The B2C Churn and Retention Playbook > Chapter 4
A win-back campaign is a sequence of messages and offers aimed at customers who have stopped buying or cancelled, timed to the window when they’re most likely to return. It’s also the one lifecycle program where ignoring most of the list is the strategy.
That sounds wrong until you look at who’s on the list. Some lapsed customers will come back on their own: they moved, they got busy, the season turned. A larger group will never come back, and no offer changes that. The group in the middle will return if you reach them with the right message in the right window. A win-back campaign exists for the middle. Sending it to all three groups pays the first for nothing, wastes money on the second, and under-serves the third.
This chapter is the campaign side of win-back: how to define “lapsed” for your business, how to find the middle group, when to reach them, what to send, and how to know it worked. If you want the argument for why winback beats acquisition and where that math gets abused, our post on how to win back customers and who’s worth the offer makes it. This page assumes you’re convinced and need to run the thing.

Most win-back programs start with a rule like “no purchase in 90 days.” The rule is wrong for almost everyone it touches, because 90 days is a long gap for a coffee buyer and a short one for someone who buys running shoes. Lapsed has to be defined against the business, and usually against the customer’s own rhythm.
| Industry | Lapsed means | Typical return window |
|---|---|---|
| Ecommerce | No order for 1.5x to 3x the customer’s own purchase gap, measured per customer rather than with a fixed day count | One more purchase cycle; past 3x the gap, the odds fall quickly |
| Subscription and streaming | Cancelled or expired, with voluntary cancels separated from failed payments | The first 30 to 90 days after cancelling; resubscribers are a large share of sign-ups |
| Gaming | No session for 7 to 14 days, depending on genre; a shorter window for payers | Days, and most of the returners come back unprompted |
| Fintech | Account open but no deposits or transactions for 60 to 90 days; primary status lost | The next one or two pay cycles; after a quarter, dormancy hardens |
| Insurance | Policy not renewed; the customer switched | The next renewal anniversary, 11 to 12 months out, when they shop again |
The subscription row deserves a number. Antenna’s analysis of premium streaming found that of roughly 169 million gross sign-ups over a year, some 57 million (34%) were resubscribers. A third of the front door is people coming back. The win-back window in streaming opens the day they cancel and is wide open for the next quarter.
The gaming row deserves a quote. When SciPlay looked at lapsed players, its VP of Marketing Technologies put it plainly: “most will return within a few days or a week”, which is the reason the studio stopped retargeting everyone who went quiet and started predicting who wouldn’t come back alone.
Take a brand with 200,000 lapsed customers and an average order of $50. The standard play is 25% off to the whole list. Here’s what that does when the list contains, as lists like this usually do, three kinds of people: say 30,000 who’d return anyway, 120,000 who never will, and 50,000 who could go either way.
Among the 30,000 returners, most come back, and most of those use the coupon because it’s sitting in their inbox. Call it 18,000 redemptions at $12.50 of margin each: $225,000 paid for behavior that was already happening. Among the 120,000 lost causes, nobody comes back, and 120,000 sends plus the paid remarketing budget behind them go nowhere. Among the 50,000 persuadables, the offer works on, say, 12%: 6,000 customers return who otherwise wouldn’t have, at $75,000 in discounts.
| Scenario | Sends | Reported “reactivations” | Incremental returns | Discount paid | Cost per incremental return |
|---|---|---|---|---|---|
| 25% off to all 200,000 | 200,000 | 36,000 | 6,000 | $300,000 | $50 |
| 25% off to the 50,000 persuadables only | 50,000 | 6,000 | 6,000 | $75,000 | $12.50 |
The blast reports 36,000 reactivations and looks like a triumph. It cost $50 per customer won back, which is the entire first order. The aimed campaign wins back the same 6,000 people for a quarter of the discount spend, and it sends the 30,000 returners a no-discount nudge (their usual product is back in stock) that they’d have acted on anyway. The 120,000 get nothing, which is the first time that group has ever been profitable.
The numbers are illustrative. The shape holds: every win-back list we’ve seen splits roughly this way, and the only thing that changes is how well you can tell the groups apart. That’s a prediction problem: a model trained on your past lapsed customers, which ones came back, and what they looked like before they did. The propensity modeling chapter in our next best action playbook covers how that model gets framed. The short version is that “will this lapsed customer return in the next 60 days, with and without an offer” is a question your historical data can already answer.

