Find what your wins, losses, and churn share.
Reads closed-won, closed-lost, and churned accounts, then hands back a ranked list of lookalike accounts to work.
Inside: The won / lost / churn pattern logic and the lookalike ranking.
Install it in one line, or paste it in.
~/.claude/skills/ and runs automatically when it is relevant.Connect your context. Set it to your motion.
the three cohorts and their fields, live.
engagement decay, which powers the predictive-churn scan.
Map each row to the actual field name in your system. If a pattern depends on a field you do not have, the skill runs the modes it can and tells you which one it skipped and why.
| Set this | What it is | Default / Example |
|---|---|---|
| CRM connector | Where your won/lost/churn cohorts live | Any CRM |
| Won/lost stage field | The field that marks a deal won or lost | Stage = Closed Won / Closed Lost |
| Loss reason field | Reason a deal was lost | A loss-reason field |
| Competitor field | Named competitor a deal was lost to | A competitor field |
| Churn signal | How you mark a lost customer vs a lost new deal | Type = Renewal AND Stage = Closed Lost |
| Segment fields | Account fields used for lookalike matching | Employee countindustrytech stackchannelprimary signal |
| Activity signal | The product-analytics metric for engagement decay | 30-day active-user trend |
Everything the skill does, in full.
This skill reads your closed-won, closed-lost, and churned accounts and finds the repeating patterns inside them. It runs in three modes. WON tells you what your winning deals have in common and generates a ranked list of active accounts that look like them. LOST tells you where, how, and to whom you lose, with the objections rolled up. CHURN groups your lost customers into named themes and then scans your active book for accounts that match those themes before they leave.
- 1WON mode
Cohort: trailing 6 months of closed-won. Extract buying-committee shape, deal velocity, primary entry point, dominant signals, tech-stack overlap. Score active accounts against the winning pattern and rank the closest matches.
- 2LOST mode
Cohort: trailing 6 months of closed-lost. Competitive head-to-head by stage and deal size. Objection theme rollup from transcripts. Stage-of-loss distribution.
- 3CHURN mode
Cohort: trailing 12 months of churned customers. Group churns into named causes (consolidation, layoff, vendor switch, product gap, champion left, acquisition). Scan active accounts for matches to each theme and surface them with a confidence tag, ranked by risk.
- Pattern extraction is statistically meaningful. The skill requires at least 5 occurrences before it calls something a pattern. Below that it returns "insufficient data, expand the window."
- Lookalike candidates are similarity-scored, with the named dimensions behind the score.
- Predictive churn calls are confidence-tagged with the evidence behind them. A low-confidence match is labeled low, not dropped.
PATTERN ANALYSIS · Q2 Lookback
WON (n=14):
Channel: Product 57% / Inbound 29% / Outbound 14%
Dominant signal: sales-team hiring (78% of wins)
Top lookalikes: Account A 91%, Account B 88%, Account C 87%
LOST (n=21):
Stage-of-loss: Solution Validation 52%
Top reason: "no compelling differentiation vs incumbent" (n=9)
Lost to Competitor 1: 5 (at Proposal, "team already trained")
CHURN (n=8, 12 months):
Consolidation / layoff (n=3), acquisition (n=2), feature gap (n=2)
PREDICTIVE WATCH:
Account F, consolidation pattern (73%): team 22 to 14, activity -47%
Account G, champion-left pattern (68%): champion changed jobs 21d agoDefaults to re-tune: win/loss windows 6 months trailing, churn window 12 months, pattern floor of 5 occurrences before something is named a pattern. Lookalike and churn-match scores are built from your segment fields; add or drop dimensions to fit what predicts outcomes in your data. The activity-decay threshold flags an at-risk account; tune it to your product's normal usage rhythm. Re-tune the windows and the pattern floor first.
Where an operator takes this next.
Running this quarterly on an export proves the patterns hold. Here's the version that watches the active book continuously.
The patterns are only useful the week you'd actually act on them.
A scheduled Claude task reruns the predictive-churn match against Salesforce weekly and only surfaces accounts whose risk score moved.
Connect Gong or a similar conversation-intelligence tool so a champion mentioning a new role gets flagged automatically instead of waiting for the account owner to notice.
Push the lookalike list into Clay or your outbound tool so the accounts that score closest to your winning pattern become a working list, not a slide.
One skill is the on-ramp.
A single skill does one job. Chained into a playbook, or run as a full build, it becomes a system. Here is where this one plugs in.