CHURN IS DEAD
The Case Against Customer Health Scores
6 min read · Strategy
Archive note: This issue predates the evidence ledger introduced in August 2026. Treat uncited benchmarks and examples as editorial analysis, not independently verified findings.
Customer health scores are the astrology of customer success.
Both promise to predict the future using complex systems of measurement. Both give their believers a sense of control over uncertain outcomes. And both consistently fail when it matters most.
The difference? When your horoscope is wrong, you lose nothing. When your health score is wrong, you lose customers.
Yet 94.8% of CS teams now use AI-powered health scoring. Platforms compete on algorithm sophistication. CS leaders present colored dashboards to executives as proof of predictive mastery. And customers still churn without warning.
This isn't a technology problem. It's a belief problem.
The False Prophet of Prediction
The health score industry has convinced us that customer behavior is algorithmically predictable. Feed enough data points into the machine—usage metrics, support tickets, payment history, engagement scores—and it will divine which accounts need attention.
This is mathematical astrology.
Real customer relationships aren't reducible to data points. They're built on context that doesn't live in your CRM. The budget freeze that happened in a hallway conversation. The new executive who hates your product category. The internal champion who just took another job.
Your health score sees the metrics. It misses the story.
Meanwhile, frontline CSMs develop an intuitive sense of account risk. They notice when stakeholder responses become shorter. They feel the energy shift in renewal conversations. They detect the subtle signs that precede every churn.
But when the dashboard shows green, leadership trusts the algorithm over the human.
The Confidence Trap
Health scores don't just fail at prediction. They actively make teams worse at retention.
Here's how:
False negatives create blind spots. Your highest-value enterprise client scores 98% healthy. Usage is up. Support tickets are down. The executive stakeholder attended your last QBR. Then they submit a 90-day cancellation notice citing fundamental product-market fit issues.
The algorithm missed what a 15-minute phone call would have revealed: they've been evaluating competitors for six months.
False positives waste resources. Mid-market accounts with 60% health scores get aggressive intervention campaigns. CSMs run discovery calls to identify expansion opportunities. Account managers get pulled into "strategic" conversations.
Meanwhile, the account was never at risk. They're just seasonal users who go quiet every Q4. The health score algorithm doesn't know their business cycle.
Gaming behaviors emerge. When CSM performance reviews include health score improvement targets, teams optimize for the metrics rather than the outcomes.
Need to boost a health score? Send more engagement emails. Schedule unnecessary check-ins. Push adoption of features that don't matter to the customer's business.
The score improves. The relationship deteriorates.
The Data Theater Performance
Every CS platform promises better health scoring through more sophisticated data inputs. User activity patterns. Email sentiment analysis. Support ticket language processing. Invoice payment timing.
More data points don't create better predictions. They create better theater.
Executives see dashboards with hundreds of weighted factors and assume the system is scientifically rigorous. CS leaders present trend analyses and statistical correlations. The complexity gives everyone confidence that this time, they've solved the prediction problem.
But complexity isn't accuracy.
The most predictive signals are often the simplest ones. How often does the primary contact respond to your emails? Do they ask questions about contract terms six months before renewal? Has their org chart changed recently?
These signals require human interpretation, not algorithmic processing.
What Actually Predicts Churn
After analyzing churn patterns across dozens of enterprise SaaS companies, the most reliable leading indicators aren't data points. They're relationship shifts:
Stakeholder availability. When champions become hard to schedule, expansion stops and renewal risk increases. No algorithm captures calendar availability or email response time patterns reliably.
Question types change. Healthy customers ask "how" questions: How do we implement this feature? How do we train more users? At-risk customers ask "why" questions: Why does this cost so much? Why can't we do this ourselves?
Internal narrative shifts. The customer stops talking about future plans with your product and starts talking about current limitations. Their language moves from partnership to vendor management.
Context changes. Budget cuts, acquisitions, leadership changes, strategic pivots. These create churn risk regardless of product usage or satisfaction scores.
None of these signals live in your data warehouse. They emerge in conversations, emails, and meetings. They require human pattern recognition, not machine learning.
The Signal vs Noise Framework
The solution isn't better health scores. It's knowing which signals matter and which create false confidence.
The Signal vs Noise Audit helps teams separate genuine churn predictors from algorithmic theater.
Relationship Signals
*What humans detect that algorithms miss*
- Communication patterns: Response time changes, email tone shifts, meeting frequency drops
- Stakeholder engagement: Champion availability, executive involvement, internal advocacy strength
- Strategic alignment: Future planning discussions, partnership language, expansion conversations
- Context awareness: Organizational changes, budget cycles, competitive evaluations
These signals require human observation and interpretation. They can't be automated but they can be systematized through structured customer conversations and relationship mapping.
Data Noise
*What systems measure that creates false confidence*
- Usage metrics: Daily active users, feature adoption rates, session duration
- Support metrics: Ticket volume, resolution time, satisfaction scores
- Engagement scores: Email opens, content downloads, webinar attendance
- Financial indicators: Payment timing, invoice disputes, contract utilization
These metrics correlate with customer health but don't predict churn timing. They're useful for operational management but dangerous for risk prediction.
Predictive Gaps
*What gets missed between algorithms and reality*
- Seasonal patterns: Business cycles that look like engagement drops
- Organizational context: Internal changes that affect tool priorities
- Competitive dynamics: Alternative solutions being evaluated
- Strategic shifts: Business model changes that affect product need
These gaps exist because algorithms optimize for historical patterns while business contexts constantly evolve.
Action Triggers
*What actually drives effective intervention*
- Relationship maintenance: Regular strategic conversations with key stakeholders
- Context monitoring: Systematic tracking of organizational and industry changes
- Value reinforcement: Ongoing demonstration of business impact and ROI
- Expansion alignment: Identifying growth opportunities within existing usage patterns
These triggers focus on relationship strength rather than risk mitigation. They prevent churn by building deeper partnerships rather than detecting early warning signs.
The Path Forward
Stop optimizing health scores. Start optimizing relationships.
The teams that prevent churn don't predict it better. They build stronger partnerships that survive the contextual changes that health scores can't detect.
This requires a fundamental shift: from algorithmic confidence to relationship intelligence. From data dashboards to customer conversations. From prediction theater to partnership depth.
Your CSMs already know which accounts are at risk. Trust their judgment over the algorithm's confidence.
The health score will tell you everything except what you need to know.
By Kuber Sethi · All issues · Subscribe