Most churn signals arrive too late to do anything about. A customer cancels, and only afterward does someone notice their order frequency had been dropping for weeks. The data was there. Nobody was watching the right comparison at the right time.

Part of the problem is that LTV and retention numbers usually get calculated as one blended average across the whole customer base, pulled from a store, an email platform, and an ad platform separately, then stitched together by hand whenever someone needs a current view. By the time that view exists, it's already a snapshot of the past, not a warning about what's happening now.

The average hides the customer who's already leaving

A company-wide average is the wrong comparison for catching an individual customer's drift. A customer who normally orders every month going quiet for six weeks is a strong signal. A customer who normally orders twice a year doing the same thing is normal. Both look identical on a blended dashboard. Only a comparison against that specific customer's own pattern catches the first one in time to matter.

What an actual cohort review looks at

Someone who's good at this doesn't start with one company-wide number. They segment by acquisition channel before calculating anything, because a blended average can hide one channel quietly losing money under a healthier one. They flag "at-risk" against a customer's own historical cadence, not a fixed calendar rule. They calculate LTV to CAC per cohort, not company-wide, and treat anything under roughly 3x as worth a look. They measure retention on a rolling window segmented by what a customer bought first, because that predicts loyalty better than signup date does. And they only look at what's changed since the last review — not the entire dataset every time, because a report that repeats yesterday's numbers gets ignored by the second week.

That's the judgment Agencize learns as a playbook — captured from watching how an analyst actually reviews cohorts and decides what counts as a real risk, not from a standard LTV formula. Once learned, it stays connected to the data sources directly and surfaces only what moved: a customer whose order frequency just dropped against their own average, a cohort whose LTV to CAC quietly slipped below the floor. The dashboard doesn't show everyone, every time. It shows what changed, with the reasoning attached, early enough to actually do something about it.

See how a playbook gets learned, or see this exact use case running as an Instant App.