Customer Health Scores That Actually Predict Churn

For subscription and recurring-revenue businesses, keeping customers matters as much as winning them. Customer success teams therefore try to spot customers at risk of leaving early enough to help. The usual tool is a customer health score: a single rating, often red, amber or green, that summarises how a customer relationship is going.

The idea is sound. In practice, many health scores fail. They are built from whatever data was easy to get, weighted by intuition, and never checked against which customers actually left. Customer success managers learn to ignore them and rely on gut feel instead.

This article explains how to build a health score that genuinely predicts churn and leads to useful action.

What a health score is for

A health score has two jobs:

1. Early warning. Flag customers at risk of leaving or reducing spend, early enough to act.
2. Prioritisation. Help teams decide where to spend limited time.

A score that does neither, because it flags everyone or no one, or because it reacts only when the customer has already decided, is not doing its job.

Choose signals that lead, not lag

The most common mistake is relying on lagging signals, which change only after a customer has decided to leave. A cancellation notice is the ultimate lagging signal. Useful health scores combine leading indicators from several categories.

Product usage

• Active users compared with licences purchased.
• Usage of the features most associated with value.
• Trends: falling usage often matters more than low usage.

Outcomes

• Whether the customer is achieving the goals they bought the product for.
• Progress against agreed success measures.

Relationship

• Engagement of key contacts and executive sponsors.
• Whether a key sponsor has left the customer organisation.
• Attendance at business reviews.

Support and experience

• Volume and severity of support cases, and especially unresolved ones.
• Satisfaction or effort scores after interactions.
• Escalations and complaints.

Commercial

• Payment delays.
• Requests for discounts or contract changes.
• Time remaining until renewal.

Not all of these will be available or relevant. Start with the ones you can measure reliably.

Validate against real outcomes

This is the step most organisations skip. Before trusting a health score, test it against history:

1. Take customers who renewed and customers who left over the past year or two.
2. Look at their signals three to six months before renewal or cancellation.
3. Ask which signals differed between the two groups.

You will often find surprises. Signals that seemed important may make little difference, while others, such as the loss of a key sponsor or a drop in usage of one particular feature, turn out to be strong predictors.

Use what you learn to choose and weight signals. Then keep testing: each quarter, check whether customers flagged red were in fact more likely to leave than those flagged green.

Keep the score explainable

A health score that nobody understands will not be trusted. Whatever method you use, customer success managers should be able to see why a customer is amber or red: which signals moved and when. That explanation is what turns a score into a conversation and an action.

Combine data with judgement

Data never captures everything. A customer may be quietly evaluating a competitor, going through a merger or changing strategy. Allow customer success managers to add their own assessment alongside the calculated score, and track whether their judgement improves predictions. Often the best results come from combining the two.

Tie scores to actions

A health score only matters if it changes what people do. Define clear responses:

• Green: maintain regular engagement, look for expansion opportunities aligned with the customer's goals.
• Amber: investigate the cause, agree a plan with the customer, and check again within a set period.
• Red: escalate, involve senior leaders where appropriate, and run a structured recovery plan with clear owners.

Track whether these actions work. Which recovery approaches actually saved at-risk customers? Over time, this builds a playbook of what to do for each kind of risk.

Common pitfalls

• Too many signals. Twenty signals with arbitrary weights produce noise. Five to eight well-chosen ones usually work better.
• Static weights. Customer behaviour changes. Revisit the model at least annually.
• One score for everyone. Customers of different sizes, segments or products behave differently. Consider separate models or thresholds.
• Score without context. Show trends, not just the current colour.
• Using it to judge the team. If customer success managers are rewarded for green scores, scores will drift green.

How mature is your customer success practice?

Updated October 2026: we have added a Customer Success assessment. It covers strategy and segmentation, onboarding and adoption, customer health and risk, value realisation, renewals and expansion, and how customer insight reaches the rest of the organisation.

Retention is shaped by the whole customer experience, not just the customer success team. The Customer Experience assessment looks across journeys, service delivery, feedback and governance.

Frequently asked questions

Do we need machine learning to build a health score?

No. A simple, well-validated score based on a handful of signals often performs well. More advanced models can help once you have plenty of historical data and a clear process for acting on results.

How far in advance can a health score predict churn?

It depends on your renewal cycle and signals. Aim to identify risk at least three to six months before renewal, so there is time to act.

Should customers see their health score?

Usually not the score itself, but sharing the underlying measures, such as adoption and progress towards goals, in business reviews is often valuable and builds trust.