Customer Health Score Formula — How to Build One That Actually Predicts Churn

A customer health score is only useful if it predicts something. Many CS teams build health scores that feel comprehensive — pulling in product usage, NPS, support tickets, CSM sentiment — but do not actually correlate with renewal or churn outcomes. The result is a number that looks meaningful on a dashboard but does not drive better decisions.

This guide covers how to build a health score formula that is predictively valid, not just visually impressive.

What Makes a Health Score Actually Good

A good health score has three properties. First, it is predictive: accounts that score red should churn at a higher rate than accounts that score green. If they do not, the weights are wrong. Second, it is actionable: each component of the score should map to a specific CSM action when it fires. A score that highlights a problem with no clear response protocol is not actionable. Third, it is timely: the score should update frequently enough that CSMs see the signal before it becomes too late to act.

Gainsight's research on health score design shows that teams using health scores with 3–5 components outperform those using 8+ components — because fewer, better-weighted signals produce cleaner predictions than many noisy ones.

Choosing the Right Components

Start with the signals from your own churn data, not industry benchmarks. Pull a list of every account that churned in the last 12 months. Look at what their metrics looked like 90 days before they cancelled. The metrics that were consistently degraded in churned accounts are your most predictive signals.

If you do not have enough historical data, use these validated components as a starting point — they hold up across a wide range of B2B SaaS products:

The Formula

Here is a specific formula you can implement immediately. Each component scores 0, 50, or 100 within its category, then gets weighted to produce a total out of 100.

Example calculation: An account with growing usage (100×0.30=30), 85% onboarding (100×0.25=25), last client action 5 days ago (100×0.20=20), NPS of 7 (50×0.15=7.5), and normal support volume (100×0.10=10) scores 92.5 — healthy.

An account with flat usage (50×0.30=15), 45% onboarding at day 30 (0×0.25=0), last client action 18 days ago (0×0.20=0), NPS of 6 (50×0.15=7.5), and 2x support tickets (50×0.10=5) scores 27.5 — critical. This is consistent with the churn signals from our churn prediction model.

How to Implement Without a Dedicated CS Platform

If you do not have a dedicated CS tool, you can implement this in a spreadsheet updated weekly. Create a tab per account. Track each component with the scoring criteria above. Use conditional formatting to show green/yellow/red. Review every account below 40 in your weekly CS team meeting.

As your team grows, migrate to a dedicated platform. Most CS tools — including Gainsight, ChurnZero, and Lyniro — calculate health scores automatically from connected data sources. The advantage is real-time updates and automated alerts rather than manual weekly updates.

Calibrating Your Score Over Time

A health score that you never validate against actual outcomes gradually drifts from reality. Every quarter, run this calibration: pull all accounts that churned in the past quarter and check what their health score was 60 days before they cancelled. If churned accounts were scoring green consistently, your weights are wrong. Adjust the weights toward the components that most consistently predicted the churn.

This calibration loop is what separates health scores that CSMs trust and act on from health scores that CSMs learn to ignore. For the broader set of metrics that work alongside the health score, see our post on the 8 CS metrics every team should track.