Skip to main content

​
Competitive Intelligence That Reps Actually Use

What this solves: Sales reps get competitive cards built from real evidence, with honest assessments of where competitors actually win — the kind reps trust enough to use mid-call.

​
The Problem

Competitive battlecards rot. A PMM builds them once, they sit in a wiki, and within three months the competitor has shipped new features, changed pricing, and shifted positioning. The card is now actively misleading. Meanwhile, the best competitive intel lives in scattered Slack messages from reps who just lost a deal — never making it back into the card. There’s a subtler failure underneath that. Ask an AI to synthesize intelligence from several sources about a competitor and it manufactures consensus. G2 reviews, sales call transcripts, and competitor marketing all get smoothed into one coherent narrative — and the contradictions, which are usually the most useful signal, quietly disappear.

​
How It Works

The system runs five phases with an integrity protocol that forces contradictions to surface rather than resolve. The integrity protocol is the key architectural decision. Phase 2 extracts from each source individually — what does this G2 review actually say, what bias does it carry, how recent is it? Phase 3 cross-references those extractions looking specifically for disagreements. “Sales says we win on ease of use, but call transcripts show deals closed on a different feature entirely” — that contradiction becomes the most valuable section of the card. Every build produces two artifacts: a rep-facing card (scannable under cognitive load, ~650 words) and a PMM-internal card (fully annotated with source attribution, confidence levels, and a changelog for automated refresh).

​
In Practice

In a single sprint, Claude built six competitor battlecards in parallel. Each card had 4-5 agents gathering intelligence simultaneously — scraping competitor sites, pulling review platform data, searching internal Slack channels, and checking existing research. The per-source extraction ran as parallel agents split by source type. The contradiction audit caught real errors. One card initially conceded recording quality to a competitor when our product had already shipped full parity. Another had three different characterizations of the same competitor’s AI capabilities across three sections — each individually plausible, collectively incoherent. A three-agent review process (simulating rep, PMM, and enablement perspectives) caught both before the cards shipped. The monthly refresh pipeline runs in about 30 minutes of PMM review time. Agents handle the gathering; the human handles judgment calls on what actually changed enough to warrant a card update.

​
What I Learned Building This

The strongest competitive framing is always category distinction, not feature comparison. “They’re a recording tool. We’re an editing platform” outlasts any specific feature claim because it describes product ancestry rather than current capabilities. Feature claims go stale in 6-12 months. Category framing is much harder to invalidate. I cut the card format down twice. The first version was too long — reps don’t read 1,700-word documents mid-call. Going to 650 words wasn’t just editing; it changed which sections existed at all. A word budget per section (50 words for quick kills, 80 for the talk track, 120 for objections) forced every line to earn its place. One honest concession is what makes the rest of the card credible. The “Where They Win” section — where you name the places a competitor genuinely beats you — built more rep trust than any winning claim, so I set a minimum of three items. If you can only find one way a competitor wins, you haven’t looked hard enough, and reps will discount the whole card as spin. The coexist column in the displacement matrix did the same thing: giving reps permission to not compete on every front made them sharper about competing where it mattered.