How to Build a Customer Success Health Score That Predicts Churn
Most health scores I've seen are about as useful as a horoscope. They're vague, they react late, and by the time an account turns red the customer has already decided to leave. A score that only explains what already happened isn't helping your CSMs do anything differently.
A good health score should predict behavior, not describe it after the fact. Here's how I'd build one that gets closer to that.
Why most scores miss
The usual problems are easy to spot once you look. The score leans on vanity inputs like logins and email opens. Nobody has checked whether it lines up with actual renewals. The model was set up once and never changed as customers changed. And too much of it runs on a CSM's gut feel entered as a number.
None of those are hard to fix. They just take some discipline.
What goes into it
I'd build from five kinds of signal:
Product adoption. Are they using the features tied directly to the value they bought? Activity on its own isn't success, which is the whole argument for measuring outcomes instead of logins.
Engagement. Executive alignment, whether they show up to business reviews, how quickly they respond. No executive engagement is one of the clearest warning signs there is.
Sentiment. CSAT, NPS and qualitative feedback. Pay attention to how customers say things as well as the score they give.
Support health. Ticket volume, severity and time to resolution. Frustrated customers aren't always loud about it until they leave.
Financial behavior. On-time renewals, upsells, budget conversations. If a customer "forgot to budget" for you, take the hint.
Building it step by step
Pick a small set of inputs. Don't jam every KPI into the score. Keep the ones that actually relate to renewal or expansion. If you're not sure which those are, start from the CS metrics that actually matter and work down.
Weight them on purpose. They don't all matter equally. As an illustration only, you might decide executive engagement deserves more weight than feature adoption for your enterprise accounts. The point is to make that choice deliberately and write down why, rather than splitting everything evenly.
Set clear thresholds. A 0 to 100 scale or green, yellow and red both work. What matters is that everyone knows what "healthy" means before a customer hits panic mode.
Check it against real outcomes. This is the step most teams skip. Pull your renewed and churned accounts from the last few periods. Did the healthy ones actually renew? Did the at-risk ones actually churn? If not, the score isn't predictive yet, no matter how good the dashboard looks.
Then keep tuning it. Revisit the weights every quarter based on what actually happened. A health score is a living tool, the same way a customer journey map is, and it drifts out of date if you leave it alone.
Mistakes to watch for
Overcomplicating it is the most common one. If your CSMs can't explain why an account is yellow, they won't trust it and they won't act on it. Relying only on surveys is another: a happy NPS response is not a safe renewal. Scoring every segment the same way is a third. SMB and enterprise customers behave differently, so their scores should be built differently, ideally lining up with however you've segmented your customers. And the last one is treating the score as finished. Markets shift, products change, and the score has to shift with them.
If you have the data to go further, predictive models can take some of the guesswork out of the weighting. I cover that in how AI is changing health scoring. But a simple score you've validated against real renewals will beat a sophisticated one nobody has checked.
Building a score that predicts churn is part science and part detective work, and it's never really done. Get it close, and your team spends a lot less time being surprised at renewal.
If you're rebuilding your health score and want someone to pressure-test it, let's talk.