Lapsed customers come back in a window, and the window closes. Your campaign calendar has nothing to do with it.
In subscription, sooner is better. The cancelled streaming subscriber is most winnable in the first weeks, while the habit is still warm and the replacement hasn’t stuck. By the time the quarterly “we miss you” goes out, the window has mostly shut. Meal kits and memberships behave the same way, with a second, smaller window at the next seasonal change.
Ecommerce windows are a multiple of the customer’s own gap. A shopper at 1.5x their normal interval is drifting, and at 3x they’ve probably found another store. The useful moment is about ten days past the expected reorder date, with the product they usually buy in the first line. Seasonal buyers are a special case: the customer who buys sunscreen every May isn’t lapsed in November, and the campaign that treats them as lapsed trains them to wait for a discount.
Gaming’s window is days, and most returners don’t need the push. The campaign is for the players the model says won’t come back on their own, reached before day 14.
Fintech windows follow pay cycles. A customer whose deposits stopped last month is reachable in the next two. After a quarter of silence the account has usually become somebody else’s primary.
Insurance has the longest window: the next renewal anniversary, eleven to twelve months after they switched. That’s the one case where patience wins: the campaign goes in the diary for 60 days before the competitor’s renewal date, with a quote in hand.
The old win-back playbook leads with the biggest discount it can afford. Ours leads with the prediction.
| Predicted likelihood to return | What to send |
|---|---|
| High (they’d come back anyway) | A reason, with no discount: a restock, a new arrival, the menu for next week, a feature they asked for |
| Medium (persuadable) | An escalating sequence: free shipping or a perk first, a real discount only if the first two messages are ignored |
| Low, but moved by price | A real offer, capped below whatever discount they last used, with a deadline |
| Very low | Nothing beyond a one-question survey; suppress from email and from paid remarketing audiences |
One rule holds for every row: never send a bigger discount than the one that lost them. A customer acquired at 20% off who lapsed and is won back at 30% off has been taught that waiting pays, and they’ll lapse again on schedule. Perks that don’t touch price (early access, a loyalty tier, a free add-on) hold up better, and in subscription a pause or a cheaper plan is usually a stronger win-back than a discount on the plan they already rejected.
The survey is worth a word, because the older advice to “ask for feedback” was right about the question and wrong about the audience. Ask the lost causes. They’re the cheapest market research you’ll ever run, they have nothing to lose by being honest, and one question (“what would have kept you?”) sent to 120,000 people with no offer attached tells you more about your product than any retention offer will.

Email is the default channel, SMS is the escalation for the persuadable middle, and paid remarketing is where win-back quietly spends the most. Lapsed-customer audiences get synced to Meta and Google by default, and then the ad platforms spend happily against the 120,000 people who were never coming back.
Suppressing the never-returning group from paid remarketing is usually the single most profitable thing a win-back program does, before any message is sent. Push the model’s lowest-probability tier to the platforms as an exclusion audience and watch the remarketing cost per returned customer fall. The same exclusion keeps the sure-things from seeing a discount ad on the way back to a purchase they were already going to make.
Channel-switch on silence. A persuadable who has ignored three emails isn’t going to open the fourth. The next touch is an SMS, a piece of direct mail for high-value accounts, or for the top of the list, a call. And know when to stop. A short, polite final message that offers one last reason and then removes the customer from the sequence protects your sender reputation and, oddly, gets replies.
Count incremental returns. Raw reactivation counts flatter everyone, and the only way to get the honest number is a holdout: a random slice of eligible lapsed customers, across every tier, who get no win-back contact at all. Compare their return rate to the treated group’s over the window. The difference, times the treated count, is what the campaign did. Divide the cost (sends, discounts redeemed, remarketing spend) by that number and you have cost per incremental returned customer, which is the only figure finance should see.
It’s also the figure that makes the case for prediction. When Hydrant, a DTC drink-mix brand, built a model on customers who hadn’t purchased for at least two months, using their activity over the prior 60 to 180 days, and aimed win-back offers at the customers predicted to respond, that group had a 2.6x higher conversion rate and a 3.1x higher average revenue per customer than a control group. The control group is what turns that from a nice campaign into a proven one.
The next chapter, on measuring retention lift with a holdout, covers how big the holdout needs to be and how to read it. The lifecycle team’s version of all this, gating the reactivation flow with a prediction inside the ESP, is in our lifecycle playbook chapter on who will actually come back.

Telling the three groups apart is the part that used to need a data science team. Now it’s one sentence: you bring your historical customer data and the question (“which lapsed customers will return in the next 60 days, and which of them need an offer to do it?”), the Predictive AI Agent reads that data as it is, trains and validates the model, and we own the outcome, with a return score and a response score for every lapsed customer live in the ESP, CRM and ad platforms your team already uses. Our customer winback page shows what those scores look like and where they land.
See who will churn in the next 60 days:
Book a demoA win-back campaign is a sequence of messages and offers aimed at customers who have stopped buying or cancelled, timed to the window when they are most likely to return. A good one is aimed only at the lapsed customers who will return because of it, and measured against a holdout group that receives nothing.
Inside the return window for your business: the first weeks after a subscription cancel, about ten days past a shopper’s expected reorder date, before day 14 for a lapsed player, within the next pay cycle or two for a dormant account, and 60 days before the next renewal for a switched policyholder. A fixed “90 days of inactivity” rule is late for some customers and early for others.
There is no useful benchmark, because raw conversion counts customers who would have returned anyway. Measure the return rate of the treated group against a random holdout, and judge the campaign on cost per incremental returned customer.