health-score-design
/health-score-design
Purpose
/health-score-design produces a customer health score model — the scoring system, not the scores themselves. It defines signal families, per-segment weights, a cold-start variant for <90-day customers, and the recalibration cadence that keeps the model honest over time. The output is a model document that a CS team can implement in Gainsight, a spreadsheet, or a custom pipeline.
This skill intentionally does NOT interpret scores against a real customer base. Interpretation (reading a live score, deciding whether to open a retention playbook, triaging at-risk accounts) is handled by the separate health-score-interpret skill that consumes the model this skill produces. Keeping design and interpretation in different skills prevents the common failure mode where a health score model gets "tuned" every time a surprising customer outcome shows up — which is calibration thrash, not learning.
When to Use
Invoke /health-score-design when you need to:
- Stand up a health scoring model for a new product that has no historical churn data
- Redesign health scores after a segmentation refresh (new segments need new weights)
- Run a quarterly calibration cycle against accumulated churn and retention data
- Perform a health-score audit after an unexpected churn where the score failed to flag the risk
- Introduce segment-specific health models where a single one-size-fits-all model has broken down