How to Build Customer Health Scores That Predict Churn
A practical guide to customer health scoring: choosing signals, weighting factors, and turning scores into actionable retention strategies.
TL;DR: Customer Health Scoring Explained
Customer health scoring is a predictive analytics approach that combines multiple data points—product usage, support interactions, billing status, engagement metrics, and relationship strength—into a single composite score indicating how likely a customer is to churn or expand. Health scores typically range from 0-100 and are categorized into tiers (Healthy, Stable, At-Risk, Critical) that trigger different retention interventions. Companies with mature health scoring systems identify 70-80% of at-risk customers before they cancel, enabling proactive retention that saves 15-30% of would-be churn.
The most effective health scores are validated against actual churn data—not based on assumptions about what matters. Product usage signals (login frequency, feature adoption, usage trend) typically carry the highest predictive weight (30-40% of score), followed by engagement signals (email opens, response rates, resource usage) at 20-25%, support signals (ticket volume, sentiment, escalations) at 15-20%, billing signals (payment failures, plan changes, renewal proximity) at 15-20%, and relationship signals (NPS, champion engagement, stakeholder changes) at 10-15%. However, the optimal weighting varies significantly by business model—B2B SaaS with high-touch relationships weights relationship factors more heavily than product-led B2C companies where usage is the dominant predictor.
For implementation, Sequenzy ($19/mo) is our top recommendation for most SaaS companies because it combines health scoring with AI-generated retention email sequences and native billing integration, automatically triggering appropriate intervention campaigns when health scores decline. Enterprise customer success platforms like Gainsight ($50k+/yr) and ChurnZero ($1,500/mo) offer more sophisticated health scoring with CSM workflow management for companies with dedicated CS teams. Start with 5-7 validated signals, weight based on churn correlation, and define clear intervention playbooks for each health tier. Review and refine quarterly based on actual churn outcomes.
Customer health scoring takes multiple signals - product usage, support interactions, billing status, engagement metrics - and combines them into a composite score that predicts how likely a customer is to churn or expand. A good health score enables proactive intervention before customers decide to leave.
Building effective health scores requires understanding which signals actually predict churn in your business, weighting them appropriately, and creating workflows that act on the scores. This guide walks through the process.
What Are Customer Health Scores?
Customer health scores are quantitative indicators that synthesize multiple data points into a single metric representing customer success likelihood. Think of them as a leading indicator—the canary in the coal mine that signals trouble before actual churn occurs. Unlike lagging metrics like churn rate or NRR, health scores enable proactive rather than reactive retention strategies.
Health scores typically range from 0-100, where higher scores indicate healthier customers more likely to renew and expand. Scores are calculated by combining weighted signals across five categories: product usage (how customers use your product), engagement (how customers interact with your company), support (the quality and quantity of support interactions), billing (payment history and contract status), and relationship (the strength of the human connection).
The power of health scoring lies in converting scattered qualitative signals into actionable quantitative intelligence. Rather than relying on CSM gut feel or random check-ins, health scores provide systematic early warning that helps retention teams prioritize their limited time on the accounts most likely to churn. When combined with automated intervention systems, health scores enable scaled proactive retention without proportional headcount growth.
Health Score Signal Categories Compared
| Signal Category | Key Metrics | Predictive Power | Data Source | Update Frequency |
|---|---|---|---|---|
| Product Usage | Login frequency, feature adoption, usage trend | Very High | Product analytics (Mixpanel, Amplitude) | Daily |
| Billing | Payment failures, plan changes, renewal date | High | Stripe, billing platform | Real-time |
| Support | Ticket volume, sentiment, escalations | Medium-High | Zendesk, Intercom, support tool | Daily |
| Engagement | Email opens, response rates, resource usage | Medium | Email platform, analytics | Daily |
| Relationship | NPS, champion engagement, stakeholder changes | Medium (varies by model) | CSM notes, survey tools | Weekly |
Choosing Health Score Signals
The signals you include should actually predict churn, not just feel important. Start with these categories and test which signals correlate with actual outcomes:
Product Usage Signals
- Login frequency and recency
- Feature adoption breadth and depth
- Time spent in product
- Key action completion
- Usage trend (increasing, stable, declining)
Engagement Signals
- Email open and click rates
- Response to outreach
- Webinar or event attendance
- Documentation or resource usage
- Community participation
Support Signals
- Ticket volume and trend
- Ticket sentiment and severity
- Time to resolution
- Escalation frequency
- CSAT scores on tickets
Relationship Signals
- NPS or satisfaction scores
- Champion engagement level
- Stakeholder changes
- Expansion conversations
- Referral activity
Billing Signals
- Payment history (failures, recoveries)
- Contract value trend
- Time to renewal
- Discount or concession history
- Plan changes (upgrades, downgrades)
Weighting and Scoring
Not all signals matter equally. A support escalation might predict churn 3x better than a missed login. Weight signals based on their predictive power, which you discover through analysis of historical churn data.
