Customer Health Scores: Where Teams Get Them Wrong
A customer health score is useful only if it changes what a team does. Too many organizations build elaborate scoring models, assign colors to accounts and then quietly continue operating exactly as before.
Start with the decision, not the data
Before adding another signal, ask what decision the score is supposed to improve. Should a CSM change engagement cadence? Should leadership join a customer call? Should product adoption become the priority? If the answer is unclear, the metric is probably ornamental.
Use multiple kinds of signal
Usage matters, but so do executive engagement, support history, stakeholder turnover, implementation progress, commercial timing and the customer’s own stated outcomes. A healthy account is not simply one with a lot of logins.
Make the score explainable
A CSM should be able to explain why an account is yellow or red in plain language. If nobody can explain the score without opening a spreadsheet of formulas, the model has become harder to trust.
Close the loop
The final step is action. Every meaningful risk condition should have an owner, a next step and a review point. Health scoring becomes valuable when it creates earlier intervention and more consistent judgment.
A practical health-score framework
Start with a small scorecard that connects evidence to a decision. Define the customer outcome, the signals that indicate progress, the owner of each signal and the response to deterioration. Keep the underlying evidence visible alongside the color. A score without a reason or a next action is incomplete.
The following framework is illustrative. Its weights and thresholds are starting points to test against your customer segments, not a proprietary process or a claim about results at a former employer.
Seven signal categories to review
- Usage and adoption: Track the workflows and roles needed to achieve the agreed outcome. Compare meaningful adoption with the customer's expected operating pattern. A seasonal decline or completed project needs different treatment from a failed rollout.
- Executive engagement: Look for an active sponsor, agreement on priorities and participation in value discussions. Meeting attendance alone does not prove sponsorship.
- Support history: Consider severity, recurrence, age of unresolved issues and the customer's experience of resolution. A high ticket count can reflect active adoption; a repeated blocker can threaten value even when ticket volume is low.
- Stakeholder changes: Record departures, role changes and new decision makers. Revalidate the success plan when a champion leaves instead of assuming the relationship transfers automatically.
- Implementation progress: Compare milestones, dependencies and acceptance criteria with the agreed plan. Identify whether delays sit with your team, the customer or a shared dependency.
- Commercial timing: Track renewal dates, notice periods, procurement steps and budget decisions. An approaching renewal increases urgency; it does not, by itself, prove poor health.
- Stated customer outcomes: Record what the customer wants to accomplish and the evidence they accept as progress. Distinguish an outcome confirmed by the customer from a benefit inferred by your team.
Weight the signals without hiding the judgment
One illustrative model assigns 25% to customer outcomes, 20% to usage, 15% to implementation and 10% each to executive engagement, support, stakeholder continuity and commercial readiness. Rate each category from 0 to 100 using written criteria, then multiply each rating by its weight and add the results. Publish those criteria with the model so two CSMs can assess similar evidence consistently.
Adjust the weights by lifecycle and segment. During onboarding, implementation dependencies may matter more than mature usage. For a complex enterprise relationship, sponsorship and stakeholder coverage may deserve more weight. Check proposed weights against past account histories before treating them as useful predictors.
Missing data is unknown, not healthy. Show data freshness and confidence separately; do not quietly replace missing evidence with a green value. Allow a documented override for a material event such as a stated intention to leave. Include who made the override, why it was necessary and when it will be reviewed.
Turn green, yellow and red into action thresholds
- Green: 80–100, with current evidence and no critical override. Continue the success plan, verify the next customer outcome and review health at the normal account cadence. Green does not mean the account no longer needs attention.
- Yellow: 60–79, or a material adverse trend. Assign the CSM to validate the concern with the customer, name a recovery action and set a review date. As an illustrative operating target, review the plan within five business days.
- Red: below 60, or a confirmed critical risk. Assign an accountable recovery owner, involve the appropriate leader and agree on the next customer update. An illustrative target is to establish the plan within one business day; an active critical incident requires immediate incident response.
These bands need calibration. A strong average must not mask a critical unresolved blocker. Keep the reason for the color visible: “Adoption has stalled because the integration is incomplete; services owns the dependency review on Thursday” is more useful than “health is 58.”
Common mistakes to avoid
Do not equate logins with value, let a renewal date dominate the model or make subjective CSM sentiment the only explanation. Avoid adding signals simply because the data is available. Too many weak inputs make a score harder to maintain without making it more useful.
Another mistake is tuning the score until it agrees with the team's existing opinions. Review both missed risks and false alarms. If a yellow alert repeatedly produces no useful action, change its definition or remove it. Keep a record of model changes so shifts in color are not mistaken for changes in customer behavior.
Implementation checklist
- Choose one segment and one decision the score should improve.
- Agree on customer outcomes, signal definitions, data owners and freshness rules.
- Write rating criteria, initial weights, confidence rules and critical overrides.
- Review historical account examples, including healthy customers and missed risks.
- Pilot with a small group of CSMs and compare explanations, not just totals.
- Attach an owner, action and review date to every meaningful yellow or red condition.
- Review false alarms, missed risks and completed interventions before expanding the model.
For the broader operating context, see Douglas Prehodka's Customer Success approach. A health score should support that work: earlier risk detection, clearer priorities and customer outcomes that a team can explain.
Discuss your Customer Success priorities
Connect with Douglas about Customer Success leadership, team development and customer relationships.