In this post:
A customer health score is a single rating, usually 0 to 100 or a red, amber and green label, that combines product usage, engagement, support history, billing and relationship signals to predict whether a customer will renew, grow or leave.
Customer success teams use it to decide where their hours go. Nobody can read every account every week, so the score sorts the base for them: these forty are fine, these six need a call this week, and these three are ready to talk about buying more.
Most of the health scores we get asked to review fail in the same direction. They stay green right up to the cancellation email, because they were built from whatever data was easy to pull (survey answers, ticket counts, the date of the last business review) instead of from what the customers who left actually had in common. A score like that does real damage, because it tells the team it can stop looking.
TL;DR
A health score blends four or five groups of signals into one rating: product usage, engagement with your team, support and billing friction, and how many people at the account you actually know. We give usage the most weight, because it shows what customers do rather than what they say.
Measure depth of use, not logins. Pick the one or two actions that mean the customer is getting value, track active seats against seats bought, and watch each account against its own baseline.
Show the parts next to the total. Two accounts can both score 72 for opposite reasons, so every score needs its components and the signal that moved it most.
HubSpot builds health scores from point-based rules in the Service Hub customer success workspace (Professional and Enterprise, with a Service seat), and Gainsight builds weighted scorecards from measures and measure groups. Both leave the choice of signals and weights to you.
Before anyone relies on the score, rebuild it for last year's renewals and cancellations as they looked about a quarter before the renewal date. If the accounts that left don't sit clearly lower, the weights are wrong.
Give every band an owner and a next step: red gets a rescue plan, amber a check-in, green a look at expansion. Patterns in who stays green also tell you which customers to go after next.
The signals behind a customer health score, and where each one lives
A health score summarizes a handful of signal groups, each pulled from a different system and each moving at a different speed. Where the data sits matters as much as what you measure, because a signal that has to be copied in by hand is out of date by the time anyone reads it.
Signal group | What to look at | Where it usually lives | How fast it moves |
|---|---|---|---|
Product usage | Active users against seats bought, how often the core action happens, which features are in use | Product analytics or your own database, synced to the CRM | Days |
Engagement | Replies to the account manager, meetings held, who turns up to reviews | CRM activity, calendar, email | Weeks |
Support | Ticket volume, severity, reopened tickets, the tone of recent conversations | Help desk | Days |
Billing | Failed payments, overdue invoices, downgrade requests | Billing system or accounting software | Each invoice |
Relationship | A named champion, the number of engaged contacts, an executive sponsor, job changes among them | CRM contacts plus job-change alerts | Slowly, then all at once |
Owner sentiment | The account manager's own read of the account | A picklist on the company record | Whenever someone remembers to update it |
No single row predicts churn on its own. A customer can use the product every day and still leave because their champion moved to another company and the replacement prefers a competitor. That's why a contact leaving is worth tracking as a trigger event in its own right, and why the relationship row deserves more attention than it usually gets. A sales champion who changes jobs takes a lot of your account's health with them.
The sentiment row needs care. An experienced account manager often knows something the data doesn't, but a manual field goes stale fast and drifts toward optimism, so give it a modest weight and show when it was last changed.
Every row also inherits the quality of the records underneath it. Duplicate company records split one customer's usage across two scores, and contacts who left a year ago make the relationship look healthier than it is. Basic data hygiene comes before any weighting debate.
Usage is the signal customers can't talk their way around
A customer can tell you on a renewal call that everything is going well. If half their seats haven't been used in six weeks, the usage data is the more reliable witness. Usage also tends to fall before anyone at the customer has consciously decided to leave, which makes it the earliest warning most teams can get.
The trap is measuring the wrong kind of usage. Logins are a weak proxy, because a browser tab left open all day counts as a login. What you want is the action that means the customer is getting what they paid for: invoices sent in an invoicing tool, meetings booked in a scheduling tool, reports viewed by someone outside the team that built them. Pick one or two of those per product and count them.
Seat use is the other number worth watching. A customer using 40 of the 100 seats they bought, eight months into a twelve-month contract, is heading for a renewal conversation about cutting seats, however friendly the calls feel. The reverse matters too: an account bumping against its seat limit is a growth conversation waiting to happen.
Finally, compare each account with itself. A customer with steady, moderate use is usually safer than one that has dropped from heavy use to slightly above average, even though the second still looks better in absolute terms. Scoring the change over the last 30 or 60 days catches drift that a flat threshold misses.
Getting usage into the CRM is the same plumbing that powers a product qualified lead setup. Product events are aggregated per account, synced on a schedule and written to company properties that the score can read.
Two accounts at 72, for opposite reasons
Here's why the total alone isn't enough. Picture two customers on the same plan, both scoring 72 this week.