Simple Approach
Start with a simple weighted average. Assign each signal a score (1-10) and a weight based on perceived importance. Calculate: (Signal1 x Weight1 + Signal2 x Weight2 + ...) / Sum of Weights.
Validated Approach
As you gather data, validate which signals actually predict churn. Run correlation analysis between each signal and churn outcomes. Adjust weights based on actual predictive power. Remove signals that don't add value.
Health Tiers
Translate numeric scores into actionable tiers:
- Healthy (80-100): Low intervention needed, focus on expansion
- Stable (60-79): Monitor closely, proactive engagement
- At Risk (40-59): Active intervention required
- Critical (0-39): Urgent save efforts needed
Acting on Health Scores
Health scores are useless without action. Define interventions for each health tier:
Automated Actions
- Trigger email sequences when health drops
- Assign tasks to CSMs for at-risk accounts
- Send alerts when critical accounts emerge
- Update CRM records with health data
Human Actions
- CSM outreach for at-risk accounts
- Executive involvement for critical accounts
- QBR scheduling based on health trends
- Escalation procedures for rapidly declining accounts
Validating Your Health Score
Regularly check whether your health score actually predicts churn:
- Do low-health accounts actually churn more than high-health accounts?
- What percentage of churned customers were flagged as at-risk?
- Are there false positives (flagged but didn't churn)?
- Are there false negatives (churned without being flagged)?
If your health score isn't predictive, adjust signals and weights until it correlates with actual outcomes.
Tools for Health Scoring
Customer success platforms like Gainsight, ChurnZero, Vitally, and Custify provide health scoring capabilities. For email-focused retention, tools like Sequenzy include health signals integrated with retention email automation.
Start simple and iterate. A basic health score that triggers action beats a sophisticated score that sits in a dashboard unused.
Frequently Asked Questions About Customer Health Scoring
How accurate are customer health scores at predicting churn?
Mature health scoring systems typically identify 70-80% of at-risk customers before they cancel, with false positive rates around 20-30%. However, accuracy varies significantly based on business model and signal quality. Product-led B2C companies with clear usage patterns often achieve higher accuracy (80%+) than relationship-driven B2B enterprise companies where human factors matter more (60-70%). The key is continuous validation against actual churn data—if your health score doesn't correlate with outcomes, adjust signals and weights until it does.
How many signals should I include in my health score?
Start with 5-7 validated signals across the major categories (usage, billing, support, engagement, relationship). More signals don't necessarily mean better scores—the goal is predictive power, not complexity. In fact, too many signals can create noise and make scores harder to interpret. Focus on signals that show strong churn correlation through analysis. You can expand over time as you validate additional predictors, but 5-10 well-chosen signals typically outperform 30+ poorly chosen ones.
How often should health scores update?
Update frequency depends on signal volatility. Product usage and billing signals should update daily or in real-time. Support signals should update daily as tickets are created and resolved. Relationship signals (NPS, champion engagement) typically update weekly or monthly. Your overall health score should recalculate whenever component signals change significantly—daily recalculation is standard for most SaaS businesses. Real-time health scores are becoming more common and valuable, especially for triggering immediate intervention when critical signals decline.
Should I share health scores with customers?
Generally no. Health scores are internal tools for retention strategy, not customer-facing metrics. Sharing can create awkward conversations and competitive dynamics. However, some companies successfully share simplified "success scores" or "adoption metrics" that show customers their progress toward value realization without framing it as a risk assessment. If you do share, focus on positive framing—what they're doing well, opportunities for more value—rather than risk indicators.
How do I choose between simple and complex health scoring approaches?
Start simple. A basic weighted average of 5-7 validated signals delivers 80% of the value with 20% of the complexity. Only invest in sophisticated approaches (machine learning models, ensemble methods, predictive analytics platforms) when you've validated that simple scoring isn't sufficient and you have the data science resources to implement and maintain complex systems. Many companies discover that simple scores work fine, especially when combined with effective intervention playbooks. The goal is actionable insight, not algorithmic sophistication.
What's the difference between health scores and churn prediction models?
Health scores and churn prediction models serve similar purposes but differ in approach. Health scores are typically transparent, rules-based systems where humans define signals and weights—this makes them interpretable and adjustable. Churn prediction models use machine learning to find non-obvious patterns in large datasets, often achieving higher accuracy but as "black boxes" that are harder to understand and adjust. Most mature companies use both: health scores for day-to-day retention operations and churn models for strategic planning and resource allocation. Tools like Sequenzy combine simple health scores with AI-powered churn prediction for comprehensive coverage.
Turn health signals into retention action
Sequenzy triggers AI-generated retention sequences based on customer health.
Try Sequenzy Free