Customer health scoring is a way of summarising how a customer relationship is going in one number or colour, built from signals such as product usage, support contact, feedback and billing. Customer success teams, account managers, agencies and membership organisations use it to decide which accounts need attention this week, before a quiet problem turns into a cancellation. This page explains what goes into a health score, how one is calculated with a worked example, how it differs from NPS and from sentiment scoring, what it looks like outside software companies, and which tools teams use to run one.
What a Customer Health Score Is
A customer health score is a composite: several signals, each scored and weighted, added up into one value per account. Most teams express it as a number from 0 to 100 or as a traffic-light tier of green, amber and red.
The score answers one question: is this account getting enough value to stay? It does not answer why. A falling score is a prompt to go and find out, not a diagnosis.
Two things it is not:
- Not a satisfaction score. A customer can give you a 9 on NPS and still be drifting, because the person who answered the survey is not the person who stopped using the product.
- Not a prediction. A low score says an account looks like accounts that left before. Many cancellations are preceded by observable signals, but not every one is, and a green account can still leave for reasons no score sees, such as a budget cut or an acquisition.
What Goes Into a Health Score
Inputs fall into four groups. Useful scores usually draw on at least three of them, because each group misleads in its own way when used alone.
Usage signals
What the customer actually does:
- Login frequency and time since the last active session
- Adoption of the features that deliver the core value
- Completion of key workflows or milestones
- Number of active users on the account, for multi-seat products
Relationship signals
The human side of the account:
- Response rate to emails and check-ins
- Attendance at review calls, training or events
- Whether your main contact is still there, and still engaged
Feedback signals
What the customer says:
- NPS, CSAT or CES responses and their trend
- The tone and recurring themes of open-text feedback and support tickets
- Support ticket volume, and whether the same problem keeps coming back
Commercial signals
The financial side:
- Days until renewal
- Late or failed payments
- Expansion or downgrade history
Usage alone tells you who logs in, not who is getting value. Feedback alone over-weights the customers who talk to you. Commercial signals arrive late. Combining them is the whole point of a health score.
How a Score Is Calculated: A Worked Example
Every health score does three things: it scores each signal against a band, weights the signals, and maps the total to a tier. The numbers below illustrate the method. They are not reference values, and copying them imports somebody else's conclusions about your customers.
1. Score each signal against a band. A three-level rubric is enough to start, and it works in a spreadsheet:
| Signal | Healthy (2) | Neutral (1) | At risk (0) |
|---|---|---|---|
| Login frequency | Several times a week | Weekly | Less than weekly |
| Feature adoption | Five or more core features in use | Two to four | One |
| Latest NPS response | 9 or 10 | 7 or 8 | 6 or below |
| Support tickets this month | None or one | Two or three | Four or more |
| Last feedback submitted | Within 30 days | 31 to 60 days | Over 60 days |
2. Weight the signals. A weight says how much each signal counts toward the total:
| Signal | Weight |
|---|---|
| Login frequency | 25% |
| Feature adoption | 25% |
| Latest NPS response | 20% |
| Support tickets this month | 15% |
| Last feedback submitted | 15% |
3. Add it up and map it to a tier. Take an account that logs in weekly (1 of 2), uses three core features (1 of 2), last scored 9 on NPS (2 of 2), opened four tickets this month (0 of 2) and sent feedback 20 days ago (2 of 2). Its score is 25 × 0.5 + 25 × 0.5 + 20 × 1 + 15 × 0 + 15 × 1, which is 60 out of 100.
With tiers of 80 to 100 for healthy, 50 to 79 for at risk and 0 to 49 for critical, this account is amber. The NPS score alone would have called it a promoter. The ticket count is what pulled it down, and that is exactly the conversation worth having.
The bands and weights are the part to calibrate. "Several times a week" is healthy for a product people work in every day and meaningless for one they open monthly. Score last year's renewed and cancelled accounts retrospectively, then put the red line where it would have caught a useful share of the departures without producing a list nobody works through.
How to Build a Health Score Without a Data Team
A first version fits in a spreadsheet and takes an afternoon. These six steps turn the worked example above into a score for your own accounts.
Step 1: Choose four to six signals
Start with data you already have, and pick at least one signal from three of the four groups above. Ask what your departed accounts had in common before they left, and look for the signals that would have shown it. Do not wait for a perfect data model; a rough score you review every week beats a precise one you never build.
Step 2: Set a band for each signal
Decide what healthy, neutral and at risk look like for each signal, as in the rubric above. The bands have to reflect how your customers normally behave: a weekly login is excellent for a monthly reporting tool and a warning sign for a daily workspace.
Step 3: Weight the signals
Give the signals that best separated renewals from cancellations in your history the largest weights. If you have no history yet, weight them equally and adjust after the first review.
Step 4: Set tiers and calibrate them
Map the total to green, amber and red, then test the boundaries against past accounts. If most of last year's cancellations would have scored green, the bands or weights are wrong, not the customers.
Step 5: Give every tier an action and an owner
Decide what happens when an account turns amber or red, who does it, and how quickly. An alert nobody owns is a report, not a system.
Step 6: Review after 60 to 90 days
Compare the scores with what actually happened. Did red accounts leave? Did anyone leave from green? Adjust the signals, bands and weights, and repeat each quarter as your customer base changes.
Examples in Different Kinds of Organisations
The mechanics are the same everywhere. The inputs change.
