Picture a grocery delivery app called FreshKart (invented for this article). Its growth head, Meenakshi, opens the Monday report and sees “monthly churn: 4%”. The retention lead, Arjun, runs his own query and gets 19%. Both are looking at the same customer table. Both are correct. They have simply never agreed what it means for a customer to have left.

Two kinds of business, two kinds of leaving

A telecom postpaid plan, a SaaS licence or an OTT subscription is contractual. There is an event: the customer cancels, the plan lapses, the mandate stops. Churn is that date. You still have edge cases (paused plans, failed payments that recover), but the definition is mostly handed to you.

A grocery app, a cab app or an online fashion store is non-contractual. Nobody tells you they are leaving. They stop opening the app, and the only evidence is an absence. So churn there is an inference: “we have not seen this person for long enough that we will treat them as gone.” Everything depends on “long enough”.

Arjun’s 19% and Meenakshi’s 4%

Meenakshi’s team counts a customer as churned if they have not ordered for 120 days. Arjun counts anyone silent for 21 days. For a customer who typically orders every week, 21 days is three missed cycles. For a customer who orders once a month for festival stock-ups, 21 days is just a normal gap, and calling them churned would be wrong.

Neither number is dishonest. The 120-day rule is too slow to be useful: by the time someone is flagged, the chance of winning them back has mostly gone. The 21-day rule fires so early that it flags people who are about to order anyway, so a retention offer sent to them is just a discount handed to a customer who would have bought regardless.

Letting the customers pick the threshold

The better way is to look at how long real customers go quiet before they come back. Take every gap between two consecutive orders in the last year and see how those gaps are distributed. Say, with illustrative figures, that 70% of gaps are under 14 days, 90% are under 35 days, and 97% are under 60 days. If you set the churn line at 35 days, then in about one case in ten you will call someone churned and they return anyway. At 60 days that falls to about three in a hundred, but you learn about real departures nearly a month later.

That is the trade-off in one sentence: a short threshold is quick and wrong more often, a long threshold is accurate and late. There is no correct value, only one that suits what you will do with it. A win-back push notification can tolerate false alarms, so a shorter line works. A board-level retention figure wants stability, so a longer one is safer.

Two refinements are worth knowing. First, make the line relative to the person rather than the same for all: flag someone when they are silent for, say, three times their own usual gap. A weekly shopper and a monthly shopper then get fair treatment. Second, consider a “reactivated” status. People who return after being flagged are not new customers, and counting them as new inflates acquisition numbers while hiding that your churn line was too tight.

What a churn rate does once you trust it

Suppose FreshKart settles on a definition that gives 12% monthly churn among active customers. Two things follow from simple arithmetic, which I checked: the average customer lifetime is roughly 1 divided by 0.12, about 8.3 months, and of any group of customers only 0.88 raised to the twelfth power, about 21.6%, are still active after a year. That second number surprises people. A 12% monthly leak sounds mild, and it still empties four-fifths of the bucket in a year.

These shortcuts assume the same rate every month, which is rarely true. New customers churn much faster than long-standing ones, so a blended rate mixes two very different groups. A cohort view, where you follow people who joined in the same month, is the fairer picture. The lifetime figure also feeds into customer lifetime value, so a loosely defined churn flows straight into how much you decide you can spend to acquire someone.

Writing it down

Back at FreshKart, the fix was not statistical. They wrote one paragraph: churn means no order for 45 days, measured at month end, among customers who ordered at least once in the prior 90 days, with reactivations tagged separately. That last condition matters. Without a denominator rule, people who signed up and never ordered pull the percentage around in ways nobody notices.

Put the definition, the threshold and its reasoning next to the metric wherever it is displayed. When someone later asks why churn “improved” by three points, the first thing to check is whether the line moved, not whether customers did.

Figures and names in the examples are illustrative.

Leave a Reply

Trending