Customer health score: green until it cancels

The account was green on Monday. On Thursday the customer gave notice that they won’t renew. Looking back, the signs were there: logins down for two months, three escalations, and a detractor score from the project lead. The health score averaged them into a 71, and 71 was green.

Scoring customer health is stage six of a customer 360. It’s where usage, support, feedback and billing come together into something a CSM can act on, and where most teams compress them into a single number nobody trusts.

What is a customer health score?

A customer health score summarizes how likely an account is to renew, shrink or grow, using signals such as product usage, support history, survey responses, payment behavior and engagement with your team. A useful score shows its categories and its reasons as well as a total, is checked against which customers churned, and alerts when a category moves.

Why every account is green until it cancels

1. The signals were chosen because they were available. The score uses whatever was easy to pipe in, such as logins and ticket counts, instead of the signals that came before churn in your own data.
2. Everything collapses into one number. Two accounts at 62 can need completely different conversations, and the total doesn’t say which.
3. Nobody checked it against churn. The weights were agreed in a workshop, and nobody compared last year’s scores with the customers who left.
4. It alerts on level, not movement. The total moves slowly, so a sharp drop in one category is diluted by the others.
5. Renewal dates aren’t in it. A falling score 300 days from renewal and the same score 45 days from renewal get the same color.
6. It only looks for risk. Accounts using more than they bought don’t trigger anything, so expansion depends on someone noticing.

What it costs

A score nobody trusts gets ignored. CSMs go back to their own judgment and their own spreadsheets, and the company pays for a health score and a renewal process that don’t talk to each other.

The worse cost is false comfort. Leadership reports a mostly green book of business, and the renewal forecast is wrong by exactly the accounts that were green until they weren’t.

What to fix first

1. Pick the signals that predicted churn in your data. Look at the last year of churned and renewed customers, and keep the signals that separated the two groups.
2. Keep the categories visible. Give usage, support, sentiment, billing and engagement each their own status and their top reason.
3. Back-test before anyone relies on it. Score last year’s customers with the model and count how many churned accounts it flagged, and how early.
4. Alert on movement. Tell the account owner when a category changes, with the reason.
5. Weight by time to renewal. Put at-risk accounts with renewals in the next 90 days at the top of the queue.
6. Flag expansion too. Send accounts at or above what they bought, with a renewal approaching, to the account owner as expansion candidates.
7. Track what happened to flagged accounts. If they didn’t churn and nobody acted, the model needs work.

The guide to building customer health signals has the prompts for each step.

How we flagged renewals and expansion at Tray.ai

Two earlier workflows at Tray.ai covered each end of this. One ran daily against Salesforce, found accounts with a contract end date approaching and no renewal under way, and sent the retention team a Slack message with the account and its owners. The other ran monthly against usage data in the warehouse, found accounts at or above their plan’s usage with a renewal approaching, and sent each account team a report of its expansion candidates. Each used one signal. A health score puts both in one place, with the reasons.

Where it fits

Health needs every stage before it: product usage, support history, survey responses and billing, all joined on one customer ID. Once it exists, reverse ETL puts the status and its reasons back in the CRM, where account teams work.

For why a single customer view rarely holds up once four teams need it, read why the customer 360 never sticks.