A health score needs a next action.

Keep the model explainable

I use five inputs to bring product behavior, relationships and account context into one assessment. Usage carries the largest weight because it is concrete and can be reviewed regularly. Stakeholder engagement comes next because a relationship that depends on one person is exposed when that person moves.

I score each input from 0 to 100, multiply each score by its weight and add the results. The resulting composite stays on the same 0–100 scale. I keep the underlying inputs visible so that the total can be explained.

Read the evidence behind each score

I distinguish activity from adoption. Frequent logins are less useful than evidence that people use the workflows connected to the customer's goals. I look at both the depth of use and whether it is improving, stable or declining.

I make the same distinction in relationships. A full meeting calendar does not establish engagement if the same contact attends every call and the executive sponsor is absent. I look at who is engaged, what is being discussed and when the last meaningful conversation happened.

Support needs context as well. A count of tickets hides severity and repetition. I want to know whether difficult issues are recurring and whether the customer considers them resolved. During onboarding, I look at milestone progress; after onboarding, I look for evidence that adoption is deepening.

The input scores should capture those differences. A composite can otherwise hide a weak relationship behind strong activity, or obscure a stalled implementation inside a seemingly acceptable average.

Attach a response to each status

I use these starting thresholds to connect the score to an operating response. The status names are written out; the response does not depend on recognizing a color.

I review status changes in the weekly cadence. The useful discussion is what changed, which input explains it and who owns the response. A change in the total without an explanation is a prompt to inspect the evidence.

Calibrate manually before automating

I start in a spreadsheet. Each CSM scores the accounts they manage, using the same definitions. I then review the edge cases together: accounts that appear healthier than the relationship feels, or accounts that different people would score differently.

1. Score the accounts. Record the input scores and the evidence behind them.
2. Discuss disagreement. Resolve ambiguous definitions before treating the composite as reliable.
3. Review weekly. Examine status changes and the progress of intervention plans.
4. Complete a cycle. Run the model manually for 90 days before making automation decisions.

That cycle provides evidence about which signals are useful in the actual business. Selecting a platform first makes it too easy to encode assumptions before anyone has tested the operating logic.

Judge the model by the decisions it supports

I look for earlier risk visibility, clearer portfolio discussions and renewal preparation grounded in account evidence. If the score does not support those outcomes, I inspect the input definitions and the response process.

A score also needs somewhere to go. The playbook establishes the ownership and cadence. The KPI hierarchy connects account-level signals to the retention and growth outcomes being reviewed.

I want a CSM to be able to explain both the assessment and the next action without relying on a dashboard to supply the judgment.