Building a Customer Health Score: Inputs, Weights and Thresholds

Pick signals that reflect whether the customer is getting value, and that you can pull without manual work:

• Product usage: logins, active users and use of the features that deliver your core value, compared with what was purchased.
• Adoption depth: how many of the intended teams or use cases are live.
• Support: open tickets, escalations and unresolved bugs.
• Relationship: whether you have a champion and an executive contact, and when you last spoke to them.
• Commercial: renewal date, payment status and any recent downgrade requests.
• Sentiment: survey scores or comments, used cautiously because response rates are low.

Skip signals you can't measure consistently. A score based on a rep's mood produces guesses that look like data. Usage-based products may need different inputs, covered in customer health monitoring for usage-based software.

Set it up in a spreadsheet:

1. List accounts in rows, and put each input in a column.
2. For each input, define what scores green, yellow and red in one plain sentence, for example "weekly active users above the level agreed at onboarding."
3. Convert each input to a score from 0 to 100, or to a simple 1, 2 or 3 scale.
4. Assign weights that add up to 100. For example, you might start with usage at 40, adoption at 20, support at 15, relationship at 15 and commercial at 10.
5. Calculate the weighted score, then set overall bands such as green, yellow and red.
6. Add an override column so a customer success manager can flag risks the data misses, with a required reason.
7. Add a column for the action triggered by each band.

The weights in step four are illustrative. Change them after the back-test below.

A score that doesn't predict outcomes is decoration. Back-test it:

1. Take accounts that renewed and accounts that churned in the past year or two.
2. Calculate what the score would have been three to six months before the renewal date.
3. Compare: were churned accounts mostly yellow or red at that point, and renewed accounts mostly green?
4. Adjust weights and thresholds where they fail. If a signal doesn't separate the two groups, lower its weight or drop it.
5. Repeat each quarter as you learn.

Say your back-test shows churned accounts had low adoption depth but normal login counts. In this example logins are a weak signal and adoption depth should get more weight. Keep the model simple, since a small dataset can't support a score with a dozen inputs.

A score only matters if it changes what people do. Write the response for each band:

• Green: standard cadence. Look for expansion or reference opportunities.
• Yellow: a check-in within a week, a review of the specific low signal and a plan to fix it, owned by one person.
• Red: an escalation with an executive contact, a written save plan and a decision date tied to the renewal.

Feed the results into your business reviews. The QBR template shows how to bring health data into a customer meeting, and the onboarding plan template covers how to prevent low adoption before it starts.

Customer success platforms such as Gainsight and ChurnZero can calculate scores from usage and CRM data and trigger tasks automatically. For a small book of accounts, a spreadsheet is enough to learn what matters first. The platform comparison covers the options.

Watch for these:

• Too many inputs, so no single signal stands out and nobody can explain a score.
• Weights set by opinion and never tested against real outcomes.
• Scores that don't trigger any action, so red accounts sit unchanged.
• Treating every customer alike. A small self-serve account and a large enterprise account show health differently.
• Stale data, such as usage pulled monthly for a metric that changes daily.
• No overrides, so an account with an unhappy executive shows green because usage is high.

For a technical treatment of usage inputs, see health score formulas from telemetry metrics, and for tooling, automated health dashboards in Gainsight.