual difference between Crayon and AlphaSense?

They serve different jobs for different buyers, and the overlap is one shared word. Crayon is a competitive-and-market-intelligence platform for product marketing and sales: it watches rivals and keeps reps armed. AlphaSense is a market, financial, and research intelligence platform for analysts and strategy teams: it searches premium content at depth and cites its answers. The honest starting point is not "which is better" but "which job are you hiring for."

Dimension	Crayon	AlphaSense
Category	Competitive and market intelligence	Market, financial, and research intelligence
Core job	Track named rivals, arm reps	Search and synthesize premium research at analyst depth
Primary buyer	Product marketing, competitive enablement, sales	Analysts, corporate strategy, IR, finance, consulting
Signature asset	Auto-updated battle cards delivered to reps	Citation-linked answers over filings, transcripts, broker research
Content base	Public web signals and first-party field intel	500M+ premium docs, 250K+ expert transcripts
Analyst recognition (2026)	Gartner Leader	Gartner Leader
Price altitude	~$30K/yr total	Five figures per seat

That difference makes the call cleaner than most head-to-heads.

If this is you	Lean
PMM tracking competitors, arming reps, keeping battle cards current	Crayon
Deep market, financial, or company research: filings, earnings calls, expert interviews, buy-side depth	AlphaSense

For most product marketing teams running a competitor analysis and keeping sales current, this is a Crayon-shaped job, not an AlphaSense-shaped one. AlphaSense is what a strategy or research desk reaches for when the question is about a whole market or a specific company's financials, not about winning next quarter's deal against a named rival.

What does each one actually do?

Where one has a feature, the other usually has a blank, which is why a straight scorecard would mislead. Crayon aggregates competitor signals and turns them into rep-ready cards. AlphaSense searches a premium research library and writes cited answers. The table below is not a parity grid: the point is how little the two columns touch.

Capability	Crayon	AlphaSense
Competitor monitoring	Core: 100+ data types across 300M+ sources (Crayon's claim)	Company and topic alerts, research-oriented, not battle-card
Battle cards	Native, auto-updated, delivered to reps	None
Win-loss	Field Agent capture plus Salesforce win/loss	None
Rep delivery (Slack, CRM, call tools)	Core	None
Broker and equity research	None	1,700+ providers
Earnings-call transcripts	None	Yes, with AI summaries
Expert-call library	None	250K+ transcripts
SEC filings and financial data	None	Yes
Generative AI search	Crayon Answers, competitive Q&A	Generative Search and Deep Research over premium content

Crayon's strength is breadth of automated capture. It pulls from a very wide net of public sources, scores the signals, and pushes them into a newsfeed, into Salesforce and Slack, and into battle cards that update as the field moves. Its newer AI, the Sparks summaries and Crayon Answers assistant, sits on that feed to tell a rep what a competitor's move means in a deal. The whole product points at one outcome: a seller walking into a competitive call prepared.

AlphaSense is a different animal: generative search over roughly 500 million premium documents (filings, earnings transcripts, broker research from more than 1,700 providers) plus the expert-interview library it gained by acquiring Tegus for $930 million in 2024, now more than 250,000 interviews. It answers the questions a research analyst asks, with citations back to the source, well enough to be valued around $7.5 billion. What it does not do is telling: its own materials carry no battle cards, no win-loss discipline, and no rep-delivery surfaces. Point it at a sales-enablement job and you are using a microscope to hammer a nail. That is a category boundary, not a knock on the platform.

How much do Crayon and AlphaSense cost?

The price gap is as wide as the category gap, and it runs in AlphaSense's direction. Both are gated, so neither prints a list price, and the figures below come from third-party marketplace data rather than either vendor. Read the ranges as roughly, not to the dollar.

Dimension	Crayon	AlphaSense
Pricing model	Gated, quote-based	Gated, quote-only, per seat
Median (Vendr)	~$30K/yr total	~$17.5K/yr contract
Typical range	~$12.7K-$46K	~$9K-$51K
Per seat	Bundled into the total	~$10K-$20K per seat
What a real team runs	~$15K-$46K	~$50K-$100K+
Enterprise top	Past $100K	$1M+

The medians look closer than the reality. Crayon's roughly $30K is a whole-team number, per Vendr's marketplace data. AlphaSense's roughly $17.5K median is a single contract, and because the platform is priced per seat at a reported $10K to $20K each, a five-seat research team runs $50K to $100K a year and the largest deployments pass $1 million. That is two to three times a typical Crayon deployment for a job most product marketing teams do not need done. When you evaluate a competitive intelligence tool, the AlphaSense number is the tell: it is priced for a research desk, not a marketing team.

What do customers say about Crayon and AlphaSense?

Both are highly rated, and the ratings say less than the complaint patterns do. On G2, both sit around 4.6, Crayon across roughly 385 reviews and AlphaSense across roughly 308, though those counts come from secondary citations rather than a live read. What matters more is that each tool draws a consistent complaint that follows directly from its design.

Dimension	Crayon	AlphaSense
G2 rating (reported)	~4.6 / 5	~4.6 / 5
Most praised	Breadth of capture, customer success, real-time alerts	Search quality on financial content, breadth, support
Most common complaint	Alert noise, heavy tuning to separate signal from volume	Cost, steep learning curve, over-scoped for GTM
The theme underneath	Captures everything, so a person has to filter it	Built for deep research, so it overwhelms a lighter job

Crayon's complaint is noise: because it tracks everything, teams get flooded with low-value alerts unless someone tunes them carefully, and that tuning is ongoing work. AlphaSense's complaint is cost and overkill, its most-cited frustration by a wide margin, tangled up with a learning curve new users describe as being lost in too many features. Neither pattern is disqualifying. They are the predictable trade-offs of a wide-net enablement tool versus an analyst-grade research platform.

