Winback & retention

What a winback programme is actually doing

Diagnosis, then treatment. "Has not bought in 90 days" is a symptom that four completely different customers share: one ran out later than average and never left, one was disappointed, one moved to somebody else, and one is no longer the right customer at all. A single discount blast tells all four that waiting is rewarded, and it is the most common winback programme in existence.

The one rule: segment by cause, never by recency alone. Recency tells you who to look at. It never tells you what to send.

How to think about it

• Derive the lapse threshold from your own repurchase cycle. Median days between orders, then the 75th percentile as the point where a customer is genuinely unusual rather than merely slow. A consumable might lapse at sixty days and a durable at eighteen months. Ninety days is somebody else's number and it is on every template because it is round, not because it is right.
• The at-risk window matters more than the lapsed one. Keeping somebody costs a fraction of winning them back. A programme that only fires after the lapse threshold is running late by design.
• Start with the cheapest thing that could work. A reminder, then an acknowledgement and a fix, then a value add, then, only where nothing else moved, a discount. Escalate on evidence, not on impatience.
• Do not mail into an unfixed problem. Winning a disappointed customer back into the same bad experience costs more than leaving them alone, and it converts a quiet lapse into a public one.
• A predictable discount is a price cut you never decided to make. If the offer always lands at day ninety, customers learn to wait until day ninety, and your best customers learn it fastest.
• Some lapses are correct. Life changed, the need ended, they were never the right fit. Suppress rather than spend, and count the suppression as a win: it protects the deliverability the rest of the programme runs on.

Failure modes worth naming: an apology so vague it reads as a template, which confirms that nobody actually noticed; a winback mailing that reaches somebody who complained, refunded, or died, which is how a retention programme becomes a reputation problem in one send; and a report of programme revenue with no holdout under it, which will always conclude that discounting works.

Numbers to hold it against

These are reactivation numbers, and they are deliberately unflattering. A winback list is the least engaged audience the business owns, and pretending otherwise is how the discount ladder gets climbed to the bottom.

• Winback opens run well below your list average, commonly in the low tens of percent, because these addresses stopped engaging some time ago. That is expected and it is not a subject-line failure.
• A well-built sequence reactivates a low single-digit percentage of a lapsed segment inside 30 to 60 days. The top of that range comes from relevance, not from a bigger discount, and the mistimed segment is where most of it lives.
• The holdout is the whole story. Hold back 5 to 10 percent of each segment and mail them nothing. A meaningful share of lapsed customers return on their own, so the untested programme claims orders it merely witnessed, and the gap between raw reactivation and lift is routinely large enough to reverse the conclusion.
• Margin, not revenue. At a 50 percent gross margin a 20 percent discount needs the units to roughly double to hold gross profit flat, and a discounted order that would have arrived at full price is a straight donation. Run that sum on the offer ladder's floor before the first send.
• Sunset after 6 to 12 months of no opens or clicks, following one final attempt. Continuing to mail dead addresses is what drags the whole programme's deliverability down: mailbox providers read engagement, not intention.
• The 90-day repeat rate of reactivated customers is the number that separates retention from a discounted one-off. A programme that produces one order and no second one has not retained anybody, it has run a sale to a small list.

What moves these: category and repurchase cycle above all, how long the segment has been lapsed (recency of lapse predicts reactivation more strongly than any copy decision), whether the cause was fixable, and how much the brand was worth to them in the first place.

What has to be true before a card asks for approval

The lapse definition. Derived from the real cycle, split per product family where the cycles genuinely differ, with the at-risk window stated as well as the lapsed one, and the exclusions written down: refunded, complained, unsubscribed, fraud, deceased. Getting the exclusions wrong is the fastest way to turn a retention programme into a reputation problem.

Churn reason cards. Evidence and a rough share, from support tickets, reviews, a survey, cancellation notes or returns data. Anything without evidence is labelled a hypothesis rather than dressed as a finding. And an honest line on whether marketing can fix it at all, because a product or delivery problem is not solved by an email.

Segment cards. An identification rule somebody could run in the tool today, a size, and the what-would-backfire line filled in. That line is the one people skip and the one that saves the margin.

The offer ladder. Cheapest first, the floor named, and the person who signs off on exceptions. State what you will not do, specifically: the offer that would win this order and lose the customer's willingness to ever pay full price again.

Emails. The acknowledgement names the specific problem and what changed, and it does not send until the problem is actually fixed. Any discount carries its terms, expiry and exclusions in the email itself. Suppressions applied and verified before launch, not after. No placeholder or unresolved bracket surviving into approval.

The seed cards are the first instance, not the ceiling

Eight cards show the shape of the thinnest honest programme. A running one is denser, and the Create phase multiplies.

Diagnose holds the lapse definition plus one card per reason, usually three to five, including the ones marketing cannot fix. Segment holds one card per cause, three to four, plus the sunset segment that most boards forget until deliverability reminds them. Offer holds the ladder and, once several sequences run, a sequence card per segment. Create is where the arithmetic bites: one email per segment per step, so three segments running a two-step sequence is six email cards, not two. Measure holds the result card with a row per segment, re-read each quarter.

Two email cards on a board with four named causes means two causes are being treated and two are being mailed the wrong thing.

The layout is one arrangement of many

The columns run by production step: diagnose, segment, offer, create, launch, measure. That fits the first build, because the diagnosis has to come first and the layout enforces it. Once several sequences are live, a column per cause reads better: each cause owns its segment, its offer, its emails and its own row of the result, and the parallel structure makes an underperforming sequence obvious. Rearrange it. Segment by cause survives any layout; the left-to-right order is only there to stop anybody starting at the email.

When this is the wrong playbook

If the product or the delivery problem is unfixed, this board will win people back into it and make things worse. If the list is small enough that a 5 to 10 percent holdout leaves no readable sample, run the sequences anyway but stop short of claiming lift you cannot measure. If the honest answer is that the customer is no longer the right customer, suppress and move on: the cheapest winback is the one you decline to run. If first orders never became second orders in the first place, the post-purchase board is upstream of this one and will do more. And if acquisition itself has stalled, there is very little to win back, and this board will spend a month proving it elegantly.