Resources · Health Scores
Building a health score your CSMs trust
Ask a room of customer success managers whether they trust their health score and you'll get polite silence. Ask them how they actually decide which accounts to worry about, and they'll tell you: memory, gut, the last angry email. The score sits in the CRM, slightly red, and nobody looks at it.
That isn't the CSMs' fault. Most health scores deserve the silence. They were written once, by someone who has since left, from a list of signals that sounded sensible. They can't explain themselves, they're wrong often enough to be ignored, and nobody ever checked them against a single real renewal.
We've helped more than 300 customer success teams rebuild their scores. The ones that work follow five rules.
1. Start from your renewals, not from a whiteboard
The most common mistake is brainstorming signals. Someone suggests NPS, someone suggests logins, someone suggests "engagement", and twenty minutes later you have eleven signals and no evidence for any of them.
Start the other way round. Pull your last eight quarters of renewals and churns. For each churned account, look at what was different 90 to 180 days before the renewal date. For each renewal, same question. You are looking for signals that separate the two groups, in your data, for your product.
When Tidewater Freight did this, they found that NPS barely mattered. Weekly active users against seats sold mattered a lot, and so did one signal nobody had suggested: errors on their tracking API. Customers whose integrations were failing churned at three times the rate of everyone else. That went into the score. NPS came out.
2. Use fewer signals than you want to
Every signal you add makes the score harder to explain and easier to argue with. Four to six signals is plenty for most teams. If a signal doesn't move the score's accuracy when you back-test it, take it out, even if it feels important.
This is the hardest rule to follow, because someone always has a favourite signal. Let the back-test settle the argument. That is what it's for.
3. Make every score show its working
A CSM should be able to open an account and say, in one sentence, why it's amber. "Usage dropped from 610 to 430 weekly users after their reorganization, and there are two escalated tickets on reporting." If they can't, they won't act on the score, and they'll be right not to.
That means showing which signals moved the score, by how much and since when, on the account page itself. Not in a separate report. Not in a tooltip nobody hovers over. Right next to the number.
A score that can't explain itself doesn't get used. In our experience, this single change does more for adoption than any amount of training.
4. Separate your segments
A 40-seat customer and a 4,000-seat customer don't churn for the same reasons. Small accounts tend to leave because they never adopted the product. Large accounts leave because a sponsor changed, a competitor got in at the executive level, or a procurement review asked questions nobody prepared for.
One score for both will be mediocre at each. Northwind Analytics runs two models, one for SMB and one for enterprise, and the enterprise model weights sponsor changes about four times as heavily as the SMB one. Two simple models beat one complicated one.
5. Check it every quarter, in public
A health score is a prediction. Predictions should be checked. Every quarter, compare what the score said 90 days before each renewal with what actually happened. How many churned accounts were green? How many red accounts renewed happily?
Share the answer with the CS team, including the misses. This is counterintuitive, because it seems like admitting the score is wrong would make people trust it less. The opposite happens. CSMs trust a score that is honest about its error rate far more than one that claims to be right.
Then adjust. Products change, customers change, and a score that was accurate last year can drift quietly. A quarterly check catches it before your CSMs do.
What good looks like
You'll know the score is working when the conversation in your weekly risk review changes. Instead of "is this account really at risk?", people start asking "what are we doing about it?". That's the moment the score stops being a CRM field and starts being a tool.
At Tidewater Freight, that happened in about two weeks. Their at-risk accounts now turn amber roughly 70 days earlier than before, and gross churn fell 18% in the first year. The score didn't save those accounts on its own. It gave the team enough time to.
If you'd like to see what your own renewal history says about your signals, we're happy to look at it with you. Book a demo and bring a CSM.