A customer health score is a leading indicator that churn is coming before the customer says.
Customer health scoring is the practice of combining product usage, engagement, and relationship signals into a single indicator of how likely a customer is to renew, expand, or churn. Health scores that work give CSMs 60-90 days of intervention time before a churn event; scores that don't work either fire too late (after the customer has already decided to leave) or too often (CSM fatigue, action inertia). The difference is signal selection, not model sophistication.
(1) Product usage — depth (which features), breadth (how many users), and frequency (how often). The most predictive signals here are usually specific 'value moment' actions rather than raw activity: 'created their first automated workflow' beats 'logged in.' (2) Engagement — support tickets, executive touchpoints, exec-team turnover at the customer, community participation. (3) Commercial — contract growth or contraction, invoice payment behavior, ARR trajectory. Skipping any of the three categories produces blind spots.
Simple weighted composite (usage 50%, engagement 30%, commercial 20%) beats sophisticated ML models at Series B/C stage. Start with 5-8 signals, each scored red/yellow/green, roll up to overall red/yellow/green. Refine weights and add signals every 6 months based on which signals actually predicted churn in the last cohort. Machine-learning models earn their keep at $100M+ ARR when you have enough churn events to train on; before then they overfit and CSMs distrust them.
Every product has one or two actions that predict long-term stickiness: for a project management tool it might be 'invited 5+ collaborators,' for an analytics tool 'built and shared a dashboard,' for a CRM 'imported deal pipeline.' Identify yours by looking at usage patterns of customers who renewed happily vs churned. New customers who haven't hit the value moment within 30-60 days are 3-5x more likely to churn. This single signal alone changes CSM prioritization dramatically.
The score is worthless without a playbook of interventions per state. Red-usage: 30-day recovery plan, executive engagement, deployment review. Yellow-engagement: quarterly business review, exec sponsor introduction, adoption program. Red-commercial: procurement early engagement, multi-year renewal proposal, discount authority review. Every state should trigger a specific action assigned to a specific owner with a specific SLA. Scores that inform without triggering action produce a false sense of understanding.
Backtest quarterly: of customers who scored red 90 days ago, how many churned? Of customers who scored green, how many actually renewed? A good score has clear separation (red churns 30-50%, green churns <5%). A useless score has red and green with similar churn rates — usually a signal that the wrong signals are weighted highly. Recalibrate weights on the data, don't defend the current scoring because it's what you built.
Investor directory · Fundraising library · Articles A–Z · Company funding database