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

> Why you can't auto-generate battle cards from Gong transcripts alone, what sales calls do sharpen, and how to keep the card true where it moves fastest.

- Source: https://calven.ai/resources/ai-for-pmm/battle-cards-from-sales-calls
- Author: David Kolinek
- Published: 2026-07-16
- Tags: competitive intelligence, battle cards, voice of customer, sales enablement, AI for PMM

---

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](https://aprildunford.substack.com/p/the-no-differentiation-illusion), 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](/platform/research#voice-of-customer) does against your recordings, pulling the buyer language, objections, and pains from each transcript into your [why-won and why-lost analysis](/resources/product-marketing/win-loss-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](https://www.growthunhinged.com/p/2025-state-of-saas-pricing-changes), and [Paddle finds](https://www.paddle.com/blog/frequency-pricing-change) 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](/resources/competitive-intelligence/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](/resources/competitive-intelligence/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](/resources/comparisons/chatgpt-claude-competitive-intelligence), 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](/resources/ai-for-pmm/how-ai-is-changing-product-marketing) 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](/platform/research#competitive-intelligence) tracks each rival's site, docs, and pricing, keeps the dossier current, and generates the sourced battlecard from it. The [Product Intelligence Agent](/platform/messaging#product-intelligence) 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.