What Is a Customer Health Score and How Do You Calculate It?

Ask an account manager "how is the Smith account doing?" and you will get a paragraph: they used the product a lot last month, one complaint is open, the renewal is in March, and the main contact has gone quiet recently. Useful, but impossible to compare with the other eighty accounts on the same list. A customer health score compresses that paragraph into a single number from 0 to 100 — a score that can be sorted, trended, and turned into rules like "alert me when anything drops below 60".

What a customer health score actually is

A health score is a weighted composite of the signals that predict whether a customer stays, grows, or leaves. It is a leading indicator: it moves before the cancellation does, which is exactly why it is useful. Done well, it answers three questions at a glance: which accounts are thriving, which need attention, and which are at risk of leaving before you next talk to them.

What goes into the score

The ingredients vary by business model, but most useful scores mix some combination of these five signal families:

• Engagement and usage. Logins, active users, core-feature adoption, and workflow frequency. The strongest predictor of retention for most products — a customer who stopped using the product has already left in every way that matters except the paperwork.
• Satisfaction and sentiment. Survey scores (CSAT, NPS) plus the tone of what customers say in feedback, reviews and support conversations. Sentiment catches the shift between surveys.
• Support experience. Open ticket count, age of the oldest ticket, repeat contacts about the same issue. Unresolved friction is a churn accelerator, and it is one of the few signals you control directly.
• Commercial relationship. Payment history, plan level, expansion behaviour, and time to renewal. A customer in an expansion conversation is a customer who is staying.
• Recency. All of the above matter more when they are fresh. A six-month-old satisfaction score should not carry the same weight as last week's behaviour.

How to calculate it: a worked example

There is no universal formula, but the standard method is simple: pick the signals, score each one from 0 to 100, assign weights that sum to 100%, and multiply. Here is a concrete example using five signals and an imaginary account called Acme:

• Engagement: 80 out of 100, weighted at 30% → contributes 24.0
• Satisfaction & sentiment: 70, weighted at 30% → contributes 21.0
• Support experience: 90, weighted at 20% → contributes 18.0
• Commercial relationship: 85, weighted at 15% → contributes 12.75
• Account tenure/stability: 60, weighted at 5% → contributes 3.0

How to read the number: bands and actions

A single number is only useful if it triggers a response. Most teams use simple bands:

• 0–39: At risk. The account is exhibiting multiple churn indicators. Assign an owner, schedule a conversation this week, and diagnose before pitching.
• 40–69: Needs attention. One or two signals are slipping. Agree a fix for the specific friction and a follow-up date; this is where most preventable churn lives.
• 70–84: Healthy. Keep the loop closed and watch for drift. A declining trend inside this band is more important than the level.
• 85–100: Thriving. Candidates for expansion, advocacy, testimonials and referrals — ask for them while the goodwill is current.

The trend matters more than the level. An account at 82 that fell from 94 in two months is a warning; an account at 60 that has climbed steadily since onboarding is a success story in progress. Score on a fixed cadence so both are visible.

Five pitfalls that quietly break health scores

• Stale data. A score built on last quarter's numbers is astrology. Refresh on a schedule short enough to act on.
• Double-counting. If satisfaction feeds both the sentiment and the support components, one bad survey unfairly drags the score twice.
• Invisible weights. If nobody knows what the score is made of, nobody trusts it — or knows which lever to pull when it drops. Publish the recipe.
• All lag, no lead. A score made only of past behaviour (churned, didn't churn) tells you what happened, not what will.
• No owner, no action. A score with no assigned response is decoration. Every band should have a named next step.

You can start a health score in a spreadsheet this afternoon, and you should — the exercise of choosing signals and weights is valuable on its own. But a manual score goes stale the moment you stop updating it, which is why ClientTell keeps it live: every customer gets a running 0–100 score from satisfaction, engagement, sentiment and support signals, with the recipe visible and alerts when anyone starts to slip — free to start, and you can try the same weighted calculation in our health score calculator right now.