Customer Health Score

There is no customer health score. When leadership asks 'which customers are at risk?' the answer is anecdotal — CSMs offer gut feelings based on their last conversation. 'I think Acme is fine' and 'I'm worried about Beta Corp' is the extent of health visibility. There is no metric, no scoring model, and no written definition of what 'healthy' means.

None — AI cannot perform any health assessment or churn prediction because no customer health score definition or records exist in any system.

CSMs assign customer health scores based on personal judgment — a red/yellow/green flag in the CRM. The criteria are unwritten and subjective. One CSM marks an account yellow because of a tough support call; another CSM would have called the same situation green. The customer health score reflects the CSM's personality more than the customer's actual risk level. Comparing health scores across CSMs is meaningless.

AI could aggregate the subjective customer health score flags into dashboards, but cannot build predictive models because the scores reflect CSM judgment variability rather than consistent risk signals.

A customer health score model exists with defined input dimensions — usage frequency, support ticket volume, NPS response, contract value trend. Each dimension has a documented scoring rubric. CSMs calculate the customer health score periodically by pulling metrics from multiple systems and applying the formula manually. The model is explicit, but computing the score is labor-intensive and happens monthly at best.

AI can validate manually computed customer health scores against the documented rubric and flag inconsistencies, but cannot auto-calculate scores because the input metrics live in disconnected systems without programmatic access.

The customer health score computes automatically from connected input signals — daily active users, feature adoption rates, support ticket severity, NPS trends, payment history, and engagement frequency. The scoring model is documented, and each component's weight is explicit. A CSM can see exactly why a customer health score dropped: 'DAU declined 30% and a P1 ticket has been open for 7 days.' The score is data-driven, not opinion-driven.

AI can calculate, explain, and act on customer health scores using the connected metrics. Can identify at-risk accounts and trigger playbooks. Cannot yet optimize the scoring model itself because there is no feedback loop from actual churn outcomes to health score accuracy.

The customer health score is a formal entity in a predictive analytics framework. The scoring model is versioned, backtested against historical churn outcomes, and continuously validated. Component weights are calibrated by machine learning. An AI agent can ask 'what is the sensitivity of Acme Corp's customer health score to a 20% DAU decline?' and get a precise, model-driven answer with confidence intervals.

AI can autonomously manage customer health scoring — recalibrating model weights, detecting scoring anomalies, identifying emerging risk patterns, and recommending model updates based on outcome validation.

The customer health score is a living, self-calibrating metric. Every product interaction, support conversation, billing event, and engagement signal recalculates the score in real-time. The scoring model continuously learns from churn and expansion outcomes, adjusting component weights and adding new signals without manual tuning. The customer health score is not a computed metric — it is an emergent property of the entire customer event stream.

Can autonomously maintain, calibrate, and evolve the customer health score model in real-time. The health score is a self-learning system that becomes more predictive over time without human intervention.

Ceiling of the CMC framework for this dimension.