Churn is a decision made long before the cancel button

By the time a customer clicks "cancel," nothing is left to save. The relationship didn't end that day — it ended weeks or months earlier, quietly, in a series of small signals nobody was watching: the usage that tapered off, the champion who stopped replying, the support ticket that closed unhappily, the renewal date that arrived with no conversation behind it. The cancellation is just the paperwork on a decision that was already made. The whole point of a customer health score is to see that decision forming while there's still time to change it, instead of discovering it the day the revenue walks out the door.

This is the difference between retention you run on purpose and retention you hope for. Without a health score, success work is reactive: you find out a customer is unhappy when they tell you, and by then "telling you" usually means giving notice. A health score makes the relationship legible — it turns the scattered, easy-to-miss signs of a customer drifting away into a single number that gets a person's attention before the drift becomes a departure.

A health score is leading indicators, not a satisfaction survey

The most common mistake is to build a health score out of the wrong inputs. A satisfaction survey, an NPS number, the warm feeling from the last call — these are lagging or unreliable. The customer who rates you a nine and churns three months later isn't a paradox; they answered a survey about how they felt that day, not whether they were getting value. A health score worth trusting is built from leading indicators — the behaviors that reliably precede a renewal or a cancellation, observable in the data you already have.

The strongest inputs cluster into a few categories, and the art is choosing the handful that actually predict churn for your product rather than measuring everything:

  • Product usage and its trend. Not just whether they log in, but whether usage is steady, growing, or fading — and especially whether they've adopted the features that deliver the core value. A customer using one-tenth of what they pay for is a cancellation waiting for a budget review. The trend matters more than the level: a heavy user trending down is in more danger than a light user holding steady.
  • Relationship depth. A single champion is a single point of failure — when they leave, the account often leaves with them. An account with multiple engaged contacts is far healthier than one resting on one person's goodwill, which is why broadening the relationship is its own retention move.
  • Engagement with you. Are they answering emails, showing up to check-ins, opening what you send? A customer going quiet is rarely a happy customer who's just busy; silence is one of the most reliable leading indicators there is.
  • Support and friction signals. A rising count of tickets, an unresolved escalation, a bug that hit them hard — friction that isn't cleanly resolved erodes the relationship even when nobody complains out loud.
  • Commercial signals. Whether they've expanded or contracted, whether the renewal is approaching with momentum or silence, whether they downgraded a tier. These are the loudest indicators because they're the customer voting with money.

The point isn't to track all of these — it's to pick the four or five that, looking back at customers who actually churned, were flashing red before they left.

Weight the inputs, and let the worst signal dominate

Once you have inputs, the temptation is to average them into a tidy number. Averaging is exactly wrong, because it lets a strong signal hide a fatal one. A customer with great usage, deep relationships, and a champion who just gave notice is not "mostly healthy" — they're in serious danger, and an average would paint them green. The math should let the worst signal pull hard, because in churn the worst signal usually wins.

A workable approach is to weight each input by how predictive it actually is — usage trend and a departed champion deserve more weight than an open rate — and to set hard tripwires that override the score regardless of everything else. A renewal in sixty days with no conversation booked should flip an account to at-risk no matter how good its usage looks, because the commercial clock doesn't care how engaged they were last quarter. The score's job isn't to be mathematically elegant; it's to surface the accounts a human should look at today, and a tripwire on the one signal that historically precedes loss does that better than any weighted average.

This is the same logic that makes a lead score work on the way in: a score is only useful if it reliably separates the cases that need action from the ones that don't, and it earns that reliability by being built and tuned against what actually happened, not against what feels right.

The score is worthless unless it triggers an action

A health score that lives in a dashboard nobody refreshes is a vanity metric. The number has value only at the moment it changes someone's behavior — when a green account turning yellow puts a specific task in front of a specific person while there's still runway to fix what's wrong. A score that merely describes health is a postmortem in slow motion; a score that triggers a save is the whole reason to build one.

That means every meaningful threshold needs an owner and a play. An account dropping to at-risk shouldn't generate a notification that decays in a feed — it should create a task for the person who owns the relationship, with enough context to act: what dropped, when, and what the likely cause is. The response itself draws on the rest of the success playbook — a real check-in, a fresh round of value delivered the way onboarding first delivered it, a business review that re-anchors the relationship to outcomes, or, if the account has already half-decided to leave, the honest save conversation that's a cousin of winning a churned customer back — except here you're catching them before they go. The score points; the play saves.

Health scores protect expansion, not just retention

It's worth noticing the score cuts both ways. The same signals that warn you about churn also surface the opposite: an account whose usage is climbing, who's adding contacts, who's leaning in is telling you it's ready for more. A health score isn't only a churn alarm; it's also the cleanest way to spot the expansion opportunities hiding in your book, which is where the lifetime value of a small team's customer base actually compounds. The green-and-climbing accounts deserve a play too — just a different one. Reading the score only for danger leaves half its value on the table.

Make the score live where the relationship does

A health score maintained by hand in a spreadsheet is a score that's already out of date — the usage data is stale, the renewal dates drift, and the whole thing decays into a quarterly guess. To predict churn the score has to be wired to the live signals of the account, and it has to push, not wait to be checked.

In Hitt CRM, the inputs already live on the customer record: the activity that tracks whether they're engaging, the renewal and expansion data on the deal, the support and check-in history on the timeline. A scoring model can turn those leading indicators into a single health number that updates as the signals do, and an automation can do the only thing that matters — turn an account slipping to at-risk into a task in front of the right owner the moment the score crosses the line, instead of a red cell nobody sees until the renewal call that never gets booked. Reports then turn the whole book into a map: which accounts are healthy, which are quietly drifting, and where the next save — or the next expansion — is hiding before either becomes obvious.

The one-sentence version

A customer health score is worth building only if it does the one thing satisfaction surveys can't — read the leading indicators of a relationship going wrong (usage and its trend, relationship depth, engagement, friction, and commercial signals) early enough to act, weighting them so the single worst signal can't hide behind a comfortable average, and then triggering a real save play on the right person's task list the moment an account slips — because churn is a decision made weeks before the cancel button, and the only version of the score that matters is the one that catches that decision while you can still change it.