Updated July 2026. This began as an answer to a reader question in 2023. The classification system at the center of it still stands. What is new is a second layer that sits on top of it, added at the end.
The question was simple. I know I should be doing some kind of churn analysis, but I have no idea where to start.
People stall on this because churn analysis sounds like it should be rigorous, and rigorous sounds like a project. So it gets scoped, then deferred, then never done. The analysis you never run is worth less than the rough one you run this week.
Churn Is Not the Problem
Churn is a symptom. It is the visible result of something that happened earlier, usually a customer who did not get where they were trying to go. Treating the symptom is why so much retention work feels like effort without movement.
The point of analysis is never to reduce a number. It is to find the thing upstream producing the number.
Low Churn Is Not a Reason to Skip This
The most common objection is that churn is too small to bother with. So walk through what one churned customer actually cost.
You spent money acquiring them. You spent your team's time serving them, which is time that did not go to customers who would have stayed. If they left before paying back their acquisition cost, that account was net negative for its entire life. If they left unhappy, they may have said so publicly, and now your sales and marketing spend has to overcome that too. And if your addressable market is finite, that logo is out of circulation for years.
None of that appears in a churn rate. All of it is real.
The Two Questions That Sort Everything
Every churned customer answers two independent questions, and the pair is the whole system.
Was it expected or unexpected? Did you see it coming. This one is not about the customer at all. It is a measurement of you: whether you were engaged, whether your signals work, whether anyone was close enough to the account to know. A customer who leaves and surprises you has told you something about your operation, separate from anything they told you about your product.
Was it avoidable or unavoidable? Could you have changed the outcome. Avoidable means something you control produced the loss, usually service delivery. Fix that and they stay. Unavoidable means the cause sat genuinely outside your reach.
Cross them and you get four boxes, and the two corners are the ones worth naming.
Unexpected and avoidable is the worst churn there is. You could have kept them and you never saw it coming. That is two failures stacked: the delivery problem, and the blindness to it.
Expected and unavoidable is the best churn available. The customer went out of business, and they told you it was coming. It still costs you everything above. It is still churn. But it is the one square where the answer is genuinely not your fault.
Unavoidable is a much smaller bucket than almost anyone wants it to be. It is the box people reach for when they would rather not look at the other three.
Be suspicious of your own sorting here. In practice you have some natural attrition, and it is less than you think.
Updated for 2026: What the Loss Tells You
The quadrant answers whether you could have stopped it and whether you saw it coming. Both are operational questions, and you need them.
What the quadrant does not answer is what a given loss means for everything you do next. That is a second layer, and it is the three-category doctrine I published in 2026. It maps onto the avoidable axis rather than replacing it.
Delivery-failure churn is avoidable. They did not get what they bought. Beyond the loss itself, this one carries a consequence the quadrant cannot show you: it precludes expansion entirely. You cannot sell more to a base that is not getting what it already paid for.
Natural attrition is unavoidable. This is the genuine floor, and it is the bucket that gets over-filled. Benchmark it against your own market rather than someone else's published average.
Outgrew-you churn is avoidable, and almost everyone files it as unavoidable. This is the important correction. A customer who succeeded past what you offer is not a customer you could not have kept. Either the next thing did not exist, or it existed and they never knew. Both are yours. By the time you notice it may well be too little too late, but too late is not the same as unavoidable, and calling it unavoidable is how the lesson gets thrown away.
That last one also happens to be the most valuable churn you have, because it is a customer telling you exactly what they wanted to buy next, right before they bought it somewhere else.
Start Smaller Than You Think
Take your last twenty departures. Put each in a box, then tag each with a category. That is the whole first pass.
Twenty is enough because you are not looking for statistical significance, you are looking for agreement. If twelve of twenty land in the same corner for the same reason, you have found something real, and no larger sample makes it more true.
Do not build the elaborate version. The elaborate version is how this stays undone.
The Analysis Nobody Runs
One addition, having done this with a lot of companies since.
Everyone runs some form of exit analysis. Almost nobody runs the opposite one. Nobody asks the customers who stayed why they stayed.
That data is easier to collect, and it describes a population still paying you. What comes back is usually a description of value those customers are not currently being sold more of.
Churn analysis tells you what to stop losing. The other conversation tells you what to go get.
