Your Churn Spike Is an Illusion

Most of the churn spikes I get called about never happened.

For years now the call has come in the same shape: a panicked leader telling me their churn has suddenly gotten much worse. It’s usually not true. When we run their actual customer data through a cohort analysis, nearly every one of those companies turns out not to have significantly worse churn. The number on the report moved. Customer behavior didn’t.

Start with what the number is made of. Churn rate is a fraction: lost customers on top, total customers on the bottom. The bottom number is your growth number. So the rate doesn’t only measure churn — it measures sales too, and it can move up or down on a change in sales even when churn didn’t change at all.

That would be tolerable if the two halves moved together. They don’t, because churn is delayed. Time passes between when customers join and when they leave — several months for most companies, and sometimes much longer. The denominator changes in response to sales before the numerator changes in response to churn. Sell a lot this quarter and the base jumps immediately while the churn from those new customers hasn’t arrived yet, so the rate reads better than reality. Sell less this quarter and the base stops growing immediately while last year’s churn keeps landing on schedule, so the rate reads worse. The general rule: “As long as sales are increasing, the churn rate will continually be distorted downward. When sales are decreasing, the churn rate will be distorted upward.”

That is the entire story of the 2023 churn panic. Our benchmark showed new customer sales dramatically down across the industry, and no significant increase in real churn. Measured properly, the multi-year creep in churn actually slowed that year: from 2017 to 2021, churn got worse for 63% of SaaS companies; from 2022 to 2023, only 54%. The year the industry was certain churn was exploding was a year churn improved a little.

Your churn rate doesn’t only measure churn — it also measures growth. When sales are rising, the rate is distorted downward. When sales fall, it’s distorted upward. Most of the churn spikes I get called about turn out to be a sales story.

The sales explanation is the easy one. There is a second one underneath it that will move your rate even when sales hold perfectly steady.

Customers don’t leave at a steady drip. A cohort of customers falls off fast at first and then flattens out — which means a customer’s chance of leaving depends enormously on how long they’ve already been with you. Younger customers churn faster than older ones, every time. (Where the curve flattens is a subject of its own — that floor is your bonded fraction, and I’ve written about it separately.)

So look again at what a monthly churn rate actually is. It’s an average taken across customers of every age at once, and the weights in that average are the ages of your customer base. Churn is a function of average age — which is why, as I noted years ago, if you’re growing, your average churn is biased down. Your rate is a blend, and the blend changes every single month whether you touch anything or not.

Once you see that, whole categories of “spikes” dissolve. A big sales month raises the base right away and pushes the rate down; a few months later that same group hits the stretch where new customers leave hardest, and the rate climbs. Nothing changed except the calendar. I once watched a set of subscribers get flagged as churning worse than everyone else when the only thing true about them was that they were all new, and churn is always higher among the new than the existing. They weren’t worse customers. They were younger ones. Aggregate rates hide that; cohorts show it immediately.

A monthly churn rate averages together customers of every age, and young customers always leave faster than old ones. Change the mix — one unusually big sales month working its way through — and the rate moves even though not one customer behaved differently.

Now the expensive part. Companies act on these numbers. The common pattern is to underinvest in customer success during the rapid growth phase, when the churn rate looks low, and then to panic when the rate appears to spike as sales slow. Both halves cost money. You skip the work when the metric is flattering you, then you buy the emergency — the save program, the reorganization, the platform, the all-hands about loyalty — for a change in behavior that never occurred. Any attempt to act on a distorted metric is literally misguided.

And the rate keeps lying after the panic is over, because it implies churn is linear. A churn rate of 10% a year suggests you’re steadily shedding about 10% of your customers in every period. That is not what’s happening. Most customers who churn tend to leave around the same stage of their lifespan — and there are different kinds of churn, which you can’t tell apart from a rate. Operating to the rate can make churn worse by concealing the real causes: wasted resources on ineffective efforts, failure to address the true causes, misunderstanding why customers stay, marketing to the wrong ideal customers, building the wrong product features.

The fix is not a better rate. There isn’t one. A rate isn’t even comparable to itself from one period to the next, and if a number can’t be compared to itself over time, it certainly can’t be compared to another company’s. You have to change what you look at.

Take the customers who started in the same month and treat them as a group. Then calculate the share of that group still with you after each passing month. That’s it. That curve has no sales number anywhere in it, so nothing about your acquisition rate can move it. Comparing this year’s starting months against last year’s shows whether churn is actually going up or down.

Then you can summarize each curve with one number. Years ago, frustrated with churn rates, I remembered from school chemistry that the decay of a radioactive isotope makes the same downward swooping curve as my cohort charts, and that the measure scientists use to compare isotopes is half-life. So I tried it on customers. Customer half-life is the amount of time it takes to lose half the customers in a cohort. It’s undistorted by whatever is happening in sales, which is why I call it Real Churn, and it stays comparable between cohorts even when the cohorts aren’t the same size. Track it cohort to cohort and you can finally see whether your retention work did anything.

Stop reading months and start reading cohorts. Group customers by the month they signed up, count the share still with you each month after, and compare the curves. That’s churn with the sales distortion taken out, and it’s the only version that can tell you whether anything actually changed.

So before the next spike triggers the next reaction, spend an afternoon on your own customer history. You need two columns you already have: the month each customer signed up, and the month they left if they did. Group by start month, count who’s still there in month one, month two, month three, and lay this year’s groups over last year’s.

If the curves sit on top of each other, your churn didn’t change and your sales rate did. If the newer curves drop faster, something real happened — and you’ll be able to see exactly when in a customer’s life it happens, which is the first useful thing anyone has told you about it. Either way you’ll know, which is more than the number on your dashboard has ever been able to give you. The answer is already sitting in your own customer list.