churn

Churn is the leak. You started the month with a certain number of paying subscribers; by the end, some of them are no longer paying, and churn is that loss measured against where you started. It is the mirror of retention, and the two get used for different jobs: churn is the figure you report, retention is the thing you work on. What makes churn awkward is that it is a single percentage covering several unrelated kinds of departure, and it moves for reasons that have nothing to do with how anyone performed.

How is churn calculated?

Subscribers lost during the period, divided by subscribers present at the start of it.

The numerator is rarely disputed. The denominator is where the number is actually made, and there are two defensible conventions:

• Starting base only. Anyone who subscribed mid-period is excluded from both halves of the fraction. This is the version that compares one month to the next.
• Starting base plus arrivals. Everyone who was subscribed at any point counts. This describes the month as it was lived.

Pick one, write it down, and never quietly switch. Most dramatic churn movements are a definition change wearing a costume.

What does the raw number hide?

Who left, and whether you can still reach them. Four very different events land in the same percentage:

Departure	What happened	Still reachable
Auto-renew switched off	Nobody wrote for weeks	Yes, an expired fan
Deliberate cancellation	Something went wrong in the thread	Sometimes
Payment declined	No decision was made at all	Yes, usually at once
Account deleted or blocked	The person left the platform	No

A month built of declined cards and a month built of deletions read identically in the metric and mean opposite things. The first is an operational problem you fix this week; the second is an audience you no longer have.

Why does it mislead on a growing account?

Because new subscribers leave faster than long-standing ones. That holds almost everywhere subscriptions exist, and the consequence catches people out: a month of heavy recruitment mechanically pushes churn up even when nothing about the work got worse. The account is not deteriorating, its mix has shifted toward the fragile end.

The reverse trap is worse. An account that stops recruiting watches its churn improve for months while its revenue falls, because everyone fragile has already gone and only the durable are left to count.

The way out is not a better formula. It is cohort analysis, which measures each arrival month against itself, so a change in churn means a change in behaviour and not a change in intake.

What do you actually do with it?

Read it beside lifetime value (losing fans who never bought is not the same problem as losing buyers) and break it down by creator before anything else. A single agency-wide churn figure averages accounts that have nothing in common.

Then work the half of the number that is still reachable. A lapsed subscriber with an intact thread is the cheapest revenue in the account, and a win-back is not a sales message. Our data shows conversion collapsing between 2am and 6am and peaking in the evening, so even the hour you send it at is a real decision rather than a queue to clear.