Customer Health Score (2026): How to Build One, a Weighting Example, and Why CRM-Only Scores Miss Churn

TL;DR

• A customer health score is a weighted composite of signals, usually product usage, support activity, engagement, feedback and renewal status, that estimates whether an account will renew, expand or churn. It is typically shown as 0 to 100 or red, yellow and green.
• There is no standard formula. Pick four to six signals that predicted renewals and churn in your own history, weight them, set bands, and check the score against what actually happened every quarter.
• Scores built only from CRM fields miss the earliest warnings: escalations rising in tickets and email, a champion who stopped joining calls, a renewal conversation that went quiet. The CRM can show green while the account is leaving.
• Sentra joins CRM, support and billing records with what customers and your team say in email, chat and meetings, and raises the account that is drifting to its owner with the evidence.

What a customer health score is

A customer health score is a single metric that combines several signals about an account into an estimate of how likely it is to renew, expand or churn. HubSpot defines it as a data-driven metric that helps customer success teams assess that likelihood, combining behavioral, engagement and outcome signals into one view of risk and opportunity.

Lorikeet adds the two cautions that matter most. There is no standard formula, so each company builds its own, weighting inputs by how strongly they correlate with renewal in its historical data. And the main risk is false precision: a score of 73 versus 75 means little unless the model has been validated against real churn.

Health scores are used for three jobs: deciding which accounts a customer success manager (CSM) works on this week, forecasting renewals, and spotting expansion. A score that does the first job well is worth more than one that is elegant on a dashboard.

Customer health score signals

HubSpot groups the usual inputs into six categories. The table adds where each signal lives and how early it moves.

Leading signals move weeks or months before a renewal decision. Lagging signals confirm what already happened. A score weighted toward lagging signals tells you an account was unhappy after it has decided.

How to build a customer health score

1. Define the outcome. Decide what "healthy" predicts: renewal at full value, renewal with expansion, or simply no churn. Pull the last 8 to 12 quarters of renewals and churns as your test set.

2. Pick four to six signals. Choose the ones that separated renewed accounts from churned accounts in that history. Start small. Lorikeet recommends three to five signals you know matter, validated against actual churn.

3. Score each signal 0 to 100. Define what earns 100, 50 and 0 for each, in writing. For example, active users as a share of licensed seats: 80% or more scores 100, 50% scores 50, under 20% scores 0.

4. Weight the signals. Weights add to 100%. Weight by predictive strength, not by how easy the data is to get.

5. Set bands. A common banding is green 75 to 100, yellow 50 to 74 and red 0 to 49. GitLab's published handbook uses exactly these bands for its Gainsight scores.

6. Handle missing data. Decide what happens when a signal is blank. GitLab redistributes a missing measure's weight to the others, and notes most measures are empty for a customer's first 30 days.

7. Add a CSM sentiment input. A human judgment, scored and dated, catches what the data cannot. Require a one-line reason for any red.

8. Validate every quarter. Compare last quarter's scores with renewals and churns. If churned accounts were green 90 days before, the weights are wrong.

Customer health score weighting example

A real, published example: GitLab's handbook weights its account health as product (license activation, user engagement and use case) 50%, risk (CSM sentiment and renewal risk) 25%, outcomes (success plan and verified outcomes) 10%, voice of the customer (support and surveys) 5%, and engagement (customer engagement, executive sponsorship, events and certifications) 10%.

Here is a worked example for a B2B software company, with illustrative weights.

Account A looks healthy. The next section shows why it might not be.

Customer health score template

Use one row per account. Update the inputs weekly and the CSM sentiment at least monthly.

The trend and the top risk with its source are the two fields most templates leave out, and they are the two a leader reads first.

Why CRM-only health scores miss risk

Most health scores are calculated from fields: seats, logins, ticket counts, renewal stage, last activity date. Those fields capture that something happened, not what was said. The early signs of churn are usually said before they are logged.

Four patterns show up again and again.