What do Crayon and AlphaSense both leave undone?

Both stop one step short of the assets a product marketer actually ships. Crayon delivers a scored signal and a current battle card; AlphaSense delivers a cited answer about a market or a company. Neither rewrites your positioning, re-tests the value props against a buyer persona, or refreshes the messaging your reps repeat in every call. That translation still happens by hand, downstream, in a document that goes stale by the next release.

A competitor's price cut or earnings call is an input, not a decision. Someone still has to work out what it means for the positioning, rewrite the value props it undercuts, and re-test the messaging against the buyer. Research tools and CI tools both stop at the edge of that work. The translation is the PMM job, and it is the part that still happens by hand in a doc that's stale by the next release.

— David Kolínek, Co-founder at Calven

The scope point is measured, not theoretical. Per Crayon's own 2026 State of Competitive Intelligence, 57.5% of teams report more competitive deals than a year ago, yet rep readiness averages just 6.3 out of 10, and in Crayon's 2024 report, 58% of CI pros said keeping battle cards and content updated was a struggle. Competitive intelligence is one slice of what product marketing owns: roughly two-thirds of PMMs own CI, versus 91% who own positioning and messaging, per the Product Marketing Alliance 2025 State of Product Marketing Report. A tool that ends at the signal owns none of that last mile, which is why the battle card goes stale between the alert and the update.

Where does Calven fit?

Calven is an AI product marketing platform: one system that runs the whole product marketing surface rather than a point tool bolted onto the rest of the stack. It overlaps with Crayon on the core competitive ground. Its Competitive Intelligence Agent tracks competitor sites, docs, pricing, and announcements, writes the dossiers, and keeps battle cards current; its Win/Loss Agent runs win-loss as a discipline; and its Market Research Agent tracks category shifts and reads analyst reports, though it does not pretend to match AlphaSense's depth of filings and broker research.

The difference is what happens to the signal after it lands. Take the same price cut. Instead of stopping at a card or a citation, it flows into the assets a PMM owns: the Positioning Agent and Messaging Agent revisit the value props the cut undercuts, drawing on the personas and ICP work, and the Voice of Customer Agent checks the new line against the language buyers actually use. The move reaches the positioning and the persona-tested messaging, not just a newsfeed.

So the real question is not only which of these two to buy. It is whether either one is buying you the job you actually need done. Crayon wins competitor monitoring and rep enablement at a fraction of the price; AlphaSense wins depth of premium research and is worth every dollar to the desk that lives in filings and expert calls. Both hand the last mile back to a person. For a team weighing whether to build that last mile out of general tools instead, ChatGPT and Claude for competitive intelligence is the adjacent decision.

Crayon vs Kompyte (2026): Where Each Wins and What Both Miss

David Kolinek — Mon, 29 Jun 2026 00:00:00 GMT

The Crayon and Kompyte shortlist comes down to one trade-off: how much competitive intelligence machinery you want to run, and how much you want to pay for it. Crayon is the broad, analyst-recognized platform built to monitor almost everything. Kompyte is the lighter, automation-first option that keeps battle cards fresh without a dedicated CI analyst, and since April 2026 it sits inside Adobe by way of Semrush. The tables below run the head-to-head. Then the piece turns to the question both shortlists skip: whether a CI tool is the right thing to buy at all.

What's the actual difference between Crayon and Kompyte?

Crayon is the broader, more established platform; Kompyte is the lighter, cheaper, automation-first one. Crayon leads on monitoring breadth, market intelligence, and analyst pedigree, and assumes an owner to curate what it captures. Kompyte leads on automating the battle-card grind at a lower price, and assumes that work gets absorbed into an existing PMM role rather than a dedicated CI hire.

Dimension	Crayon	Kompyte
Category framing	Market and competitive intelligence	Competitive intelligence and sales battle cards
Primary audience	CI and PMM analysts, cross-functional	Sales enablement and PMM, SMB to mid-market
Core strength	Breadth of automated monitoring	Automation-first, auto-updating battle cards
Ownership	Independent	Adobe, via Semrush
Analyst recognition (2026)	Gartner Leader	No publicized MQ placement
Best fit	Cross-functional CI with a curator	Lighter CI without a dedicated analyst

That resolves into a fairly clean call on who each one is for.

If this is you	Lean
Broad market and competitor monitoring, and you have someone to curate it	Crayon
Automated battle cards without a dedicated CI analyst, and price is deciding	Kompyte
Analyst-recognized pedigree matters to your buying committee	Crayon
You already run on Semrush and want CI as an extension of it	Kompyte

On the core discipline itself, running a structured competitor analysis and keeping it current, neither is a weak choice.

How do Crayon and Kompyte compare on product and features?

Both monitor competitors and keep battle cards current; they diverge on scope and on how much they claim to automate. Crayon casts the widest net it can, more than 100 data types across 300M+ sources by its own account, and pairs it with market intelligence. Kompyte automates aggressively on a narrower surface, leaning on Semrush's web, ad, and SEO data.

Feature	Crayon	Kompyte
Competitor monitoring	Broadest reach, 100+ data types across 300M+ sources (Crayon's claim), catches subtle page changes	Automated tracking of sites, pricing pages, ads, reviews, and job posts, built on Semrush's data
Battle cards	Native, auto-updated	Native, AI auto-updated (Kompyte's claim); reviewers say real cards still take manual effort
Delivery surfaces	Newsfeed, Salesforce, Slack, Microsoft Teams, Highspot	Slack, Microsoft Teams, Salesforce, HubSpot
Win-loss	Field-intel capture, pinged in Slack and linked to Salesforce	Thin, routed to a third-party partner rather than native
AI (2026)	Sparks and Answers, signal summaries and an in-tool assistant	Battle-card updater and daily summaries; no conversational assistant
Market intelligence	Part of the pitch	Limited, reviewers flag it as missing

The split follows each tool's design. Crayon captures more and gives you an assistant to interrogate it. Kompyte captures less and tries to turn what it captures into a card with less human touch, though the win/loss analysis side is where it is thinnest, since it routes that work to a partner instead of owning it.

