Winback starts with an honest definition of lapsed

A furniture retailer runs a winback programme on a ninety-day rule. Someone who bought a sofa in November receives an email in February saying "we miss you", followed by twenty per cent off. They were never gone. They bought a sofa, and the next one is eight years away.

In the same week, a coffee subscription business uses the same ninety-day rule. Its customers reorder every three weeks. By the time the winback fires, the person has been buying from a competitor for two months and the relationship is over.

Ninety days is not a finding. It is a default that arrived with someone's previous platform and never got questioned. Every winback programme we review starts here, because the definition of lapsed decides who is in the audience, and nothing downstream can fix an audience that is wrong.

Ninety days is a habit, not a measurement

Ask where the number came from and the answers are consistent. It was in the template. It matched the reporting period. Someone senior said quarterly felt right.

The cost is not obvious from the reports, which is why the number survives. A winback sent too early goes to active customers who would have returned anyway, and the revenue is attributed to the campaign. A winback sent too late goes to people who have already replaced you, and the low response rate is blamed on the creative.

Both errors make the programme look like it is working while it is not. The first inflates results, the second explains away failure. A measured window removes both arguments.

Derive the window from your own repeat curve

The honest version of lapsed comes from the interval between purchases in your own data. It takes an afternoon.

Work through it in this order:

1. Take everyone with at least two purchases in the last two years, and compute the gap in days between consecutive orders.
2. Plot the distribution of those gaps and read off the median and the upper quartile.
3. Take customers whose gap has already exceeded a candidate threshold, and measure what share ever purchase again.
4. Find the point where that share falls away sharply. That is where a normal gap becomes a genuine lapse.
5. Add a margin, because acting slightly early costs less than acting after someone has switched.

The result is a number you can defend in a room. For a category with a twenty-one day reorder cycle the answer might be forty-five days. For a category people buy twice a year it might be nine months.

Do this as cohort analysis, not as a single blended figure. First-time buyers, repeat buyers and subscribers have different curves, and averaging them produces a window that fits nobody.

Two segmentations usually justify a separate window. Product category, because a consumable and a durable have nothing in common. And customer tenure, since the gap after a first purchase behaves differently from the gap after a fifth.

Lapsed, disengaged and unreachable are three different states

Most programmes collapse three distinct problems into one audience, then wonder why one message cannot serve it.

• Lapsed: still opening and clicking, has not purchased within the window. The gap is commercial, so the message is about the product or the reason they stopped.
• Disengaged: no opens or clicks for a long period, purchase status unknown. The gap is attention, so the task is re-permission or a change of channel.
• Unreachable: hard bounces, spam complaints, or a mailbox that has stopped accepting mail. There is no message to send, only records to suppress.

The states need separate windows and separate exits. Someone can be a perfectly good customer who stopped opening email while still buying in store, and mailing them a discount teaches them that ignoring you is profitable.

Track each state as a count of profiles, updated weekly. Movement between states is the most useful retention signal most programmes are not looking at, and it predicts the churn rate before the revenue number moves.

Build the winback around the reason, not the discount

Once the audience is right, the temptation is to reach for a percentage. A discount is the fastest thing to approve and the easiest thing to measure, which is why it becomes the whole programme.

It also trains people to wait. A customer who learns that lapsing produces fifteen per cent off has been given a reason to lapse. The same problem shows up in checkout recovery, and we argued the case against reflexive discounting in basket recovery without discounts.

Work from the plausible reasons instead. There are rarely more than five, and you can usually name them from support tickets, returns data and cancellation reasons.

• The product ran out and they simply forgot to reorder.
• They had a bad experience, such as a late delivery or a failed return.
• The thing they bought is not due for replacement yet.
• They found a competitor, on price, range or convenience.
• Their circumstances changed and the category no longer applies.

Each reason implies a different message, and two of them imply no message at all. A reorder reminder timed to consumption beats a generic discount. An apology with a resolution beats both, when the data says the last order went wrong.

A sunset policy belongs inside the winback, not after it

The end of the winback sequence is a decision point, and it should be written into the design rather than left open. If three attempts across a defined period produce nothing, stop mailing that profile on the marketing stream.

That is what a sunset policy is for. It protects deliverability, keeps the retention curve honest, and stops the audience filling with addresses that have not responded in two years.

Suppression does not have to mean deletion. Keep the profile, keep it available for paid audiences and for service messages, and allow re-entry on any real signal such as a site visit or a purchase.

Set the volume ceiling here too. Winback sequences run alongside broadcasts, and without central frequency capping a lapsed customer can receive the reactivation series and the weekly newsletter in the same afternoon.

What to do this week

• Compute the gap between consecutive purchases for your repeat buyers, and find the point where the probability of returning drops away.
• Split your current winback audience into lapsed, disengaged and unreachable, and count how many profiles sit in each.
• Write the three most likely reasons your customers stop, and check whether any message in the current sequence speaks to any of them.

The part that stays difficult is the discount. Removing it costs measurable short-term revenue and returns margin that is harder to point at, and that argument needs the numbers ready before you start it. The execution work is straightforward once the definition is agreed.