Churn is the share of your subscribers you lose in a period. That sounds too simple to get wrong, and yet I rarely audit a program where the number on the dashboard means what the team thinks it means.

The definition

Monthly churn is the number of subscribers you lost during the month, divided by the number who were active on the first day of the month. Count only losses from that starting group. Someone who subscribed on the 10th and canceled on the 25th belongs in the survival table for this month’s starts, not in this month’s churn rate.

Then split it. Subscribers leave in two ways, and the fixes have nothing in common:

• Voluntary: the subscriber chose to cancel.
• Involuntary: a payment failed and every retry after it failed too, so the subscription ended without anyone deciding anything.

Peloton’s filings are a good model of a metric defined with care. It reports average net monthly churn for its connected fitness subscriptions: the quarter’s churn events, minus subscriptions that came back from a pause, minus reactivations, divided by the average number of subscribers at the start of each month, divided by three. A pause counts as churn on the day it starts Filed. Its recent quarters show why a single quarter misleads: 1.9% for the quarter ending December 2025, 1.2% for March 2026 and 2.2% for June 2026, with the summer quarter highest Filed. Whatever definition you choose, write it down, count a pause as a loss until the subscriber comes back, and compare the same month year over year before you panic about a seasonal one.

Three impostors

1. Cancels against starts“We started 400 and lost 250 this month.” That ratio tells you whether the program grew. It says nothing about how fast you lose people, because the two numbers come from different groups. A program can have a healthy ratio and terrible churn, if it’s buying lots of new starts to replace the people leaving.
2. Churn on the wrong baseDividing losses by the end-of-month count, or by a base that includes this month’s new starts, makes churn look lower in a growing program. The faster you acquire, the better it looks, which is backwards.
3. Half the lossesSome reports count only cancellations; others count only subscriptions that stopped billing. Either way you see one leak and miss the other.

From my workThe first of these is the most common and the most expensive. I’ve seen a same-month ratio of cancels to starts presented as a churn rate, and a save campaign sized against it. It was the wrong instrument measuring the wrong thing. Split the losses by cause and by the month each subscriber started, and the right fix is usually obvious.

How much is failed payments?

More than most teams guess. In Recurly’s benchmarks, updated with July 2026 data from its network of subscription businesses, failed payments made up about a third of ecommerce subscription churn: 1.38 points of 4.25 Reported. Recurly doesn’t make clear whether those are monthly or annual rates; the share is what matters. The share was about 30% for subscriptions averaging $10 to $25 per customer, and about 6% for those over $250 Derived. The cheaper the subscription, the bigger the share of its losses that come from declined cards. Stripe has said that 25% of lapsed subscriptions are “purely due to payment failures” Reported. Both are billing vendors reporting on their own networks, and neither publishes its sample, so treat the range as a guide and measure your own.

You’ll also see “up to 48%” and “53%” quoted for involuntary churn. I couldn’t trace the first to any data at all, and the second contradicts the numbers in the same article that states it. Don’t use either.

Churn isn’t constant

The average hides the most important fact about subscription churn: it’s heavily front-loaded. Recharge, which runs subscriptions for more than 20,000 Shopify brands, published its active churn rate by renewal for renewals from July 2025 to March 2026, counting only subscribers who canceled Reported:

ReportedRecharge, “Subscription churn is front-loaded,” 2026. Vendor data; subscriber-initiated cancellations only, so failed payments add to these.

Two consequences. First, an average monthly churn rate is a blend of new subscribers leaving fast and old subscribers leaving slowly, so it moves with your mix. A program that just had a big month of new starts will show rising churn next month even if nothing got worse. From my workI pin a note to the scorecard of any program coming off a big launch month: aggregate churn will spike as that group reaches its first renewals, so judge the save and payment fixes on their own numbers, not on the total. Second, the place to fight is the first two renewals, which is chapter 9.

An average churn rate is a blend of new subscribers leaving fast and old ones leaving slowly. Report it by start month.

The survival table

The fix for the blend is a table with one row for each start month and columns for the share still active after renewals one, two, three, six and twelve. Read across a row to see how one group decays. Read down a column to see whether newer groups survive better or worse than older ones. Leave a cell blank when the group isn’t old enough to fill it. The query is in Appendix A.

Do this

• Recompute the last three months with this definition, split into voluntary and failed-payment losses. Then build the first version of the survival table, even if it’s by hand. Put both next to the churn figure your billing app shows and write down why they differ.