Customer Health Scoring
Identifying at-risk customers before they churn—using automated health scores to prioritize customer success efforts.
Why Reactive Churn Prevention Fails
Most companies discover their customers are at risk when those customers cancel. By then, it's too late—retention efforts at the moment of cancellation rarely succeed, and even when they do, the relationship is damaged. Reactive churn prevention also wastes resources. Customer success teams are finite. Without data on which customers are at risk, they end up evenly distributing attention across their entire book of business—or chasing the loudest complainers. Automated health scoring changes this. Instead of reacting to churn signals, you proactively identify at-risk customers and intervene before they cancel. This shifts customer success from reactive firefighting to proactive relationship management.
What Health Scoring Enables
A good health scoring system tells you: which customers are at risk in the next 30, 60, 90 days, which specific risk factors are driving that risk, and what interventions have the best chance of working. This turns customer success from a gut-feel art into a data-driven discipline.
Defining Health Score Components
A customer health score is a weighted combination of signals that together indicate customer wellbeing. Different businesses weight these differently, but common components include: Engagement metrics: How actively is the customer using your product? Declining usage is often the earliest sign of risk. Track login frequency, feature adoption, and key workflows completed. Support patterns: High support ticket volume often correlates with dissatisfaction—or at least confusion. A sudden increase in support tickets after a period of low engagement is a red flag. Contract and billing signals: Late payments, plan downgrades, and contract renewal hesitation all signal risk. External signals: Job changes at the customer company, company layoffs, funding news—these affect customer health even when not directly related to your product. NPS and CSAT scores: Low satisfaction scores predict churn even when current usage looks fine. No single metric tells the full story. Health scoring's value comes from combining multiple signals into a single actionable score.
Building a Rules-Based Health Score
Start with a rules-based health score before investing in ML models. Rules-based scores are interpretable, auditable, and good enough for most use cases. Define thresholds for each metric: weekly active users dropping below 2 per week is warning, below 1 is critical. Support tickets exceeding 5 in a month is warning, 10+ is critical. Combine thresholds into a composite score. Assign each customer a color (green/yellow/red) or numeric score based on how many thresholds they've crossed. Review and adjust thresholds quarterly. What counts as 'warning' will change as you learn from your data. A threshold that's too sensitive creates false positives that waste CS team time; too insensitive and you're missing real at-risk customers.
The Baseline Problem
New customers often have artificially low health scores in their first 30 days—not because they're unhappy, but because they haven't finished onboarding. Build a grace period into your model. Don't flag customers as at-risk until they've been customers long enough to establish a baseline.
ML-Based Health Scoring
For more accurate predictions, machine learning models can identify patterns that rules miss. ML models learn from your historical churn data—which customers churned and what their signals looked like 30, 60, 90 days before—and predict churn risk for current customers. ML health scoring requires: sufficient historical data (at least 100 churned customers to train on), accurate labeling of churned vs retained customers, and regular retraining as your product and customer base evolve. The tradeoff: ML models are less interpretable than rules. You know a customer is high-risk but not exactly which factors drove that score. For some use cases this is fine; for others, interpretability matters. Many teams use both: rules-based scoring for day-to-day triage and ML for identifying subtle patterns that rules miss.
Acting on Health Scores
Health scores are useless without workflows that act on them. Define intervention playbooks based on risk level: High risk (red): Immediate outreach from senior CS team member. Consider executive sponsorship involvement. This customer needs hands-on attention now. Medium risk (yellow): Proactive outreach with value reminders. Share relevant feature announcements, success stories, or check-in calls. Aim to address concerns before they escalate. Low risk (green): Standard relationship maintenance. No special intervention needed. Automate the triggering of these playbooks. When a customer crosses from yellow to red, the CS team should be notified immediately. When a customer stays green for 90 days after renewal, trigger a success story opportunity review. Track intervention effectiveness: are red customers you intervened with retained at higher rates than red customers you didn't reach out to? This tells you whether your playbooks work.
Key Takeaways
- •Reactive churn prevention fails—automated health scoring enables proactive intervention
- •Combine engagement, support, billing, and satisfaction signals into a composite score
- •Start with rules-based scoring for interpretability; add ML for subtle pattern detection
- •Define intervention playbooks for each risk level and automate their triggering
- •Track whether interventions actually improve retention for high-risk customers