How do Crayon and Kompyte compare on services and support?

Both are well supported and invest in different things beyond the software. Crayon publishes the category's longest-running industry benchmark and sells professional services around the platform. Kompyte pairs onboarding with a dedicated CSM and routes deeper research, including win-loss interviews, through a partner.

Dimension	Crayon	Kompyte
Support reputation	Well reviewed, praised for smooth rollout	Positive, sold by demo and run with a dedicated CSM
Time to value	Weeks to implement, full value needs ongoing tuning	Lighter setup, though no clean public figure exists
Where the effort lands	Ongoing, in tuning to fight alert noise	Ongoing, in feed and battle-card upkeep, absorbed into an existing role
Managed win-loss	Not a managed service	No in-house service, routed to a third-party partner
Signature research	State of Competitive Intelligence, the category's longest-running report	None

The one dimension the software tables miss is who stands behind each product, and it cuts both ways. Semrush acquired Kompyte in 2022, and Adobe completed its roughly $1.9B acquisition of Semrush in April 2026, so Kompyte now sits inside Adobe through Semrush, rarely sold standalone. For a multi-year commitment, the absence of a public post-Adobe roadmap is a fair continuity question. The other side is real: Adobe and Semrush bring deep web and ad data and the balance sheet of a large platform. Crayon, by contrast, is an independent, single-focus vendor. Neither ownership model is safer in the abstract; a buying committee should price both.

How much do Crayon and Kompyte cost?

Kompyte is cheaper, and both gate their pricing. Crayon's public data is firmer: Vendr's marketplace puts its median at roughly $30K/yr. Kompyte has no comparable Vendr median, and every credible source describes it as priced below Crayon at similar deployment sizes rather than quoting a trustworthy figure.

Dimension	Crayon	Kompyte
List pricing	Gated	Gated
Median (Vendr)	~$30K/yr	Not published
Typical range	~$12.7K-$46K	Below Crayon at comparable size
Entry point	Essentials ~$15K-$25K	Lighter-priced, per-competitor and per-seat
Top of range	Enterprise past $100K	Gated
Avg negotiated savings	~20%	Not published
Billing	Independent vendor	Through Semrush, now Adobe

Source: Vendr marketplace data, February 2026, for the Crayon figures; Kompyte's are gated and quote-based. The number neither vendor prints on the quote is the human. When you evaluate a competitive intelligence tool, the fully loaded cost is the tool plus the owner it assumes, and Kompyte's lower sticker partly reflects that it expects that owner to already be on your payroll.

What do customers say about Crayon and Kompyte?

Both are well rated, and Crayon reviews a little higher. On G2, Crayon sits around 4.6 across roughly 385 reviews, and Kompyte lands a few tenths below. Each tool earns praise for what it is built around and draws a consistent complaint that follows from its design.

Dimension	Crayon	Kompyte
G2 rating	~4.6 / 5 (around 385 reviews)	A few tenths below Crayon
Most praised	Breadth of automated capture, customer success, real-time alerts	Intuitive interface, fast setup, automation cutting research time
Most common complaint	Alert noise, heavy tuning to separate signal from volume	Battle cards still need manual work, misses smaller rivals, sometimes pulls parent-company data
The theme underneath	Captures everything, so a person filters it	Automates the easy part, so the depth stays manual

Neither pattern is disqualifying. They are the predictable trade-offs of a wide-net monitoring tool versus an automation-first one, both worth weighing against the readiness gap battle cards are supposed to close.

What do Crayon and Kompyte both leave undone?

Both stop at the feed and the battle card, and for a product marketing team that is the first step, not the deliverable. Roughly two-thirds of PMMs own competitive intelligence, versus 91% who own positioning and messaging and about 79% who own sales enablement, per the Product Marketing Alliance 2025 State of Product Marketing. CI is a real slice of the job, but not the slice where most owned assets live.

Watch a single signal. A competitor cuts prices; the alert fires and the battle card gets a note. Then a person decides what the move means for positioning, rewrites the value props it undercuts, checks whether it shifts a buyer persona, and refreshes the messaging sales repeats in every call. The tool delivered the alert; the translation lives outside it. For Kompyte that gap is wider than for Crayon, because reviewers flag that it lacks market intelligence altogether.

A monitoring tool tells you a competitor moved. It cannot tell you what that means for your positioning, and it will not rewrite the messaging your reps repeat in every call. That decision is the actual product marketing work, and it still lands on a person. A tool that only fills the feed leaves the hardest part on your desk.

— David Kolínek, Co-founder at Calven

That gap is measured, not theoretical. 58% of CI pros said keeping battle cards and content updated is a struggle, per Crayon's 2024 State of Competitive Intelligence, and 57.5% of teams report more competitive deals than a year ago, per Crayon's 2026 report. More deals, harder to keep current, and the downstream work still done by hand. This is a scope point, not a knock on either tool doing CI well: competitive intelligence is one slice of what product marketing owns, and a CI-first platform does not own that rest.

Where does Calven fit?

