All articles

Customer success

Health scores that predict churn, not noise

A hand holding a phone

A health score is only useful if a red account is more likely to churn than a green one. Surprisingly often, that is not the case.

Fewer signals, better weighted

Teams add signals because they are available, not because they predict anything. Start with usage trend, support severity and payment status, and add others only if they improve the back-test.

Back-test before you trust it

Score last year’s accounts with the model and check how many churned accounts it flagged ninety days before renewal. Anything under 60% needs work.

The best health score we have seen used just four signals and caught 78% of churn three months out.

Give every score a reason

A score without an explanation gets ignored. Show the two or three signals that moved it, so the account manager knows what to do next.

Keep reading

Free guide

Get the Forecast Playbook

Twelve practices from 600 revenue teams, in a 40-page guide.

The Forecast Playbook on a desk

Run your revenue on Keel