Churn

There was a period when growth meant the new-signup number and nothing else. A report goes up saying 500 people joined this month. Another goes up next month saying 500 more joined. Read that alone and the company grows by 500 people every month.

Then somebody counts total users half a year later and finds the figure parked right around 3,000. As many left as arrived. The name for that state is a leaky bucket: open the tap as wide as you like and the water still won't rise if the bottom has a hole. With the hole out of view the response is always to open the tap wider by spending more on ads, which only raises what each arriving user costs.

Churn is the hole turned into a number. Counting who stayed and counting who left describe the same event, and both words survive because they answer different questions. The staying side shows which kind of person the product worked for. The leaving side shows how much drained out this month and what it would take to plug it.

Put the customer count at the start of the period in the denominator and the count that left during it in the numerator, then divide. In order, four steps.

1. Set the period. Products billed monthly use a month; products sold on annual contracts use a year. Measure annual-contract customers monthly and the number spikes in whichever month the contract ends and reads zero for the other eleven.
2. Set the denominator. Count only customers present when the period opened. Customers who arrived during it don't belong there — include them and the denominator swells in every high-signup month, making churn look lower than it is.
3. Define what leaving means. For a paid product, decide whether the date is when cancellation was requested or when access actually ended; those two are usually about a month apart. A free product has no cancel action at all, so it needs its own line, such as no visit for 90 days.
4. Divide. Start the month with 1,000 accounts, lose 30, and monthly churn is 3 percent.

Those same 30 accounts produce different numbers depending on what you count. Counting customer accounts is logo churn; counting the money that walked out is revenue churn. If the 1,000 accounts paid $80,000 a month and the 30 that left paid $12,000 of it, the count says 3 percent and the money says 15 percent. A handful of large customers leaving produces exactly that gap. The lost recurring portion has its own name, churned monthly recurring revenue (MRR).

The reason for leaving gets split too. A customer who decided to go is voluntary churn; a subscription cut off because a card expired or hit its limit is involuntary churn. Leave the two merged and there's no way to tell whether the product needs fixing or the billing system does.

Comparisons against other companies come first on the list of things to be careful with. An $8 per month personal plan and a $150,000 annual enterprise contract are not equally easy to leave. The first ends with one button; the second carries internal approvals and a data migration, so customers keep paying for a year after they've mentally moved on. That's why even one product splits the number by plan.

Reading it means stretching the period out. Five percent monthly churn sounds small, but held for twelve months it removes 46 percent of the customers who started the year. Two percent monthly leaves 78 percent standing. A gap of three percentage points becomes a 30 percent difference in customer count within a year.

Next comes the comparison against new arrivals. Fifty accounts in and thirty out is a net gain of twenty. Forty in and forty-five out means the company shrinks no matter how the marketing numbers looked. The bucket from earlier becomes visible in exactly this comparison.

The direction over time matters more than the absolute value. One month on its own is noisy, though: at 200 customers, each additional departure moves the rate half a percentage point. So teams read a three-month average, or group customers by when they joined and compare how many remain at the same month mark.

"Logo churn was 3 percent this month but revenue churn was 15 percent. We lost a big one."

Few accounts left, but one of them carried a large share of the billing.

"A third of our churn is failed payments. This isn't a product problem."

Fix the billing path before going looking for causes in the product.

"Churn is bigger than new business. This quarter we work on keeping, not acquiring."

Total customers shrink even as acquisition holds, so the priority has to change.

This came up at Pagedeck about a year after paid plans launched. New teams signed up every month, yet the count of paying teams sat near 400 and wouldn't move. Ravi wanted to raise the ad budget; Mia wanted to count the teams that had left first.

Monthly churn turned out to be 6 percent. Sofia contacted the twenty-four teams that had gone and found that half of them had been cut off by an expired payment card, with nobody aware it had happened. Switching to three failure notices spaced three days apart brought the next month's churn to 4 percent. Not one additional dollar went to advertising.

The most common mistake is treating churn as a number to drive to zero. A lost customer is a loss, so eliminating losses sounds like the goal. But every product takes in customers it was never going to fit. Someone who signed up to work alone has no reason to stay on a collaboration tool for long. Chasing those people costs discounts and feature work, and they leave the following year anyway. The question isn't zero, it's which customers are leaving. Poor-fit accounts departing is the system working; well-engaged accounts departing is the signal.

Looking for the cause only inside the product is the other one. "They left" reads as "the product fell short." But one of the two branches above, the subscriptions cut off by failed payments, has nothing to do with the product. That branch reaching a third of total churn is not unusual. Without the split, a team can spend two quarters on feature work and watch the number hold steady, when what it needed was a reminder email a few days before cards expire.