Calven is an AI product marketing platform: one system that runs the whole product marketing surface, competitive intelligence included, rather than a dedicated CI tool bolted onto the rest of the stack. It overlaps with Crayon and Kompyte on the core ground. Its Competitive Intelligence Agent tracks competitor sites, docs, pricing, and announcements, writes dossiers, and keeps battle cards current, and its Win/Loss Agent runs win-loss as a discipline. That is the same job Crayon and Kompyte do.

The difference is what happens to the signal after it lands. Take that same price cut. Instead of stopping at a card, it flows into the assets a PMM owns: the Positioning Agent and Messaging Agent revisit the value props the cut undercuts, drawing on the persona and ICP work, and the Voice of Customer Agent checks the new line against the language buyers actually use. One signal reaches the positioning and the persona-tested messaging, not just a newsfeed.

Because it spans the whole surface, Calven is a different kind of bet from a point tool. It is also a young product, and that cuts both ways: moving fast means it gains ground quickly, but it also changes more than a tool that has held the same shape for a decade. For a team that wants one platform to run the whole surface and values that pace, the trade is worth making.

So the real choice is not only which competitive intelligence tool to buy. It is whether competitive intelligence is the whole job you are buying for. For a team weighing whether to build that last mile out of general tools instead, ChatGPT and Claude for competitive intelligence is the adjacent decision.

Klue vs AlphaSense (2026): Two Different Jobs, One PMM Gap

David Kolinek — Mon, 29 Jun 2026 00:00:00 GMT

AlphaSense is a market intelligence tool, not a competitive intelligence tool. Klue is the CI tool here: it arms sales reps with battle cards and runs win-loss as a discipline. The two get compared for one reason: Gartner's inaugural 2026 Magic Quadrant is titled Competitive and Market Intelligence Platforms, and folding those two categories into a single report puts AlphaSense on the same shortlists as Klue, Crayon, and Kompyte. They are not the same category. AlphaSense searches more than 500 million premium documents so an analyst can take a market or a company apart; Klue helps a rep win a specific deal next quarter. A product marketer weighing one against the other is not comparing two versions of the same tool. The real decision is which job to fund. This piece is clear about what AlphaSense does and does not do, then turns to the step neither tool takes.

What's the actual difference between Klue and AlphaSense?

The difference is category, not quality. Klue is a sales-enablement competitive intelligence tool built to get competitive intel to reps and run win-loss. AlphaSense is an analyst-grade market and financial research platform built to search filings, earnings transcripts, and broker research. One is where you go to win a specific deal. The other is where you go to understand a market or a company in depth.

That shows up cleanly in the same Gartner evaluation that groups them. Both were named Leaders in the inaugural 2026 Gartner Magic Quadrant for Competitive and Market Intelligence Platforms. Klue ranked #1 for Revenue Enablement in the companion Critical Capabilities report, the enablement use case AlphaSense does not compete in. AlphaSense, by its own account, was placed highest on Ability to Execute and furthest on Completeness of Vision for research breadth. Same quadrant, two different centers of gravity.

Dimension	Klue	AlphaSense
Category	Sales-enablement competitive intelligence	Market and financial research intelligence
Primary audience	Product marketing, competitive enablement, sales reps	Analysts, corporate strategy, finance, consultants
Core strength	Battle cards and rep delivery, native win-loss	Generative search over 500M+ premium documents
Signature asset	Auto-updated battle cards in the rep's workflow	Citation-linked answers over filings, transcripts, broker research
Analyst standing (2026)	Gartner Leader, #1 Revenue Enablement	Gartner Leader, research breadth and depth
Price altitude	~$30K/yr total (Vendr median)	Five figures per seat, a small team ~$50K-$100K+/yr

For a PMM, that resolves into a fairly clean call about which job you are actually buying for.

If this is you	Lean
Sales enablement, battle cards, and win-loss are the point, and reps need intel in the deal	Klue
You need deep market and financial research: filings, earnings transcripts, expert calls, broker research	AlphaSense
You want to run win/loss analysis as a discipline	Klue
You are sizing a market or a category, not arming a rep for a call next week	AlphaSense

What does each one actually do?

They do almost non-overlapping work, so the honest picture is two separate columns, not a scorecard. Klue captures competitive signals and gets them to sellers. AlphaSense searches a research library that dwarfs anything a CI tool holds. Where one has depth, the other has a blank.

Klue's surface is the rep in the deal. Battle cards auto-update and get delivered where sellers already work: Slack, Salesforce, HubSpot, call tools like Gong, and enablement platforms. Its Compete Agent runs in the background, detects which competitor is in a deal, and pushes talk tracks and objection handling. And win-loss is native rather than bolted on, backed by Klue's DoubleCheck managed research service. The whole design points at one outcome: the rep knows what to say against a named rival.

AlphaSense points somewhere else entirely. Its library spans more than 500 million documents: SEC filings, earnings transcripts, broker and equity research, plus the expert-call transcript library it gained by acquiring Tegus in 2024, now more than 250,000 interviews. Generative search, Deep Research, and its Generative Grid run natural-language queries across all of it and return citation-linked answers. This is research infrastructure for a hedge-fund analyst or a corporate strategy team, built for a research desk's daily work.

Here is the part a feature grid would hide. AlphaSense's own product materials describe no sales battle cards, no win-loss analysis, and no rep-delivery surfaces. That is a verified absence, not an oversight: a superb research instrument pointed at a task it was not built for. The two tools do not sit on a spectrum; they answer different questions.

Job	Klue	AlphaSense
Auto-updated battle cards delivered to reps	Native	Not offered
Win-loss as a discipline	Native, plus managed research	Not offered
Rep delivery in Slack, CRM, call tools	Core design	Not offered
Search over filings and earnings transcripts	Not the focus	Core strength
Broker and equity research, expert-call library	Not offered	Core strength
Generative search across premium documents	Narrower, CI-scoped	Core strength

How much do Klue and AlphaSense cost?

