A Claude Skill that scans every CSM call and usage spike across your customer base and surfaces concrete expansion signals: new use cases mentioned, additional teams onboarding, integration questions about premium-tier features, and asks for capabilities that map to your upsell SKUs. Drops a weekly digest with a ranked list per CSM.

What you’ll need

Gong access to CSM calls (separate from sales calls)
Gainsight for usage anomaly detection
Claude Code or Claude.ai with the Skill installed
A signal taxonomy mapping mentions to specific upsell SKUs
A Slack or email destination for the weekly digest

Setup

Build the signal taxonomy. Edit references/expansion-signals.md with each upsell SKU and the trigger phrases or usage patterns that indicate intent. Be specific: not “more users,” but “asked about pricing for 50+ seats” or “mentioned compliance requirements.”
Install the Skill. Drop expansion-signals.skill into ~/.claude/skills/. Set GONG_API_KEY and GAINSIGHT_TOKEN.
Run weekly. detect_expansion(window_days=7). The Skill scans every CSM call from the window plus every usage spike of more than 30% week-over-week, then maps signals to SKUs and ranks by deal-size potential.
Route the digest. Output goes to a #csm-expansion Slack channel and an email to each CSM with their personalized list. Top 3 per CSM, not 30.

How it works

For each CSM call, the Skill runs an extraction prompt: “Find any mention of additional teams, use cases beyond the current scope, premium feature interest, or stated business goals that imply expansion.” Each match is logged with a quote and a confidence score.

For usage, it pulls Gainsight usage events and finds accounts with anomalous spikes — particularly in features that are gated by tier. A free-tier account hitting an enterprise-only feature limit is a signal. A team that just doubled active users is a signal.

Claude then synthesizes per-account: combines call signals with usage signals, scores expansion likelihood, names the most likely SKU, and suggests an opening line for the CSM to use in the next conversation.

Watch-outs

Signal fatigue. Don’t surface 30 signals per CSM. Rank to 3. Anything beyond is ignored.
CSM call coverage. If CSMs aren’t logging calls, the conversation half of the signal set is empty. Audit coverage before relying on this.
Misclassified signals. “We’re thinking about leaving” can sound like “we’re thinking about expanding” to a careless prompt. Build negative examples into the Skill prompt.

Stack

Gong — CSM call corpus
Gainsight — usage anomaly source
Claude — signal extraction, mapping, ranking