A Claude Skill that mines win/loss conversations in Gong and produces living competitive battlecards: where you win, where you lose, the most-cited objection patterns, the pricing gap quotes, and the migration concerns. Replaces battlecards that go stale the day they ship.

What you’ll need

Gong API key with deal and call read access
Salesforce with a competitor field on Opportunity (or a custom multi-pick)
Claude Code or Claude.ai with the Skill loaded
A list of named competitors to track (start with 3-5)
Optional: a canonical battlecard template in Notion or Confluence

Setup

Tag deals. Make sure every closed-won and closed-lost deal in the last 12 months has at least one competitor populated. If your data is sparse, run a one-time backfill — Claude can help by reading deal notes.
Install the Skill. competitive-battlecards.skill into ~/.claude/skills/. Set GONG_API_KEY and SFDC_TOKEN.
Configure scope. Edit references/competitors.md: list each competitor, common aliases, and any product modules to differentiate (e.g., Competitor X — Core versus Competitor X — Enterprise).
Generate. build_battlecard(competitor="competitor-x", lookback_days=180). The Skill produces a Markdown battlecard with quotes, win/loss rate, average deal size delta, and a recommended objection-handler script.
Refresh monthly. Schedule the run; old battlecards rot fast.

How it works

The Skill queries Gong for every call where the named competitor was mentioned, scoped to the lookback window. It separates won deals from lost deals using Salesforce as truth, then runs a structured extraction pass: pricing mentions, feature gaps named, integration questions, support concerns, and migration objections.

Claude then writes the battlecard in two voices. The first is data-led — “in 47 of 89 won deals against Competitor X, the deciding factor was X.” The second is conversational — short objection handlers reps can actually deliver in a call. Both voices reference real customer quotes, with the call ID linked back to Gong for verification.

The killer feature: the Skill also surfaces what changed since the last battlecard. New objection patterns get flagged as “emerging.”

Watch-outs

Selection bias. Reps don’t always log the competitor they actually lost to. Pair the Skill output with a quarterly rep gut-check.
Quote accuracy. Gong transcription is imperfect, especially with technical terms. Claude is instructed to mark low-confidence quotes; honor those flags.
Battlecard sprawl. Three to five live competitors is plenty. Beyond that, you’re producing artifacts no rep reads.

Stack

Gong — call corpus and customer voice
Salesforce — deal outcome ground truth
Claude — extraction, classification, battlecard authoring