• Escalations that never become tickets. A customer's VP of Operations emails your CEO directly about a recurring failure. The help desk shows two routine tickets. The support score stays green.
• The renewal goes quiet. Last quarter, the champion replied within a day. For six weeks, three emails have gone unanswered and the renewal call was moved twice. The CRM still shows the opportunity at "Commit" with a recent activity date, because your team kept logging outreach.
• A champion leaves. The customer's sponsor mentions on a call that she is moving to a new role. Usage stays high for a month on habit. The CRM contact record is unchanged.
• Competitor evaluation. A user mentions in a shared channel that their team is trialing another product. No field captures it.

In each case the CRM is accurate. It records activity. The risk lives in the content of email, tickets, chat and meetings, and it is spread across systems that the score does not read together. The score is green because each source, on its own, looks fine.

Worked example: a top account escalating while the CRM shows it healthy

Assumptions, illustrative only. Account A is a top-ten customer with $420,000 in annual recurring revenue and a renewal date in five months. Its health score is 85.5, as in the table above.

Over the last 45 days:

• The help desk shows 6 tickets, up from 2 a month. None is tagged as an escalation, so the support score drops only from 90 to 80.
• Two of those tickets reference the same integration failure, and the customer's operations lead wrote in one of them that this is "the third time this quarter."
• The customer's VP of Operations emailed the account executive asking for a call about "whether this is working for us."
• On the last QBR, the customer's champion said she would be handing the relationship to a new director.
• The renewal opportunity is at "Commit" in the CRM. Three follow-ups on pricing have had no reply.

Every one of these facts exists inside the company. None of them changes the score much, because the score reads ticket counts and CRM stages, not what the tickets, emails and meetings say. Read together, they describe an account at risk of not renewing: repeated failures, executive doubt, a champion leaving and a quiet renewal. The right band is red, and the right owner is the CSM lead with an executive sponsor, five months before renewal rather than five weeks. A focused QBR and a root cause analysis on the integration failure are the next steps.

Common mistakes

• Too many signals. Fifteen inputs average each other out. Start with four to six.
• Weighting by data availability. Logins are easy to count, so they get weight. Executive engagement is hard, so it gets ignored.
• Never validating. If last year's churned accounts were green 90 days before they left, the model is decoration.
• No trend. A score of 72 that was 88 last month is more urgent than a steady 60.
• No reason attached. A red with no explanation and no source gets ignored.
• Reading only the CRM. Tickets, email and meeting notes hold the earliest signs.

How Sentra catches this automatically

Sentra understands how a company really runs, and acts on it before anyone asks. It reads email, chat, meetings and documents alongside ERP, CRM, HRIS and BI data and other tools connected through its 200+ integrations, and works out what is true now and why.

For customer health, Sentra reads what the score cannot: the content of the tickets, the VP's email to your account executive, the champion's comment on the last QBR call, and the renewal thread that stopped getting replies. It joins them with the CRM record and the usage data for the same account. Because Sentra knows when a fact became true and when it stopped being true, it knows the champion named in the CRM is no longer the person who will decide the renewal, and that the renewal stage has not been confirmed by the customer in six weeks.

When those signals point the other way from the score, Sentra raises the account to its owner with the evidence: the escalation emails, the repeated ticket theme, the meeting where the champion change came up, and the dates. A named person approves each finding before it reaches anyone else, and before Sentra acts or sends anything. Your CRM and customer success platform keep the score and the workflow. Sentra supplies the context they cannot see.

Permissions follow each source, so a CSM sees only what they could already see. Sentra holds SOC 2 Type II and ISO 27001 and runs in the cloud, in an isolated VPC or air-gapped on premises. Through REST and MCP, the same account memory is available to ChatGPT, Claude, Copilot, Gemini or your own agents. Related reading: Sentra for customer support teams and contract compliance for commitments made to the customer that the record does not show.

To see Sentra find the at-risk accounts your health score shows as green, book a demo.

How to start

Pull your last four quarters of churned and renewed accounts. Score each one as it would have looked 90 days before the renewal date. If the churned accounts were mostly green, read their emails, tickets and meeting notes from that window and write down what was said. Those statements are the signals your score is missing. Add a CSM sentiment input with a required reason, add a trend field, and review the top 20 accounts by revenue every week with their top risk and its source.

FAQ