They sit at genuinely different price altitudes, and the gap is a real decision factor. Both are gated, so neither publishes a list price, and every number below is a third-party estimate rather than an official rate.

Klue lands at roughly $30K/yr total on Vendr's transaction data, with a typical range around $16K to $60K. AlphaSense is the trickier read. Its reported median contract on Vendr sits near $17.5K, which looks cheaper until you notice it is priced per seat, commonly around $10K to $20K per seat. A five-seat research team therefore runs roughly $50K to $100K+ per year, and enterprise deployments reach well past that, with the largest deals cited above $1M. So the headline median understates it: for any real team, AlphaSense is the more expensive line by a wide margin.

Dimension	Klue	AlphaSense
Pricing model	Gated, quote-based, roughly per-user plus tier	Quote-only, priced per seat
Reported median (Vendr)	~$30K/yr total	~$17.5K/yr per contract
Reported range	~$16K-$60K/yr	~$9K-$51K/yr, seats ~$10K-$20K each
What a real team pays	Mid-five figures total	~$50K-$100K+/yr, enterprise past $1M

Source: Vendr marketplace data, Klue and AlphaSense, 2026. The point is not that one is cheap. It is that AlphaSense is priced for a research desk and Klue for a go-to-market team. When you evaluate a competitive intelligence tool, the altitude tells you which job the vendor built for.

What do customers say about Klue and AlphaSense?

Both are well reviewed, and the praise splits along the line this comparison draws. Klue sits around 4.8 on G2 from roughly 535 reviews; AlphaSense is reported near 4.6 on G2 from about 308 reviews and around 4.5 on Gartner Peer Insights, though those come from secondary citation and are worth confirming against the live listings. Klue reviewers praise the rep experience, clean battle cards, and native win-loss; AlphaSense reviewers praise the research search, content breadth, and support.

The complaints are the predictable shadow of each design. Klue draws the note that the rep experience is easy while the curator side takes work, the trade-off you would expect from a rep-delivery-first tool. AlphaSense's number-one complaint by a wide margin is cost and price opacity, trailed by a steep learning curve and the recurring verdict that it is the wrong tool for sales enablement. That last one is not a defect. It is the market telling AlphaSense buyers what AlphaSense is for.

What do Klue and AlphaSense both leave undone?

Neither one turns its output into positioning, personas, and persona-tested messaging. That is the shared gap, a scope point rather than a knock on either tool doing its job well. Klue stops at the battle card, AlphaSense one step upstream at the research answer. A competitor cuts prices, and both tools deliver the input; then a person decides what the cut means for the positioning, rewrites the value props it weakened, checks whether it shifts a buyer persona, and refreshes the messaging sales repeats in every call. The assets that go to market still get assembled by hand after both tools have done their part.

The research got automated. The part that decides what a signal changes in your positioning, and rewrites the messaging sales repeats in every call, did not. That's the expensive half, and every tool in this category still hands it back to a person and a doc.

— David Kolínek, Co-founder at Calven

For a PMM, AlphaSense is over- and under-scoped at once: analyst-grade research depth priced for a finance desk, more than a PMM competitive intelligence job needs, and no battle cards, win-loss, or messaging output at the other end. Klue covers the enablement half it was built for and stops at the card. Competitive intelligence is one slice of what product marketing owns, and a signal that should reach the positioning, value props, personas, and battle card instead reaches a feed or a research answer, and someone copies the rest into a doc that is stale by the next release.

Where does Calven fit?

Calven is an AI product marketing platform that runs the whole product marketing surface, not a CI tool or a research library bolted onto the rest of the stack. It overlaps with both of these. Its Competitive Intelligence Agent tracks competitor sites, docs, and pricing and keeps battle cards current, and its Win/Loss Agent runs win-loss as a discipline. That is Klue's territory, and it reads market and analyst signals too, with less depth than AlphaSense brings to the same question.

What Calven does that neither does is carry the signal forward. Take the same price cut. Instead of stopping at a card or a citation, it flows into the assets a PMM owns: the Positioning Agent and Messaging Agent revisit the value props the cut weakened, pressure-tested against the buyer personas it would affect. Whether that update lands as a draft for your review or propagates straight to live assets is a setting you control. The practical difference is timing: the messaging sales repeats is current for the next call, not three weeks and one stale doc later.

AlphaSense wins research breadth, Klue wins sales enablement, and each is the right buy when that is the job you are funding. Calven's claim is narrower and, for a PMM, more useful: it finishes the last mile both leave undone. So the real question is not only which of these two to buy. It is whether the budget is going to a research desk, a sales-enablement tool, or the product marketing work that turns either signal into messaging that is live in the deal. For the build-it-yourself version of that last mile, ChatGPT and Claude for competitive intelligence is the next read.

Klue vs Crayon (2026): Where Each Wins and What Both Miss

David Kolinek — Mon, 29 Jun 2026 00:00:00 GMT

Klue and Crayon are the two names that land on almost every competitive intelligence shortlist, and the honest starting point is that both do the core job well. Which one wins comes down less to features than to where competitive intelligence sits in your organization, because on price, on time to value, and on the fact that each needs a dedicated owner, they are effectively tied. The tables below run the head-to-head across product, services, cost, and what customers say, then the piece turns to the question the shortlist skips: whether a competitive intelligence tool is the right thing to be buying at all.

What's the actual difference between Klue and Crayon?

They are different in shape, not in quality. Klue is built around sales enablement and rep delivery, with native win-loss as its signature. Crayon is built around the widest automated monitoring surface it can offer, with market intelligence as its pitch. Both are dedicated, enterprise-priced platforms, so the real question is which job you are hiring the tool to do.

