Designing CS health scores that actually predict churn

A weighted-average single score hides more than it reveals. An account can be green on usage and red on relationship and net-out 'yellow' — which tells the CSM nothing useful. Healthy scoring breaks the model into pillars that each carry their own threshold and recommended play.

Adoption: depth, breadth, and recency of product usage relative to entitlement. Outcomes: attainment of the business outcomes the customer bought for. Relationship: multi-threading depth, executive engagement, and stakeholder stability. Sentiment: signal from support, surveys, and unstructured channels.

Each pillar has segment-specific thresholds, and each red threshold maps to a specific play. This is the unit of work, not the rolled-up score.

Health scores decay. New product features change usage patterns. New segments shift outcome definitions. New competitive pressure shifts what 'engaged executive' looks like. A quarterly tuning ritual — owned jointly by CS Ops and Product Analytics — keeps the model honest.

A health score that only appears in a slide is operational furniture. A health score that triggers tasks in the CSM workflow, surfaces in the QBR brief, and shows up in the renewal narrative is a real system. Build for the second; the first will follow.