Customer health scores are useful only when they help a team decide what to do next. A single number can create false confidence if nobody understands which evidence moved it or what action it should trigger.
Before building a complex customer health model, start with a small set of observable signals and a clear review process.
1. Product adoption
Track whether the customer uses the workflows that create value, not only total logins. A decline in a core action may matter more than a stable weekly active-user count.
2. Change from the customer’s normal pattern
There is no universal healthy usage level. Compare each account with its own baseline and look for meaningful changes over time.
3. Unresolved customer friction
Open support issues, repeated questions and delayed implementation steps are operational signals. They often explain risk earlier than a quarterly survey.
4. Stakeholder engagement
A missing champion, fewer meetings or a new stakeholder entering the conversation can change the account context. Capture the change and give the CSM a reason to review it.
5. Promises without visible progress
A customer commitment or internal follow-up that has no owner or next step is a risk to trust. Keep these items visible until they are resolved.
6. Expansion intent
Requests for access, adjacent use cases or new-team workflows can indicate expansion potential. Treat them as hypotheses to validate, not automatic sales opportunities.
7. Evidence quality
Every health signal should show its source, date and confidence. If the team cannot explain why the score changed, the score will not improve decision-making.
A practical AI customer health workflow can monitor these signals, summarize what changed and prepare a review queue. It should not silently change account status or send messages without approval.
The goal is not the most sophisticated model. The goal is a health review that helps a lean customer success team spot the right accounts earlier, understand the reason and take a better next step.