Dimension	Klue	Crayon
Category framing	Competitive enablement	Market and competitive intelligence
Primary audience	Sales enablement and reps in the deal	CI and PMM analysts, cross-functional
Core strength	Rep delivery and native win-loss	Breadth of automated monitoring
Analyst recognition (2026)	Gartner Leader, #1 Revenue Enablement	Gartner Leader
Best fit	Sales-led orgs anchored on Salesforce	Cross-functional CI with a curator

That difference resolves into a fairly clean call on who each one is for.

If this is you	Lean
CI owned by sales enablement, Salesforce-anchored, win-loss is central	Klue
Broad market and competitor monitoring, and you have someone to curate it	Crayon

On the core discipline itself, running a structured competitor analysis and keeping it current, neither is a weak choice.

How do Klue and Crayon compare on product and features?

Both cover the CI essentials, monitoring, battle cards, and integrations, and diverge on where they go deep. Klue goes deep on getting intel to reps and on win/loss analysis. Crayon goes deep on the breadth and automation of what it captures.

Feature	Klue	Crayon
Competitor monitoring	Curated, solid coverage	Broadest reach, more than 100 signal types (Crayon's claim), catches subtle page changes
Battle cards	Native, auto-updated, delivered to reps	Native, auto-updated
Rep delivery surfaces	Slack, Salesforce, HubSpot, email, call tools, "Ask Klue," browser extension	Newsfeed, Salesforce, integrations
Win-loss	Native discipline, built in	Field-intel capture and a Salesforce-linked win/loss module
AI (2026)	Compete Agent, deal-level, reads CRM and call data	Sparks and Answers, signal summaries and an in-tool assistant
Analytics	Battle-card usage tied to win rates	Real-time alerts and reporting across the feed

How do Klue and Crayon compare on services and support?

Beyond the software, the two invest in different things. Klue runs a managed win-loss research service; Crayon publishes the category's longest-running industry benchmark. Support is well reviewed for both, but the onboarding burden lands at different times.

Dimension	Klue	Crayon
Support reputation	Well reviewed, CSMs curate content for clients	Well reviewed, praised for smooth rollout
Time to value	About 2 to 3 months, admin setup is the harder part	Weeks to implement, full value needs ongoing tuning
Learning curve	Low for reps, steeper for curators and admins	Low to use, ongoing effort to manage signal volume
Where the effort lands	Front-loaded, in setup	Ongoing, in tuning to fight alert noise

Offering	Klue	Crayon
Managed win-loss	DoubleCheck, human-led win-loss interviews and reports	Not a managed service
Signature research	Not a focus	State of Competitive Intelligence, the category's longest-running report
Add-on AI	Compete Agent	Sparks, Answers

How much do Klue and Crayon cost?

They cost about the same, and the sticker is the smaller part of the true cost. Both are gated, so neither publishes a list price. Vendr's marketplace data puts each at a roughly $30K/yr median.

Dimension	Klue	Crayon
Median (Vendr)	~$30K/yr	~$30K/yr
Typical range	~$16K-$60K	~$12.7K-$46K
Entry point	Gated	Essentials ~$15K-$25K
Top of range	~$60K	Enterprise can run past $100K
Avg negotiated savings	~18%	~20%
The owner it assumes	An enablement or PMM curator	An analyst to curate the volume

Source: Vendr marketplace data, February 2026. The line neither vendor prints on the quote is the human. When you evaluate a competitive intelligence tool, the fully loaded cost is the tool plus the owner it assumes, and that owner is where the two prices converge.

What do customers say about Klue and Crayon?

Both are highly rated, and the gap is a few tenths of a star, not a verdict. On G2, Klue sits around 4.8 and Crayon around 4.6. Reviewers describe the same split the rest of this comparison does: each tool is praised for what it is built around, and each draws a consistent complaint that follows from its design.

Dimension	Klue	Crayon
G2 rating	~4.8 / 5	~4.6 / 5
Most praised	Ease of use for reps, unified win-loss, usage analytics	Breadth of automated capture, customer success, real-time alerts
Most common complaint	Curator and admin interface can be clunky, and it needs a dedicated owner	Alert noise, heavy tuning to separate signal from volume
The theme underneath	Easy for reps, more work for the curator	Captures everything, so a person has to filter it

These patterns show up across G2 and Gartner Peer Insights, and neither is disqualifying. They are the predictable trade-offs of a rep-delivery tool versus a wide-net monitoring tool.

What do Klue and Crayon both leave undone?

Both stop at the feed and the battle card, and for a product marketing team that is the first step, not the deliverable. Competitive intelligence is one responsibility among many. Roughly two-thirds of PMMs own competitive intelligence, versus 91% who own positioning and messaging and about 79% who own sales enablement, per the Product Marketing Alliance 2025 State of Product Marketing. CI is a slice of the job. The assets it should change are the ones most PMMs actually own.

Watch what happens to a single signal. A CI tool flags that a competitor cut prices. The alert fires, the battle card gets a note, and then a person has to decide what the move means for the positioning, rewrite the affected value props, check whether it shifts a buyer persona, and refresh the messaging that sales repeats in every call. The tool delivered the alert. The translation lives outside it.

Competitive intelligence is one input, not the deliverable. The moment an alert lands, someone still has to decide what it means for the positioning and rewrite the messaging. That translation usually happens in a doc that's already out of date by the next release.