The first has excellent usage. Their team runs the core workflow every day and support tickets are rare. But their champion left the company a month ago, nobody has replied to the last two emails from the account manager, and the renewal is in eleven weeks. That 72 is falling, and the problem is relational.
The second dipped in usage over the summer and has a few open tickets about an integration. But three people there are actively engaged, the head of department joined the last review, and they just asked about adding another team. That 72 is noise around a healthy account.
One needs an executive call this week. The other needs the integration fixed and probably a conversation about expansion. If the account manager only sees "72, amber", both get the same generic check-in email. Show the component scores next to the total, plus the signal that moved it most in the last month. That one line turns a number into an instruction.
How HubSpot and Gainsight build a health score
Most teams configure their score in a CRM or customer success platform rather than building it from scratch, and the two most common setups work quite differently.
In HubSpot, health scores live in the customer success workspace in Service Hub Professional and Enterprise. According to HubSpot's documentation on customizing a health score, you build the score from groups of events or groups of properties, and each rule either adds points for healthy behavior or subtracts points to flag risk. HubSpot offers ready-made groups for one-to-one engagement, call engagement, support tickets, CSM sentiment and NPS, and you set the thresholds and names for three status labels yourself. A score can apply to up to five company or contact segments, and running several separate scores is an Enterprise feature. Teams that have already built lead scoring in HubSpot will recognize the points logic.
One practical detail: HubSpot's article on the health score card on records says a Service seat is required to see it. The card shows the score, the status and recent changes, and the score history lists the events that moved it. If account executives own renewals but don't have Service seats, check what they can actually see before you promise them a health score.
Gainsight takes a weighted approach. Its scorecard documentation describes measures, each tracking one health metric, collected into measure groups. Each measure gets a percentage contribution inside its group, each group gets a percentage of the overall score, and neither set can add up to more than 100%. A measure without a percentage is left out of the calculation. Scorecards can mix automated measures, updated by rules, with manual ones that an account manager fills in.
Neither tool decides which signals matter for your customers. When a scorecard doesn't reflect real health, Gainsight's scorecard FAQ advises adjusting the measures, thresholds and weights and checking that both automated and manual measures are current. That's sound, but which way to adjust is a question about your own churn history.
Test a customer health score on customers who already left
A health score is a claim about the future, and the cheapest way to check it is against the past. Before anyone builds a process around the score, run it backwards.
Pull every account that came up for renewal in the last twelve months, and split them into renewed, downgraded and churned.
Rebuild each account's score as it would have looked roughly 90 days before its renewal date, using only the data available at that point.
Compare the groups. The accounts that churned or downgraded should sit clearly below the ones that renewed. If the ranges overlap heavily, find the signals that did separate them and move the weight there.
Look at the false alarms too. Red accounts that renewed without fuss tell you which signals are scaring people for no reason.
Step two is where most teams get stuck, because nobody saved last year's usage per account. If that's you, start a monthly snapshot of every input now, and in a year you'll have the history to test properly.
Small customer bases need a lighter touch. With fifteen churned accounts a year you can't fit weights statistically, so read every churned account by hand, note what was visible a quarter earlier, and let that shape the weights. Start with three to five signals you trust rather than fifteen you can't explain. Statistical models, the same approach as predictive lead scoring, only start paying off once you have a few hundred renewals and cancellations to learn from.
Repeat the check after each renewal cycle. Products, pricing and the customer base all shift, and a model that separated churners well two years ago can drift without anyone noticing.
What each color should set in motion
A score that changes color and triggers nothing is decoration. Each band needs an owner, a deadline and a default action, written down before the first account turns red.
Red. The account owner gets a task the same day, with the components that drove the drop. The default response is a rescue plan: find out what broke, bring in someone senior on both sides if the relationship has gone cold, and fix the usage problem before the renewal date gets close. Nobody pitches an upgrade to a red account.
Amber. A check-in within a week or two, aimed at whatever moved. Often that's a refresher session for new users, an introduction to the person who replaced the champion or a nudge on an open ticket.
Green. Watch for signs that the account is outgrowing its plan: seats near the limit, a new department asking for access, usage close to a tier ceiling. A thriving account behaves much like a product qualified lead for expansion revenue, and it's where a land and expand strategy gets its second and third deals.
The alerts should be automatic. A workflow that opens a task when a score crosses a threshold is built from the same parts as automated prospecting, just pointed at customers instead of strangers. Nobody should have to remember to open a dashboard.
The score is the early read, and your retention metrics are the verdict that arrives later. Your customer churn rate counts the logos that left, revenue churn and gross revenue retention count the money that walked out, and net revenue retention shows whether expansion outran those losses. A health score is how you influence those numbers, and the customer lifetime value behind them, while there's still time. Healthy customers who grow also tend to recommend you, which feeds the GTM flywheel.
