March, 2026
We know churn but do we understand it We learn the concept of customer churn the same way we learn gravity. We take an observation and give it a name. Things fall to the ground and that’s gravity. Customers leave and that’s churn. We end up knowing both names but understanding neither. Churn can come from a good experience A bad experience is not churn. In fact, a good experience can be churn. This seems absurd but lets take an example. In my neighbourhood there was a popular food spot whose owner decided to upscale the place, add glass front and ACs. But the popularity tanked immediately. It turned out that the customers of that spot wanted cheap food that was good (almost all of them were university students). And fancy seating being correlated with higher prices led to them churning without even the prices having been raised. Why customers say "sayonara" Churn is when a user runs into something that is the opposite of what they wanted - and then never come back. Higher prices, bugs, reliability and speed are only sometimes reasons for users leaving - but only when these were what the users wanted from the product. But different users want different things from a product. Even the same user may want different things from the same product. Some people want payment wallets because their bank app is slow. Some people want wallets because they want to separate the money that they can spend from the money they already have allocated for other things. Some people want wallets because they want to keep their main account from being exposed to risk as they make purchases online. The person that uses a wallet app for its speed will leave if it starts showing unnecessary pop-up screens while the one that uses it as a risk limiting system might not leave when the pop-ups start showing up. Different people leave for different reasons because they want different things. When you cancel a subscription, there is sometimes a form asking for the reason. That form is trying to figure out this reason (usually unsuccessfully). What this means is you have different sets of customers who get different values from your product. And one benefit of churn data is that you can use it to tell you what different values users get from your product for as you make changes. If your UX gets worse and users don't leave it means users don't come to your app for the UX. The flip side of this is that if you don't use churn data to learn what values users come to your product for, you are likely to increase your churn without realising it. Churn rides an interaction-cycle If customers want a specific things from a product, changing those thing for the worse should immediately churn customers. But it doesn't. Customers leaving because of a feature depends on customers interacting with that feature. And customers interact with different features in different cycles. Some features they interact with every time they use the product and with some once a month. If the interaction cycle of what you change is low enough that only 1% users interact with it in a month, you won't be able to see the change in total churn from that even if the churn from that change is as high as 20%. Because if your average churn is 10% per month, this new change would only move it by 0.2% One example is the re-login flow for your app. A re-login flow getting worse does not show up as high enough churn in the short term because very few users face it every week. Losing customers from an important flow just because it's interaction frequency is low is a dumb way to lose customers. One second order effect of this cyclic churn is that the cycles for features can change and lead to unexplained increase in churn. If all users of your app get logged out for some security reason and have to log back in, the problem in the flow that was only causing 0.2% in total churn would now cause 20% of total churn as 100% of your users go through that experience. To avoid losing users from dozens of small leaks and making the product worse, you should track churn across feature changes by tracking what ratio of users going through an experience churned before you made the change and what ratio of users going through an experience churned after you made the change instead of just tracking total churn across all users when you make changes. Blast radius for churn Not all churn is the same. Two things can have a different blast radius for churn even if both are causing only 5% total churn. Let's take an example, if 5% of your churn is coming from people who are trying to buy screw drivers from your e-commerce website, the blast radius of this churn is limited to people who want to buy a screw driver online and there is no risk for a general user. But lets say a change was made in the return policy for your e-commerce store that is causing 5% users to churn. This has a nuclear blast radius because every customer who orders enough products will end up going through the return procedure and every single customer of yours faces a risk of churning A high blast radius will apply churn to bigger set of users while a small blast radius will apply churn only to a very specific and small set of users. The only problem is that they both seem same in the short term. And so your job is to be able to separate small blast radius churn reasons from high blast radius churn reasons and focus on the latter. Learn stats for churn and not GPA You can figure out a great chunk of what your customers don't want from you. But you need to be able to see the data you already have and predict how it will move and then create small experiments to test your predictions. You cannot do any of this if you don't understand some of the basics of statistics and probability. You need to understand Statistical Significance to tell if the change you're seeing is actually meaningful, you need to understand Bayesian Probability to double check an assumption that you created from looking at the data and you need to understand how to create Controlled Experiments to check what customers say they want don't actually cause an increase in churn instead of reducing it. One more thing The last reason for a user leaving is not necessarily 100% of the reasons the user is leaving for. There might be a number of reasons that when combined make the user leave. But churn usually only captures the last reason for the user leaving - sometimes even not that. What we need to do is to record not just customer churn for the product as a whole but customer churn across individual values of the product. If the product has multiple values and features, keeping track of when customers stop using specific features will help you catch, predict and reduce churn. If you can figure out what the straws are stacking up on the customer's back, you might be able to remove a few before the last straw breaks the customer.
If you want to be notified of new articles, this is my substack where I send out unscheduled emails containing new essays and things that I am thinking or reading.