- A marketing agency scores each retainer client on email response rate, attendance at the monthly strategy call, on-time approval of deliverables and the last satisfaction rating. The score decides which clients get a proactive call that week, so the renewal meeting is not the first sign of trouble.
- A B2B software product scores each workspace on active users, tasks created in the last 30 days and quarterly CSAT. A low score triggers an offer of a product walkthrough, because low health there is usually an adoption problem before it becomes a relationship problem.
- A professional association or non-profit scores members on event registrations, content downloads and forum activity. Members below the threshold get a personal re-engagement message instead of the general newsletter.
- An online school scores learners on course progress, quiz participation and forum posts, so tutors can reach out before a student quietly drops out.
Health Scoring vs Sentiment Scoring
A health score is a composite. Sentiment scoring is one possible ingredient in it, and often a minor one.
| Customer health scoring | Sentiment scoring | |
|---|---|---|
| Inputs | Usage, feature adoption, support volume, billing history, tenure, contacts engaged, and sometimes sentiment | Text: tickets, feedback, survey comments |
| Unit | An account | A message, a user, or an account's messages over time |
| Updates | On a schedule, from several systems | When someone writes something |
| Fails when | Weights are set by intuition and never checked against actual renewals | Tone is read literally, or too little text carries too much weight |
The mistake worth naming is a health score that is mostly sentiment. Tone is the noisiest input available and the easiest to collect, so it tends to dominate scores by accident. An account can be quietly, politely on its way out, and a sentiment-heavy score will call it healthy right up to the cancellation.
Common Mistakes With Health Scores
Building the score from whatever data is easy to get
Login counts are easy to pull, so they end up carrying most of the weight, and the score ends up tracking activity rather than renewal. Start from the question "what did our departed accounts have in common?" and look for data to answer it, not the other way round.
Never checking the score against what happened
A health score is only as good as its validation. If nobody has checked whether last quarter's red accounts actually left, and whether the ones that left were red, the team has a dashboard, not a model.
Using too many inputs
Fifteen signals make a score that nobody can explain and nobody can act on. Four to six well-chosen signals beat a long list.
Leaving out what customers say
A customer who logs in every day but keeps sending frustrated feedback is not healthy. Purely quantitative scores miss the reason behind the numbers.
No owner for a falling score
A score only has value when a drop triggers a response. Decide in advance who acts on an amber or red account, and how quickly, or the system becomes a report nobody reads.
Tools for Customer Health Scoring
Dedicated customer success platforms calculate the score for you. Other tools supply one layer of input.
| Tool | Best for | What it contributes |
|---|---|---|
| Gainsight | Enterprise customer success teams | Health scoring built into customer success workflows |
| ChurnZero | Mid-market customer success teams | Health scores, alerts and playbooks |
| Totango | Customer success teams working by segment | Segment-based health scoring |
| HubSpot | Small businesses and agencies already using its CRM | Account health tracked alongside CRM records |
| Mixpanel | Product teams | Usage and engagement data to feed a score |
| FlagUp | Teams collecting feedback from clients or users | Feedback themes and sentiment to feed a score |
No single tool covers every input. A common setup is a product analytics tool for usage, a feedback tool for what customers say, and a CRM for relationship and billing data, with the score itself calculated in a customer success platform or a spreadsheet.
The feedback layer is usually the weakest, because open text is harder to collect consistently than login counts. FlagUp collects client feedback and requests in one place, and on paid plans groups duplicates so a problem is counted once and scores the sentiment of incoming feedback, which you can feed into your own model as one input among several. Its risk view is built from feedback sentiment and inactivity only, so it is an input to a health score rather than a health score, and it does not predict cancellations. See what each plan includes.
Frequently Asked Questions
What is a customer health score?
A customer health score is a composite metric that combines usage, relationship, feedback and commercial signals into a single number or tier describing how engaged and stable an account is at a given point in time.
How is a customer health score different from NPS?
NPS measures how likely one person is to recommend you. A health score is broader: it includes what the account actually does, such as logins and feature adoption, alongside survey answers. NPS can be one input to a health score, but on its own it misses accounts where the person answering is not the person using the product.
Does a health score predict churn?
No. It shows which accounts currently resemble accounts that left in the past. That is useful for deciding where to spend time, but not every departure is preceded by a visible signal, and a score calibrated on last year's customers can drift as your customer base changes.
Can small businesses use customer health scoring?
Yes. An agency or small team can run a useful score in a spreadsheet with three or four signals per client. The principle scales down easily, and the discipline of reviewing it weekly matters more than the tooling.
How often should health scores be updated?
Weekly suits most teams. Daily makes sense with many accounts and automated data feeds. Monthly is usually too slow to leave time to act before a renewal.
What score should trigger outreach?
There is no standard number. The threshold depends entirely on your own bands and weights. Score past renewed and cancelled accounts with your model, see where the cancelled ones clustered, and set the alert just above that point.
FlagUp helps teams collect client feedback in one place, spot the problems each account keeps raising, and show clients what happened to their requests. Start free or compare plans.
Related articles
- What is Churn Prevention? Definition, Examples, and Tools
- What is Churn Signal Analysis? Definition, Examples, and Tools
- What is User Sentiment Scoring? Definition, Examples, and Tools
- NPS vs CSAT vs CES: Picking the Right Metric for Your Stage
- The Feedback Metric That Reveals What Users Won't Say Directly