A health score that turns red the month before a customer churns is a report, not a tool. By then the decision is made. The job of a score is to flag risk while you can still change the outcome.
Building a predictive score
- Weight leading indicators, not lagging ones. Logins last month are lagging. Trend in core-action frequency, breadth of users, and time-to-value are leading.
- Track the slope, not the level. A high-usage account in steady decline is at more risk than a low-usage account that's climbing.
- Include relationship signals. Champion departure, support sentiment, and unanswered emails predict churn as well as product usage.
- Calibrate against actual churn. Backtest the score on accounts you've already lost. If it didn't flag them early, the weights are wrong.
Bottom line: score the trajectory, not the snapshot — a health score earns its keep only if it fires while you can still act.