— David Kolínek, Co-founder at Calven

That gap is measured, not theoretical. 58% of CI pros say keeping battle cards and content updated is a struggle, per Crayon's 2024 State of Competitive Intelligence, and 57.5% of teams report more competitive deals than a year ago, per Crayon's 2026 State of Competitive Intelligence. More deals, harder to keep current, and the downstream work still done by hand. A signal that should reach the positioning, the value props, the personas, and the messaging instead reaches a feed, and someone copies the rest into a Google Doc that is stale by the next release. This is a scope point, not a knock on either tool doing CI well. Competitive intelligence is one slice of what product marketing owns, and a CI-first platform does not own the rest, so its insight stops one step short of the assets that go to market. It is also why the battle card goes stale between the alert and the update.

Where does Calven fit?

Calven is an AI product marketing platform: one system that runs the whole product marketing surface, competitive intelligence included, rather than a dedicated CI tool bolted onto the rest of the stack. It overlaps with Klue and Crayon on the core ground, and it is worth saying that plainly. Its Competitive Intelligence Agent tracks competitor sites, docs, pricing, and announcements, writes dossiers, and keeps battle cards current, and its Win/Loss Agent runs win-loss as a discipline. That is the same job Klue and Crayon do.

The difference is what happens to the signal after it lands. Take the same competitor price cut. Instead of stopping at a card, it flows into the assets a PMM owns: the Positioning Agent and Messaging Agent revisit the value props the cut undercuts, drawing on the personas and ICP work, and the Voice of Customer Agent checks the new line against the language buyers actually use. The move reaches the positioning and the persona-tested messaging, not just a newsfeed.

Because it spans the whole surface, Calven is a different kind of bet from a point tool. It is also a young product, and that cuts both ways. Being new and moving fast means it is dynamic and gains ground quickly. It also means it changes more than a tool that has held the same shape for a decade. For a team that wants one platform to run the whole surface and values that pace, the trade is worth making.

Klue vs Kompyte (2026): Where Each Wins and What Both Miss

David Kolinek — Mon, 29 Jun 2026 00:00:00 GMT

Klue and Kompyte both keep your battle cards current, so the shortlist question is rarely whether the monitoring works. It is how much platform you are buying and who owns the company selling it to you. Klue is the enterprise, sales-enablement-first option with analyst pedigree and native win-loss. Kompyte is the lighter, automation-first, generally cheaper option, and since April 2026 it sits inside Adobe by way of Semrush. The tables below run the head-to-head across product, services, cost, and what customers say, then the piece turns to the question the shortlist skips: whether a competitive intelligence tool is the right thing to be buying at all.

What's the actual difference between Klue and Kompyte?

They are not the same class of tool. Klue is a dedicated, enterprise-priced competitive enablement platform built around rep delivery and native win-loss. Kompyte is a lighter, automation-first competitive intelligence and battle-card tool, cheaper and narrower, and now a component inside a much larger marketing suite. One is a system a CI team runs. The other is automated monitoring a marketing ops person keeps an eye on.

Dimension	Klue	Kompyte
Category framing	Competitive enablement	CI and sales battle cards, automation-first
Primary audience	Sales enablement and reps in the deal	Marketing ops, PMM, and growth in SMB and mid-market
Core strength	Rep delivery and native win-loss	Automated monitoring and auto-updating battle cards
Analyst recognition (2026)	Gartner Leader, #1 Revenue Enablement	No publicized MQ placement
Ownership	Independent	Adobe, via Semrush
Best fit	Sales-led orgs anchored on Salesforce	Lighter CI without a dedicated analyst

Ownership is doing real work in that table. Semrush acquired Kompyte in 2022, and Adobe completed its roughly $1.9B acquisition of Semrush on April 28, 2026, per Adobe's own announcement. So Kompyte now runs through Semrush inside Adobe, is rarely sold as a standalone CI tool, and has no publicly stated roadmap for where it lands in the Adobe portfolio. That is a continuity question a buyer should price in, not a reason to rule it out. The difference still resolves into a fairly clean call.

If this is you	Lean
CI owned by sales enablement, Salesforce-anchored, win-loss is central	Klue
You want automated battle cards on a lighter budget and have no dedicated CI analyst	Kompyte

On the core discipline itself, running a structured competitor analysis and keeping it current, both do the job.

How do Klue and Kompyte compare on product and features?

Both cover the CI essentials, monitoring and battle cards and CRM delivery, and diverge on depth. Klue goes deep on getting curated intel to reps and on win/loss analysis. Kompyte goes wide on automated capture and leans on Semrush's data infrastructure to do it, then narrows to the battle card.

Feature	Klue	Kompyte
Competitor monitoring	Curated, solid coverage	Automated across sites, pricing, ads, reviews, social, and job posts; leans on Semrush data
Battle cards	Native, auto-updated, delivered to reps	AI-generated and auto-updated (Kompyte's claim); reviewers say cards still need manual work, and note no PowerPoint export
Rep delivery surfaces	Slack, Salesforce, HubSpot, Highspot, Gong, "Ask Klue," browser extension	Slack, Microsoft Teams, Salesforce, HubSpot; alert-to-feed
Win-loss	Native discipline, built in (DoubleCheck)	Thin, routed to a third-party partner
AI (2026)	Compete Agent, deal-level, reads CRM and call data	AI noise-filtering, battle-card updater, summaries; no conversational AI interface
Analytics	Battle-card usage tied to win rates	Feed alerting and battle-card engagement tracking

Two gaps are worth naming. Kompyte's auto-updating battle card is its headline pitch, but independent reviewers report the cards still need manual effort and lack Klue's editorial depth, so treat "no manual work" as marketing, not a spec. And Kompyte's win-loss is not a native discipline: it runs through a partner, while Klue's is productized in the platform. If win-loss is central to why you are buying, that is the whole ballgame.

How do Klue and Kompyte compare on services and support?

