Customer churn: how to count it and why growth hits a ceiling

Churn is the share of customers who stopped paying during a period. It's simple to count:

monthly churn = customers lost in the month ÷ paying customers at the start of the month

You had 200 customers and 16 left during the month — churn is 8%.

The number is simple, but it implies something many people learn too late.

The leaky bucket

Imagine 20 new customers arrive every month and 10% of those you have leave. With 50 customers, five leave — the base grows. With 150, fifteen leave — it grows more slowly. With 200, twenty leave — exactly as many as arrived. Growth has stopped.

customer ceiling = monthly inflow ÷ monthly churn

20 ÷ 0.1 = 200. The base won't rise above that, however many months pass. To double the ceiling you need either twice the inflow — roughly twice the ad budget — or half the churn.

That's why a company with high churn looks alive: new customers keep coming, sales go on, the team is busy. But the paying-customers chart has already flattened out, and every next dollar of advertising only holds it in place.

Weeks and months aren't the same thing

If churn is counted per week, it's easy to underestimate. 2% a week sounds modest, but it's about 8% a month and over 60% a year. Compare numbers in the same units.

Why overall churn lies — cohorts

Overall churn mixes old and new customers, and they behave differently. Usually the most leave in the first weeks: a person tried it, didn't figure it out, or realized the product isn't for them. Whoever survives the first two months stays much longer.

So you look by cohort: take everyone who started paying in the same week or month and track how many of them are left after 4, 8, 12 weeks. That gives two important answers:

Do cohorts level off. If each group's curve flattens over time, the product has a core of people who need it. If it slides to zero, it doesn't.
Are new cohorts better than old ones. If fresh groups hold better after product changes, the changes work. Overall churn will show this months later.

Where churn comes from

There are usually several causes, and each needs a different cure:

The wrong customer. The product solves a problem that barely hurts these people. The cure is choosing a segment, not a discount.
A raw product. Outages, bugs, slowness. The cure is quality, and sometimes a pause on new features.
A price at the ceiling. The customer pays, but at the edge of what they're willing to give — and leaves at the first excuse.
Somewhere to go. A competitor has appeared, and your customer now compares.
Discount customers. People who came for a promotion leave more easily than those who came for the product. A price war doesn't bring the same customers cheaper, it brings different ones.

The main mistake is treating the wrong cause. For example, investing in quality when the customer is actually stumbling on their first sign-in to the product.

How it works in Runway 200

In the game, churn isn't set as a number — it comes from the same causes: how well you fit the segment, product quality, price relative to what the segment is willing to pay, competitors and outages. Every week of new customers is a separate cohort: its stickiness depends on how it arrived (on a discount, through a weak channel, or on the advice of a happy customer), and old cohorts leave less often than young ones.

To get out of the first phase you need fifteen paying customers, with seven in ten surviving to week eight. You can see that number exactly only with cohort analytics — without it, the screen shows a noisy estimate. More about the instruments in Unit economics in plain words.

At the end of a run, the review breaks churn down into causes and names the turning-point week. Here's a sample from world 1007:

Cash hit zero in week 94. The main thing that broke: customers were leaving almost as fast as they arrived.

Up to week 52, paying customers grew by 1.4 a week; after that, by 0.2. That's the leaky bucket: inflow didn't drop, churn caught up with it. And the truth of this world was that the product responded to onboarding — the customer's first sign-in — while the player invested in quality.

A world number can be replayed. Start world 1007 for free and see whether you last longer than 94 weeks.