Health scores point back at who you sold to
After a year or two of scoring, look at which customers stay green. Usually they share a profile (a company size, an industry, a reason for buying), and the ones that keep sliding into red often share one too, visible before the contract was signed.
That makes the health score a feedback loop on targeting. If customers outside your ideal customer profile keep ending up red however hard the team works, the fix sits in sales, not in customer success.
The three scores in a revenue team share the same logic and point at different stages. Lead scoring ranks people you haven't won yet, account scoring ranks the companies worth pursuing, and the health score watches the customers you already have. When the health data flows back into the fit criteria behind the first two, every new customer starts closer to the green end.
Wrapping up
A customer health score turns a lot of scattered evidence into a short list of accounts that need a person this week. The hard part isn't the formula. It's choosing signals that reflect real value, pulling them in automatically, showing why a score moved and proving that the score separates the customers who left from the ones who stayed.
Get those right and the score changes how a customer success team spends its time. Rescue work starts months before a renewal instead of days, expansion conversations go to accounts that are ready for them, and the patterns in who stays healthy sharpen who sales goes after next. Of all the moving parts in a revenue setup that spans sales, marketing and customer success, this retention loop gets the least attention and often pays back fastest.
If you'd like a health score your team will actually act on, we build the scoring, the usage sync and the alerts on top of your existing CRM and product data, then leave behind the history test and a note on every weight, so the model can keep improving after we've stepped away.
Frequently asked questions
How is a customer health score different from a lead score?
They point at opposite ends of the customer lifecycle. A lead score estimates how likely a prospect is to become a customer and decides who sales should call first. A health score estimates whether an existing customer will renew, grow or leave and decides where customer success should spend its time. Both combine several signals into one ranking, but they use different data and answer different questions.
What is the difference between account scoring and a customer health score?
Both look at the whole company rather than one contact, but at different stages. Account scoring ranks target companies you haven't won yet, using fit and intent, so sales knows where to aim. A health score ranks companies that already pay you, using usage, engagement, support and billing, so the team knows who to protect and who to grow. Many teams run both, and the health data improves the account score over time.
How does a customer health score relate to customer lifetime value?
Healthy customers tend to stay longer and buy more, and those are the two things that drive lifetime value. So a health score works partly as an early forecast of value: a high score suggests a long, growing relationship, a falling one suggests a shorter, smaller one. The link also runs the other way, because accounts with high lifetime value justify more effort when their score starts to slip.
Why do some customer health scores give a false sense of security?
Usually because they were built from easy data instead of predictive data. Survey scores, ticket counts and meeting dates are simple to pull but often say little about who leaves. Other causes are weights that were never checked against real churn, a model nobody has revisited in years and manual fields that drift toward optimism. The result is a base that looks mostly green until the renewals come in.
How far ahead can a health score warn you about churn?
That depends on your contract length and on which signals you track. Usage and relationship signals often start moving months before a customer announces they're leaving, while billing problems may only show at the last invoice. A practical aim is to flag at-risk accounts at least a full quarter before renewal, because that's roughly how long it takes to fix an adoption problem or rebuild a relationship.
Should NPS be part of a customer health score?
It can be, with a small weight. NPS captures how one person felt on the day they answered, and response rates in B2B are often low, so a single detractor or a missing answer can swing the score. It works best as a supporting signal next to usage and engagement. A drop in NPS from someone who signs the renewal deserves a call regardless of the overall score.
Can you build a health score without product usage data?
Yes, but it will be weaker. Services firms, on-premise software and hardware vendors often lack usage feeds. In that case lean on what you can measure: how often the customer engages, how many contacts are involved, support and billing patterns, project milestones met and the owner's assessment. Even a rough proxy, such as reports downloaded or orders placed, is usually better than leaving usage out entirely.
Is a red, amber and green label better than a score from 0 to 100?
For day-to-day work, the colored label is easier to act on, because each color can map to a clear response. The underlying number is still useful for spotting movement inside a band, such as an account sliding from 74 to 61 while staying amber. Most tools store the number and display the label, which gives you both. Keep the number of bands small so each one means something.
Who should own the customer health score?
Customer success usually owns the score and acts on it, but RevOps or operations should own the model itself: the data feeds, the weights and the history check. Splitting it that way stops the people being measured by the score from tuning it to look better. Sales and finance should be able to see it, because renewals, expansion and forecasts all depend on the same accounts.
Can machine learning predict churn better than a rules-based health score?
It can, once you have enough history. A model needs a few hundred renewals and cancellations, with the signals recorded at the time, to learn which patterns matter. Below that it tends to find patterns that aren't real. Many teams start with a transparent rules-based score, collect clean history for a year or two, and then test a predictive model against it.
Share this post
RevOps
CRM
Glossary
Related glossary items
Understanding the core GTM concepts
Ready to build your pipeline
