Health Scores Are a Vanity Metric

Health scores are junk. They lag actual churn by three to six months, they create false confidence in renewal safety, and they're a vanity metric that lets CS leadership sleep at night while renewal risk compounds in the background.

At enterprise scale, health scores are actually worse than useless. They're actively harmful because they substitute measurement theatre for real retention work.

The Lag Problem

A health score tells you about a customer's past. It aggregates backward-looking data points: support tickets, feature adoption, login frequency, NPS responses. It converts them into a number that feels predictive. It isn't. By the time a health score reflects genuine risk, the customer is already in your renewal window, and your options are constrained. At that point, a health score doesn't prevent churn; it confirms it.

The math is straightforward. Most B2B companies operate on long renewal cycles. A health score calculated early is based on past behaviour. By the time that score actually matters, when renewal conversations begin, the customer's business context has shifted. Budget decisions changed. Procurement moved teams. Use cases evolved. The customer's technical team just onboarded new stakeholders who haven't been through your training. None of that appears in a historical health score. The score tells you what was true; you need to know what's true now.

This gap isn't a measurement problem you can solve with better variables. It's a structural problem with the premise itself. You're trying to replace ongoing customer intelligence with a static model. The model will always lose that bet. And the longer the renewal cycle, the worse the lag becomes. By the time renewal risk becomes commercially relevant, the score is often based on behaviour from far too long ago. That's not prediction; that's archaeology.

False Confidence at Scale

The second failure is worse: health scores create false confidence. A green dashboard is seductive. It tells the executive team, and your own CSMs, that everything is under control. The customer has good adoption, positive sentiment, consistent engagement. Renewal is probably safe. Except the customer's CTO just announced they're migrating to a competitor. That information doesn't flow into your health score system. It flows through a Slack message to one CSM, who escalates it to their manager, who maybe gets around to telling you if the timing is right. By then, you've already spent two weeks operating under the assumption that the account was healthy.

You see this pattern constantly in enterprise software. Accounts show green health scores in the last calculated period and still churn or contract because the data is stale. The health score didn't predict churn because the data it relied on was stale. The actual churn drivers, contract renegotiation pushback, technical debt accumulating, a new buying committee that didn't trust the existing implementation, emerged after the score was calculated. The score was measuring the wrong time horizon.

More importantly: those accounts with green scores weren't deprioritized in the renewal cycle because the score looked safe. They got standard attention. They didn't get the kind of aggressive, early engagement that actually prevents churn, the kind you'd apply if you knew the customer was at risk. The green score caused active harm through false negative. You don't fight for accounts you think are locked.

The Perverse Incentive

When you grade a CSM's performance partly on health score movement, you've created an incentive to optimize for the score, not for the customer outcome. That's not because your CSMs are bad operators. It's because you've built a system where gaming the visible metric is the rational thing to do.

A customer's engineering team files six support tickets in a month because they're building on your platform and hitting edge cases. That's actually bullish: they're investing in your product. But a health score system might interpret ticket volume as a negative signal. The CSM knows this. The rational move is to discourage the customer from filing tickets and instead direct them to documentation, to avoid score drag. The customer doesn't get the support they need, and you get a health score that looks better than the reality it's obscuring.

Or consider feature adoption. A customer's team reduces their login frequency because they've built custom integrations and automated their workflows. That's a success; they're more efficient. But a health score might flag it as declining engagement. The CSM's incentive is now to push the customer to log in more, to maintain the engagement metric. You've created friction in your best customer relationships.

These aren't edge cases. This is what happens at scale when you make a single metric both a CSM performance measure and a risk indicator. The incentives collapse into optimization for the metric itself. You're not measuring customer health; you're measuring how well your CSMs can perform behaviors that move the health score.

What Health Scores Actually Measure

The deeper problem is that health scores give you the illusion of predictability in a domain that isn't very predictable. Customers are bought and sold. Teams reorganize. Technical decisions get made that make your use case obsolete. Budgets get slashed three months out. A health score can't predict any of that because those events aren't driven by adoption or engagement patterns; they're driven by external business factors that don't show up in your platform at all. You're measuring the things you can measure and calling it prediction.

This is seductive because it feels rigorous. You have a formula. You can track it over time. You can set targets. You can build a dashboard. But you've optimized for measurability, not for accuracy. The health score becomes a KPI because it's easy to track, not because it's actually predictive of renewals or expansion.

The Real Cost

The vanity isn't in the metric itself. It's in the comfort it provides. A green health score tells you that your CS operation is working, that you have visibility into your at-risk accounts, that churn is under control. It tells you this with perfect confidence and zero actual accuracy. And because you're staring at a dashboard instead of having hard conversations with customers, you don't find out how wrong the score was until the renewal is already at risk.

That's what happens when you automate away the intelligence gathering that actually prevents churn. The numbers look good. The accounts don't.