So most businesses find out a customer left when the customer tells them. The cancellation email comes in, or the renewal doesn't, and then somebody scrambles to call, and now you're trying to win back a person who already made up their mind. And the thing that bothers me about that is the signal was sitting in your own records for weeks. Nobody was looking at it.

Where it shows up first

A customer who's about to leave almost never leaves suddenly. They go quiet first, which is different, and quiet has a shape you can measure.

There's the order gap. Somebody who's bought every three weeks for a year skips one. Their usual interval was 21 days and now it's day 35. Nothing happened, no complaint, no email, they just didn't order. That's the earliest signal you're going to get, and it's sitting in your invoices.

There's the failed payment nobody retried. A card declined on the first, the system sent one automated email, and that was it. Ten days later the customer is technically past due and technically nobody has talked to them. Often this isn't churn at all, it's an expired card, and it becomes churn because it got ignored.

There's the support thread with no reply from your side. A customer asked a question on Tuesday, somebody meant to get back to them, and now it's Friday. The customer didn't follow up, and people read silence as an answer.

And there's the renewal thirty days out with no contact. If a contract renews next month and nobody from your business has talked to that customer since the last renewal, that's not a relationship, that's a subscription that hasn't been cancelled yet.

The rules

None of this needs a model to predict anything. It needs four rules, and I'd write them down exactly like this.

For each customer, compute their usual interval between orders or jobs or payments. If the current gap is more than twice their usual, flag them.

Any payment that failed and hasn't been retried or replaced within three days, flag it.

Any inbound message from a customer with no reply from your side in 48 hours, flag it.

Any renewal inside 30 days with no outbound contact in the last 60, flag it.

That's it. Four rules, and they'll catch most of the customers who are about to leave, weeks before they leave. The problem has never been that the rules are hard. It's that the data for them lives in four systems, the invoices, the payment processor, the inbox and the CRM, and nobody is going to check four systems every morning looking for a pattern.

What you do with the list

The point of catching it early is that early is cheap. A customer whose gap just doubled gets a text asking how things are going, not a discount. A failed payment gets a friendly note with the update link. An unanswered thread gets answered. A renewal gets a call. None of that is a win-back campaign. It's just paying attention at the moment it's still easy to.

And it's the opposite of what happens when you find out late, which is that you're now trying to follow up with everybody who cancelled this quarter, all at once, after the fact, and most of them aren't picking up.

How this runs

This is one of the things Liaison does on its overnight pass. It reads the invoices, the processor, the inbox and the CRM, runs those four rules across every customer, and in the morning the quiet ones are on your list with the reason next to each name: gap doubled, payment failed on the 3rd and not retried, thread unanswered since Tuesday. Each one has a draft note ready in your voice. You look at the three names, you send the two that make sense, and you call the third.

We wrote about the five numbers I'd want on Monday morning, and this is the fifth one expanded, because it's the one no system reports on its own. It's also the one where being proactive instead of reactive pays off most obviously. The customer is still yours on day 35. They're not on day 90.

If you'd like to know who on your list went quiet this month, bring your invoices and your inbox to a call. It's usually three names, and one of them usually surprises people.