Churn Prevention

When software was sold once and installed on a customer's own servers, losing a customer wasn't really a thing that happened month to month. The license changed hands, the installation finished, and the deal was closed. Whether that company opened the product every day or forgot it existed made no difference to money already in the bank. The next time anyone checked was two or three years later, when a new version went on sale.

Subscriptions turned that structure inside out. The contract is signed once, but the payment happens again every month, and the customer decides again every month whether to keep going. Leaving stopped being a rare event and became a recurring number. The phenomenon itself is called churn; churn prevention is the name for everything a team does to push that number down.

The arithmetic explains why it got its own name. A product losing 5 percent of its customers every month keeps 54 percent of today's customers a year from now. At 1 percent a month, 88 percent are still there. Two products can add new customers at exactly the same rate and end the year at very different sizes. Closing that gap with new customers costs advertising money; keeping the ones already paying does not.

The hard part used to be timing. The only record of a departure was the moment someone clicked cancel, so what a team held in its hands was always a finished event. Month-end reviews ran the same way every time. Someone would read out "eight accounts cancelled this month," someone else would ask why, and the answer was "let me look into it." Calling those accounts afterward usually reached a team that had already moved its documents into another tool, and there was nothing left to reverse.

Looking back, the warning was always there weeks in advance. A team that logged in daily went quiet for two weeks. Nobody had been invited to the workspace in three months. After the account owner changed jobs, the login records thinned out to a single person. All of it sat in the data, and nobody was reading it. Cancellation was the last click of a long fade, and the team was counting only that click.

Churn prevention moves the moment of attention back into those weeks. Instead of counting who left, you look for who is about to, and instead of arguing on the cancellation screen you act well before anyone reaches it. Retention work inside growth splits in two: engagement pushes active users deeper into the product, while churn prevention pulls back the ones already drifting out. Both look at the same users, but one is raising a number and the other is stopping a loss, which changes what you touch and how fast you have to move.

Five terms sit under churn prevention. They divide into working out who is likely to leave (diagnosis) and deciding what to do about it (response). The order matters, because a wrong diagnosis sends every response to the wrong people.

Diagnosis has two terms, split by whether you are looking at one behavior or one account.

1. Churn signal — a behavior pattern that suggests someone is close to leaving. Logins spreading further apart, a heavily used feature going untouched, the billing contact changing, support tickets spiking or suddenly stopping — all of these are candidates. Which ones actually predict anything differs by product, so teams look at customers who genuinely left over the past six months and find what those accounts did in common beforehand.
2. At-risk user — an account flagged as likely to leave because several signals stack up at once. If a signal is a single behavior, an at-risk user is the verdict you reach after rolling those behaviors up to one account. Weekly active seats dropping below half, plus no new document in 30 days, might be enough to earn the flag. Without this step, the question "who should we contact" has no names attached to it, and the thresholds are worth writing down, because a flag that catches half the customer base is the same as no flag at all.

Response has three terms, and which one applies depends on how far gone the customer is.

1. Reactivation campaign — a set of messages that calls quiet users back. The audience hasn't cancelled yet but hasn't shown up in a while, so these usually go out automatically by email or in-app notification. Timing is wired in advance: a first email 14 days after the last session, a second at 30 days.
2. Save offer — the discount or alternative you put in front of someone who has decided to cancel. Two months at half price, a one-month billing pause, a move down to a cheaper plan. It's the last card and it costs real money, so handing the same offer to everyone who reaches the cancellation screen means paying customers who would have stayed anyway.
3. Cancellation survey — the short questionnaire asking why someone is leaving. It isn't a device for keeping this customer; it collects the material for keeping the next one. One reason picked on the cancellation screen, answers piling up, and after a few hundred you can see which reason dominates. If half say the price, you look at the plans. If half say they never figured out how to use it, you look at the first week after signup.

The five form a loop. Survey answers tell you which behaviors are worth treating as signals, signals pick out at-risk accounts, reactivation reaches those accounts first, save offers meet whoever still arrives at the cancellation screen, and the survey collects reasons again. When the loop doesn't close, the pieces sit apart from each other. Signals without responses produce a risk list that grows every week and gets read by nobody; responses without signals send discount emails to people who never considered leaving.

Which half a team builds first is usually settled by what is cheapest to start. Most products have some version of a reactivation email long before anyone has defined a single signal, because sending email requires no analysis. The response layer then fires at a rough audience — everyone inactive for 30 days — until somebody goes back and works out which of those people were ever going to leave anyway.

The same work happens on the customer success side, and the dividing line is how much one customer pays. Thousands of accounts on free or low-priced plans can't have a human assigned to each, so they are handled with signals and automated messages. Enterprise customers on annual contracts get a named person who calls and visits. The first is churn prevention as growth teams practice it; the second belongs to customer success. The vocabulary is shared, but the time available per account differs by orders of magnitude.

"Twenty accounts landed on the risk list this week. Let's start with the six whose renewal date is next week."

There aren't enough hands for all of them, so the ones with the least time left go first.

"People open the reactivation emails but they don't come back. I don't think this is an email problem — they never had a reason to use it in the first place."

The cause sits earlier in the funnel, in the first week after signup, not in the churn prevention step.

"Only send the save offer when the stated reason is price. Discounting a customer who's leaving over a missing feature just moves the same cancellation two months out."

Separate the churn a discount can stop from the churn it can't, and spend the money only on the first kind.

At Pagedeck the phrase comes up most often in the Monday retention review. Ravi puts up the accounts flagged as at-risk during the previous week, and Sofia pulls out the ones on annual contracts to work through herself. Everything left over goes to automated messages. Mia reads the same list looking for whether users of one particular feature are overrepresented, because a cluster like that is a product problem rather than a set of individual customer situations. Last quarter, fourteen of the twenty-three flagged accounts were teams that had been using an outside integration and stopped; the integration had quietly been broken for a month.

Grace asks the same question in that meeting every time: are the accounts reported as saved last month still here this month? Delaying a cancellation with a discount and genuinely getting a team back into the product look identical in the month they happen, and only separate two or three months later. So Pagedeck tags saved accounts for three months and recounts what share survived once the tag expires. Three months isn't an arbitrary window — most save offers there run for two, and a shorter window would count discounted accounts as saved while the discount is still paying their bill.

The most common mistake is treating churn prevention as something that happens on the cancellation screen. That's where the saving is visible, and where the discount popup appears, so the assumption is easy to form. But anyone who reaches that screen decided weeks ago and has usually already picked a replacement. The share who reverse course there tends to be in the single digits. The stretch that actually moves the number comes earlier, during the weeks when someone is still technically using the product but showing up less. A team that works only on the cancellation screen can rewrite the popup copy a dozen times and watch next month's churn rate hold exactly where it was. One way to see the gap is to look at where those accounts stood a month before they cancelled: most were still logging in, and most had already stopped doing the one thing that made the product worth paying for.

The second mistake is scoring a save as a success regardless of how it was won. The result shows up as a number the same month, which invites that reading. But what happens next depends entirely on the method. A customer held with a discount comes back to the cancellation screen the month the discount ends, and by then the decision has already been made once, so the second attempt almost never lands. A customer who stayed because they finally started using the feature they were missing tends not to appear on the risk list again. Counting both as the same outcome pushes the team toward whichever is cheap and fast, which is always the discount. Six months on, saves have doubled while average revenue per account has dropped and the churn rate hasn't moved.