Customer Health Score Rubric Designer

---
name: health-score-rubric-designer
description: Use when designing a customer health scoring rubric or fixing one that fails to predict churn. Triggers on "health score", "customer health", "churn prediction", "health rubric".
---

# Customer Health Score Rubric Designer

A health score is only useful if it forces real decisions. A score where 90% of accounts are green and yet 8% churn isn't a health score — it's a dashboard.

## Required inputs

1. **Signals you have access to** — product usage, NPS / CSAT, support volume, contract data, exec engagement, payment behavior
2. **Past churn data** — accounts that left, with what signals preceded
3. **Customer segments** — different segments may need different rubrics
4. **Use case** — health score driving CSM action, executive reporting, or both

## Output

### Signal weighting
Typical weighted rubric:
- **Product adoption (30%)** — usage breadth + frequency
- **Engagement (20%)** — active users / total seats
- **Support sentiment (15%)** — ticket volume + severity + CSAT
- **Executive engagement (15%)** — QBR attendance, exec NPS, calls accepted
- **Renewal proximity (10%)** — time to renewal × current relationship strength
- **Payment behavior (10%)** — invoice timeliness, finance escalations

Adjust per segment. Justify each weight with the historical churn signal it captures.

### Per-signal thresholds
Concrete numbers, not vibes:
- "Active users below 30% of seats logged in last 30 days → 0/30 points"
- "NPS < 6 from primary contact → 5/20 points"

### Scoring bands
- **Green (80-100)** — healthy, expansion candidate
- **Yellow (60-79)** — watch list, CSM should act this month
- **Orange (40-59)** — at-risk, exec intervention warranted
- **Red (< 40)** — churn likely, special handling

### Calibration
Take last 12 months of accounts that churned. Score them with this rubric using their data from 6 months before churn. The rubric should have flagged most of them. If it flagged < 60%, the weights are wrong.

### What the rubric will get wrong
Honest. Accounts that score green but will churn (likely cause: champion leaves, M&A, off-product signal). Accounts that score red but renew (likely cause: legacy account with deep contract lock-in).

### Per-band CSM action
- Green: nothing routine; opportunistic expansion conversation
- Yellow: monthly health touch, specific issue to address
- Orange: weekly check-ins, escalation to senior CSM, exec sync
- Red: war-room mode, save plan, exec sponsor

### Review cadence
- Weekly: scores recalculated, deltas reviewed
- Monthly: accounts that changed bands → why
- Quarterly: recalibrate weights against actual churn outcomes

## Anti-patterns
- 90% green band (the rubric doesn't discriminate)
- Signals that don't actually correlate with churn (added because they were easy to measure)
- One rubric across very different segments
- Manual overrides that defeat the rubric

## When a score won't help
- Highly relationship-driven enterprise accounts where one CSM knows everything
- Pre-PMF customers where churn is product-driven, not relationship-driven
- Single-decision-maker buying motion where the signal is binary