Customers have never had more options. With so many DTC brands offering similar products and benefits, it’s harder than ever to stand out—and customers are less likely to stay loyal for long.
When established customers cancel their subscriptions or otherwise leave your business, that is called “churn.”
Churn is costly: your brand spends a lot of money up and down the funnel to acquire the customer. The customer buys from you for a time, creating a steady cash flow. Then the customer vanishes—and the cash flow with them.
You notice the impact on both your CAC and your bottom line. The longer a customer stays with you, the greater the chance they’ll achieve high LTV. But if a customer defects soon after acquisition, you forfeit their payback period and see your margins erode.
Churn modeling helps marketers measure a customer’s likelihood of defecting—and estimate when churn risk increases.
Churn Isn’t Random
Customers don’t “just” disappear: they often leave clues before defecting. Order sizes get smaller. The length of time between orders gets longer. Purchases become less frequent. Basic RFM analysis can help you spot this pattern.
If you provide a service, watch how customer usage changes. For example, a wireless subscriber who is likely to churn may suddenly use less data, make fewer calls, or log in to fewer apps. An at-risk subscriber may also register more customer support complaints, contact customer service to ask about contract terms or renewal dates, or miss payments.
These early warning signals give you an opportunity to act before the customer leaves.
Who Might Churn (Logistic Regression)
We’ve discussed how logistic regression is used in customer acquisition. The same idea applies to retention. Logistic regression can be used to predict a customer’s probability of churn at a specific point in time. But therein lies a limitation: logistic regression answers, “Will they churn?”, not “When will they churn?”
Knowing which customers will churn is important—but knowing when they will churn is key. That’s where survival analysis comes in.
When Customers Churn (Survival Analysis)
Survival analysis measures the time until a specific event—like churn or cancellation—occurs. While logistic regression asks whether a customer will defect, survival analysis asks how long they will stay.
One survival analysis method, the Kaplan-Meier Estimator, is applied to cohorts of similar customers and measures their probability of remaining active over time. Often used in the pharmaceutical industry to measure drug survival during clinical trials, Kaplan-Meier can also be used in marketing to benchmark retention across customer groups.
As customers churn over time, those events are plotted on a survival chart like the one below, which depicts a winery’s retention of wine club members over several months.
Looking at the chart, notice that one month after joining the wine club (Time=1), the survival probability is 0.95—that is, a member has a 95% probability of remaining one month after sign-up.
Retention declines steadily throughout the year, with the steepest drop between months six and seven—from 57% to 38%.
That is a key decision point. The winery needs to intervene before this drop to reduce churn risk. In the months leading up to it, the winery should look for members who are skipping shipments, pausing subscriptions, or switching to one-time purchases—and target them for retention.
To reduce churn, the winery might send members a “skip or swap” reminder, offer customized shipments, or give them the option to gift their next shipment to a friend. It could also send a small preview sample of upcoming wines to build anticipation or provide recipes and pairing suggestions to deepen engagement.
When combined with customer transaction data, Kaplan-Meier not only highlights when churn risk spikes but helps you identify and act on at-risk customers before that point.
Prediction Without Action is Useless
There are multiple survival analysis methods besides Kaplan-Meier, but their purpose is the same: to help you identify periods of elevated churn risk, target high-risk customers, and act.
The value of churn modeling isn’t prediction—it’s intervention.
We saw how the winery could reduce cancellations by identifying members approaching the six-month mark who are showing signs of disengagement, and targeting them with relevant offers and messaging.
Churn doesn’t happen all at once—it builds over time. The brands that win aren’t the ones that react after customers leave, but the ones that recognize the signals early and intervene at the right moment. Survival analysis gives you that timing. What you do with it determines whether customers stay—or walk away.
