Home › Case Studies › Customer Health Score & Churn Prevention Framework

Customer Health Score & Churn Prevention Framework

Worked sample: the company and data are fictional. The model and method are the ones I use in my work.

A weighted composite health score, automated risk tiering, and a CSM playbook with SLAs, built to surface preventable churn 60–90 days before renewal, not 60–90 days after notice. The model targets $3.2M in at-risk ARR.

The problem

CSMs were getting notice of an at-risk account when the customer sent a non-renewal email, too late to actually save it. Health signals existed scattered across product analytics, support tickets, NPS, executive engagement, and billing, but no single owner watched them together. Net Revenue Retention was leaking ~7 points off-target.

The approach

Define a weighted composite score across 6 signals: product usage trend, support ticket sentiment, NPS, executive sponsor activity, billing-on-time, and contract-renewal proximity. Each signal carries a weight that reflects how predictive it is of churn at this customer segment. Auto-bucket accounts into Green / Yellow / Red. Yellow triggers a CSM playbook with a 7-day SLA; Red triggers an exec-sponsor escalation with a 48-hour SLA. The model recomputes weekly, and CSMs see their book sorted by score in their morning view.

What the model is built to do

• Tracks how often Yellow accounts return to Green when the playbook runs within SLA, so the team can test whether early intervention works.
• Flags at-risk enterprise renewals before the customer raises the issue, with the 60–90 day lead time the playbook needs.
• Anchors renewal forecasting on health-score trends instead of CSM gut feel, with the aim of narrowing NRR variance quarter over quarter.
• Gives CSM, Sales, and the exec sponsor a shared definition of "at-risk": when the system says Red, everyone means the same thing.

What I would do differently

Tune the weights with the CSM team in the room before launch, not after. The first weight set was correct in aggregate but misweighted product-usage for the smallest segment (where low usage is normal and not a churn signal). Three weeks of false-Red alerts eroded CSM trust in the score before we recalibrated. A pre-launch backtest with a domain-expert review would have caught it.