Both are well supported, and both assign a person to get you live: Klue's customer success managers actively curate content for clients, and Kompyte is sold by demo and run with a dedicated CSM whose onboarding reviewers praise. The difference is what the vendor is, not just how it answers a ticket.

Dimension	Klue	Kompyte
Support reputation	Well reviewed, CSMs curate content	Well reviewed, strong onboarding with a dedicated CSM
Time to value	About 2 to 3 months, admin setup is the harder part	Automation-first setup, but battle cards need manual effort to be useful
Learning curve	Low for reps, steeper for curators and admins	Lighter, though reviewers say steeper than the marketing implies
Where the effort lands	Front-loaded, in setup	Ongoing, absorbed into marketing ops rather than a dedicated analyst
Managed win-loss	DoubleCheck, human-led win-loss interviews and reports	No in-house service; win-loss routed to a third-party partner

How much do Klue and Kompyte cost?

Kompyte is the cheaper of the two, but neither prints a price and one barely leaves a public trail. Vendr's marketplace data puts Klue at a roughly $30K/yr median, per Vendr. No comparable Vendr median surfaced for Kompyte, and third-party estimates conflict enough that quoting a figure would be guessing. What is consistent is direction: Kompyte prices below Klue at similar deployment sizes.

Dimension	Klue	Kompyte
Median (Vendr)	~$30K/yr	No published Vendr median
Typical range	~$16K-$60K	Gated; positioned below Klue
Pricing model	Per-user, plus tier and integration level	Per-competitor and per-user, which rises as coverage expands
Entry point	Gated	Gated, quote-only
The owner it assumes	An enablement or PMM curator	Absorbed into existing marketing ops

Source: Vendr marketplace data, February 2026. Two things the sticker hides. Kompyte's per-competitor model means the price climbs as you widen coverage, so the cheap tool gets less cheap the more competitors you track. And on both platforms the real line item is the human. When you evaluate a competitive intelligence tool, the fully loaded cost is the tool plus the person it assumes, and that person does not appear on either quote.

What do customers say about Klue and Kompyte?

Both are rated well, on very different review volumes. Klue sits around 4.8 on G2 from roughly 535 reviews, the highest in the category; Kompyte reviews well too, a few tenths below, on a much smaller base of about 100 reviews. Each tool draws the complaint its design predicts.

Dimension	Klue	Kompyte
G2 rating	~4.8 / 5 from ~535 reviews	Rated well, a few tenths below Klue, on ~100 reviews
Most praised	Ease of use for reps, unified win-loss, usage analytics	Intuitive interface, fast competitor data, automation cuts research time
Most common complaint	Curator and admin interface can be clunky, needs a dedicated owner	Pulls parent-company data, misses smaller rivals, cards still need manual work
The theme underneath	Easy for reps, more work for the curator	Fast and automated, but narrower and still hands-on

One Kompyte-specific pattern is worth flagging for a PMM. Reviewers note it can surface parent-company data instead of the direct competitor, and that it misses smaller and niche rivals while covering the big names well. For a team whose hardest battles are against the scrappy startup nobody has heard of yet, that is exactly the coverage gap to weigh.

What do Klue and Kompyte both leave undone?

Both stop at the feed and the battle card, and for a product marketing team that is the first step, not the deliverable. Roughly two-thirds of PMMs own competitive intelligence, versus 91% who own positioning and messaging and about 79% who own sales enablement, per the Product Marketing Alliance 2025 State of Product Marketing. CI is a slice of the job, and the assets it should change are the ones most PMMs actually own.

Watch a single signal. A CI tool flags that a competitor cut prices; the alert fires and the battle card gets a note. Then a person decides what the move means for the positioning, rewrites the affected value props, checks whether it shifts a buyer persona, and refreshes the messaging sales repeats in every call. The tool delivered the alert; the translation lives outside it. With Kompyte that gap is wider by design: reviewers flag that it lacks market intelligence, and it is now one component inside a marketing suite rather than a PMM system of record.

The tool that tells you a competitor moved is doing the easy part. The hard part is deciding what that move does to your positioning and rewriting the messaging before the next call. That work belongs to a PMM, and it almost always happens in a doc that's already out of date.

— David Kolínek, Co-founder at Calven

That gap is measured, not theoretical. 58% of CI pros say keeping battle cards and content updated is a struggle, per Crayon's 2024 State of Competitive Intelligence, so a signal that should reach the positioning, value props, personas, and messaging instead reaches a feed, and someone copies the rest into a Google Doc that is stale by the next release. This is a scope point, not a knock on either tool doing CI well: competitive intelligence is one slice of what product marketing owns, and a CI-first tool does not own the rest, which is also why the battle card goes stale between the alert and the update.

Where does Calven fit?

Calven is an AI product marketing platform: one system that runs the whole product marketing surface, competitive intelligence included, rather than a dedicated CI tool bolted onto the rest of the stack. It overlaps with Klue and Kompyte on the core ground. Its Competitive Intelligence Agent tracks competitor sites, docs, pricing, and announcements, writes dossiers, and keeps battle cards current, and its Win/Loss Agent runs win-loss as a discipline. That is the same job Klue does well and Kompyte does more lightly.

Because it runs positioning, messaging, and audience work alongside CI, Calven is a different kind of bet from a point tool. It is also a young product, and that cuts both ways: moving fast means it gains ground quickly, but it also changes more than a tool that has held the same shape for years. For a team that wants one platform to run the whole surface and values that pace, the trade is worth making.

So the real choice is not only which competitive intelligence tool to buy, or whose portfolio you want to be a line item in. It is whether competitive intelligence is the whole job you are buying for. For a team weighing whether to build that last mile out of general tools instead, ChatGPT and Claude for competitive intelligence is the adjacent decision.