How to build a customer health score that warns you before churn.

By the time a customer tells you they are leaving, the decision is usually made. A health score is a way of spotting the signs months earlier, while you can still help. It does not need software or a data team. A spreadsheet updated weekly will do until you have hundreds of accounts.

Choose your signals

Pick four to six signals from these groups. Favour ones you can measure reliably and that you believe predict renewal.

• Usage depth: is the customer using the core feature that delivers their outcome, and how often?
• Usage breadth: what share of the licensed seats or sites are active?
• Outcomes: are they reaching the result agreed at onboarding?
• Support: a rise in tickets, unresolved urgent issues or a poor satisfaction score.
• Relationship: do you have an engaged champion and access to the person who signs the renewal?
• Commercial: late payments, requests to reduce seats, or questions about notice periods.

Build the score step by step

1. Start with traffic lights. For each signal, define what red, amber and green mean. For example, core feature used weekly is green, monthly is amber, not at all in 30 days is red.
2. Combine simply. A customer with any red in outcomes or relationship is red overall. Two or more ambers is amber. Avoid complex weightings until you have data to justify them.
3. Test it against the past. Score every customer who churned in the last year as they looked three months before they left. If most would have shown red or amber, the score is useful. If not, change the signals.
4. Set triggers. Define events that need action within days regardless of the overall score: the champion leaves, usage of the core feature drops by a third over a month (a rule of thumb; tune it to your product), an urgent ticket stays open for a week, or a renewal is 90 days away with no review booked.
5. Write a short play for each trigger. A champion leaving means: find and meet their replacement within two weeks, re-run a short onboarding, and confirm the outcome is still valued.
6. Review weekly. Look at every red and every account that changed colour. Agree an owner and a next action for each.

A worked example

Brindle HR, a fictional Manchester tech startup, sells employee onboarding software to firms of 50 to 500 staff. It scored customers on four signals: new starters processed through the system, managers logging in, open support issues and champion status.

When it tested the score against the previous year, nine of the twelve customers who left had been red or amber three months earlier. Two of the three it missed had been bought by larger groups, so Brindle added “ownership change” as a trigger. The following year, the weekly review caught a large customer whose HR director had left. A meeting with her successor and a short retraining session kept the account.

Common mistakes

• Too many signals. A score with fifteen inputs is hard to explain and nobody trusts it.
• Scoring without acting. A red customer with no owner and no action is just a label.
• Relying on satisfaction surveys alone. Happy users can still churn if the budget holder sees no value.
• Never testing the score. If it does not predict churn in your data, it is decoration.

What to measure

• Share of customers and of recurring revenue in red, amber and green each month.
• Prediction rate: of the customers who churned, how many were flagged at least 60 days before.
• Recovery rate: of red customers, how many returned to green within a quarter.
• Time from trigger to first action.