Glossary/Methodologies and Frameworks

Health Score

May 27, 2026·By Jeff Moss

A composite metric meant to summarize a customer's likelihood to renew, expand, or churn. Adopted near-universally in customer success operations and customer success platforms. Near-universally ineffective in practice. The fix is not to perfect the composite; it is to abandon the composite and track risk specifically, not generically.

"An account at 28 percent at-risk that drops to 24 percent at-risk does not tell anyone what changed or what worked. Composite scores are not operable."

What it is

A Health Score is a composite metric meant to summarize a customer's likelihood to renew, expand, or churn. Adopted near-universally in customer success operations and customer success platforms. Near-universally ineffective in practice.

The logic of a composite health score makes sense on paper: identify the factors that play into whether a customer succeeds, track them, weight them by impact, render the result as a single score. In practice the construct fails for eight reasons.

One. It assumes you know what factors matter. To create a health score, you have to know what creates risk, and you have to know the inverse: what factors lead to success. The mistake everybody makes is focusing on factors derived from failure: looking at churned customers and saying "they never turned this feature on, therefore that feature matters." Customers who fail do not use a lot of things, and their non-use is not why they failed. Successful customers do not use everything either; they use a few things religiously, and those are the factors. Identifying success factors is harder work, and most teams skip it.

Two. Most teams do not know those factors.

Three. They do not know how to approach the weighting.

Four. Almost everyone chooses lagging indicators. By the time the lagging indicator fires (they removed all their users, they stopped using the platform), the customer has already evaluated another solution and it is over.

Five. Identifying leading indicators is hard. You have to know what success looks like before you can identify the early signals of its absence.

Six. Composite weighted scores add too much nuance. What is the difference between a customer at 68 percent at-risk and 74 percent at-risk? No one knows how to operate against that.

Seven. You cannot remediate risk generically, only specifically. An account at 28 percent at-risk that drops to 24 percent at-risk does not tell anyone what changed or what worked. Composite scores are not operable.

Eight. Everyone misunderstands what the health score is for. One number cannot do four jobs: predict churn, surface risk, show what successful customers look like, and predict renewal. A customer can have real risk in the account and still renew year after year. The point is to track the amount of risk that exists; renewal probability is a separate calculation.

The fix is to track risk in two specific categories.

First, customers who have never achieved a performance threshold. They never set up the feature, never used it the way success requires (10 times a day, if that is what success looks like). This category dominates the first 90 days or first year.

Second, customers who achieved performance but dipped below the threshold. Post-onboarding sustainment risk.

Across both categories, less is more. Use a core set of essential factors: behaviors outside the system, behaviors inside the system, technical accomplishments, outcomes (first results and full results), and a time scale for when each should be accomplished. If a customer has not accomplished something within the recommended time, that is the specific named risk. The CSM no longer reports "this account is 32 percent at-risk." They report "this account is overdue on the day-30 deployment milestone. The intervention is a working session with their integration team this week." Specific. Time-bound. Operable.

Source: methodology-canon/methodology-reference.md > Health Score

Key Takeaways

•Composite health scores are near-universally ineffective for eight specific reasons.
•The fix is not to perfect the composite. It is to abandon the composite and track risk specifically, not generically.
•Track risk in two categories: customers who have never achieved a performance threshold, and customers who achieved performance but dipped below the threshold.
•Less is more. Use a core set of essential factors (behaviors outside the system, behaviors inside the system, technical accomplishments, outcomes, and a time scale for each).
•The CSM should report named risks ('overdue on the day-30 deployment milestone'), not composite scores ('32 percent at-risk'). Specific. Time-bound. Operable.

The Mistake Everybody Makes

Focusing on factors derived from failure. Looking at churned customers and saying 'they never turned this feature on, therefore that feature matters.' Customers who fail do not use a lot of things, and their non-use is not why they failed. Successful customers do not use everything either; they use a few things religiously, and those are the factors. Identifying success factors is harder work, and most teams skip it. Combined with composite weighting, lagging indicators, and the conflation of four separate jobs (predict churn, surface risk, show success patterns, predict renewal), the composite health score is not operable.

How to Apply

1.Replace the composite score with a specific risk model.
2.Identify the essential factors for your customer's success (success factors, not failure factors).
3.Define a time scale for each factor.
4.Flag any account missing a factor at its current time threshold with a specific named risk.
5.Run interventions against the named risk, not the score.
6.Treat renewal probability as a separate calculation that consumes the risk picture, not the same metric.

Examples

From 75-factor composite to 5 named milestones

A CS team replaces its 75-factor composite health score with a specific risk model: five outcome milestones the customer should hit by specific times. Integrated with their CRM by day 14, deployed first workflow by day 30, hit good benchmark on target metric by day 60, expanded to a second team by month 9, hit best benchmark by month 12. Any account missing a milestone at its current time threshold gets flagged with a specific named risk factor. The CSM no longer reports 'this account is 32 percent at-risk.' They report 'this account is overdue on the day-30 deployment milestone; the intervention is a working session with their integration team this week.' Risk becomes specific, time-bound, and operable.

Where Used in Product

Future product surface:Health and risk surfaces (post-GA): the specific-risk model replaces the composite score across CS operations.
Coaching:The AI coach uses the named-risk flag to recommend interventions per account.
Methodology:The Customer Strategy Planner's phase-position diagnoses where each customer is on the risk timeline.

Jeff Moss

Founder, Expansion Playbooks. A decade studying customer retention and expansion data from 75+ recurring revenue businesses.

Back to glossary