Dormant is not dead: reading a lapsed segment

Most operators hand us a lapsed file as one thing. One export, one date filter, one message written for all of it. It is not one thing. It is several groups of people with different reasons for going quiet, and treating them as one audience is why the last win-back campaign returned a few hundred euros and a pile of opt-outs.

A well cut segment with a mediocre offer beats a brilliant offer sprayed at everyone, every time we have measured it. This is what we do to a lapsed file before a word gets written.

Quiet is a state, gone is a different one

A player who has not deposited in sixty days has not necessarily left. Plenty are seasonal, or on a break they chose themselves, or waiting for a sport that is out of season. They still open messages. They are winnable.

The tell is not the gap in days, it is what the account did on the way out. A player who tapered off across five sessions is quiet. A player who deposited hard, lost it and closed the app inside an hour is usually gone. So the first cut is behavioural: the shape of the last three sessions, whether deposits trended down or fell off a cliff, whether a withdrawal happened. That separates a pause from an exit better than any recency band we have tried.

A lapsed VIP is not a casual with a bigger number

Someone takes the ladder that works on the mass file, swaps in a larger offer, points it at the top tier, watches it underperform, and concludes the high-value base is unreachable. It is not. It leaves for different reasons. A casual player drifts because nothing held their attention. A high-value player almost always leaves after a specific event: a withdrawal that took four days, a limit applied without warning, a host who stopped answering. They remember the event, and a generic we miss you lands as proof nobody noticed.

They also return on a different timetable. Casual returners mostly come back inside three weeks. High-value returners cluster later, often past day fifty, and respond to something closer to an acknowledgement than to an offer. That is why a person reads the top tier before anything sends. The value per head pays for the hour.

Cut tiers from deposit history, not a lifetime figure

Lifetime value carries too much history. It flattens a player who deposited heavily for one month two years ago into the same tier as one who has deposited steadily every month since. Those two need different messages. We cut from the deposit sequence instead, on four inputs:

1. Median deposit across the last twelve months of activity, not the mean. One outlier weekend should not promote a player.
2. Deposit frequency during the active period, so a monthly depositor and a daily one separate even when the totals match.
3. Direction of travel over the final six deposits. Rising, flat or declining.
4. Days since the last deposit, applied last, as a spacing input rather than a tier boundary.

That gives four value tiers crossed with three recency bands. Twelve cells, not one list. We work three or four hard and leave the rest, because a cell that has never returned above one percent is not a segment, it is a cost.

The return curve, and where it flattens

Every brand has its own curve and the only way to get it is to look. Plot returns against days since last deposit for players who came back with nothing sent. You get a steep early slope, a shoulder, then a long flat tail. The shoulder is where messaging earns its money. Before it you are paying to contact people who were coming back anyway. After the flat begins you are contacting people who have decided.

Our median reactivation gap across licensed markets is forty-one days, but that is blended and it is not your number. On one sportsbook brand the shoulder sat at nineteen days. On a slots-heavy brand it sat past seventy. Ladder spacing comes from that curve, not from a weekly cadence inherited off an email calendar.

What changes after roughly day forty

Somewhere around the six-week mark offer-led messages stop performing. Before it, the player still holds a live sense of the account. They remember the last session, and a bonus reads as a reason to do what they were already half thinking about. The message can be short and transactional. Offer, amount, expiry.

After it, that sense has gone. An offer arrives with nothing to attach to and reads like an advert from a company they used to use. What works is a message that reinstates the context first: what is in the account, what changed since they left, what is on this weekend in the thing they actually played. The offer moves to the end, or drops out of the first touch. Past day forty, context-first has beaten offer-first in most of our splits, by margins wide enough that it is now the default.

Three unanswered touches, then the segment rests

A hard ceiling of three. If a player has not answered three messages in a ladder, they leave the programme and the cell rests. There is no fourth touch. It converts almost nobody and costs something that never appears on the campaign report.

Burning a list is the expensive mistake here, and it is expensive because it stays invisible for a while. Push contact frequency up and deposits rise for two months. Underneath, the people who would have returned in month four are opting out in month two. You do not lose a campaign, you lose the asset. Opt-out rate is the instrument, which is why it sits on the front page of every report. Ours runs at 0.42% blended. When a cell drifts above about 0.9% we stop it the same day, because deposits lag and opt-outs lead.

Exclusion state is checked at send time, every time

A lapsed file is the file most likely to contain someone who has self-excluded or taken a cool-off since anyone last looked. These are the accounts nobody has touched in months.

So the check happens when the message is queued, against live state, for every destination. Never from an overnight export, never from the file the segment was cut from, never once at the top of a ladder that then runs three weeks. A player who excludes on Tuesday must not receive the wave that goes out on Wednesday, and the only way to guarantee that is to resolve the audience at send time. The cost is that a segment size is not final until the wave fires. We would rather explain a segment two percent smaller than defend a message that should never have been sent.