A customer health score you can build this quarter.

You can build a customer health score this quarter. Skip the vendor template with its red-amber-green light. Build a real one from the accounts you already renewed and the ones you already lost.

Most scores are built the wrong way round. Someone picks a few inputs that feel important and weights them by instinct, then acts surprised when the green accounts churn and the red ones renew. Build it backward instead, from what already happened.

Start with the accounts you already lost

Pull two years of renewals and two years of churns. Line them up side by side.

Look at the months before each decision, because the month it happened tells you least. The customer who left in month eleven usually stopped using the thing around month three, so the warning was there long before anyone worried. What you are hunting for is what separated the two groups back when the room was still calm. Forget the stated reason each account gave for leaving, and look at what the leavers had in common that the people who stayed did not.

The four inputs that usually survive

Across the accounts I have looked at, four behaviours tend to do most of the separating. None of them is exciting, but each one moves before the renewal does, which is why it belongs in the score.

Adoption depth means real active users, spread across the team. Seats sold do not count, and neither does one power user carrying the whole account on their back, since breadth is what makes you hard to rip out.

Time to first value is how long before they hit a real outcome, which is not the same as their first login. When that first genuine win comes fast, the account tends to stay stuck to you for years.

The champion matters too. Is the person who fought to buy you still in the building? When the champion leaves, a clock starts, and whoever inherited the account never wanted it in the first place.

Last, the support pattern. Ticket counts matter less than the shape of the contact behind them. A flood of tickets is an engaged customer working hard to succeed with you, and the account that used to talk every week and has gone quiet is the one that should scare you.

Keep the first version embarrassingly simple

The instinct is to build something clever - fifteen inputs and a machine-learning layer someone read about on a flight. Resist it.

Three or four inputs you trust will out-predict fifteen you half-measure. You can compute the whole thing in one CRM property or a spreadsheet this month, and elegance is beside the point. What you want is a number that moves before the customer does, and that a real person will look at on a Tuesday.

Weight by what separated them, not by what feels important

Do not weight by gut. Look at which of the four best split your own renewals from your own churns, and weight that one heaviest.

If you have no time to model it, equal weights still beat a feeling, so start crude. A rough score you act on beats a precise one you are still designing when the renewal arrives. Tune it every quarter as more accounts close and you learn which inputs were carrying the prediction.

Wire it to act before the renewal, not after

A score that only lives on a dashboard changes nothing, so the real work is the trigger.

A score that drops below your line should create a task, with an owner and a date, in month three while you can still do something about it. Do that, instead of finding it in a review three weeks before the renewal, when you are only narrating the loss. That gap is why retention is decided far earlier than the renewal date.

Give that trigger a real owner and a real threshold. Assign it to the whole customer success team and it belongs to nobody, so pick one person and one number that means act, and give the task one place to land. The account manager should hear that an account is slipping from the system, well before the customer's cancellation email does the telling.

A score built on stale data lies

Your score is only as honest as the fields feeding it. If the champion field points at a contact who left eight months ago, the score glows green while the account dies.

And data goes stale on its own. HubSpot's own data puts B2B contact decay at about 22.5% a year. One CRM audit I ran found 722 custom properties, 203 of them completely empty. Feed a health score from a CRM in that state and all you have automated is a guess dressed up as a predictor. Clean the inputs first, because trustworthy data is the thing every downstream number depends on.

So the fix is unglamorous: get the underlying fields right before you lean on the score at all.

Check whether the score was right

Every quarter, look back. Did the accounts it flagged red really struggle, and did the green ones renew. If the green accounts keep leaving, an input is lying or a weight is wrong, and the fix is to go back to the accounts you lost rather than bolt on three more inputs.

Most health scores only turn an account red the week before it leaves, by which point you are reading an obituary.

Common questions