The first ninety minutes of a reactivation push

Our published median from a signed brief to the first handset is 38 minutes. People read that as speed. It is not. It means the decisions were made before the brief arrived. The segment definition, the sender identity, the template clearance and the contact caps are standing work that already sits ready. What happens on the day is a choice, not a build.

Here is a real push, minute by minute. One market, one brand, a lapsed file nobody had touched in eleven weeks.

Minute zero: read the brief, and be willing to send it back

A workable brief is three lines. The market, the segment we are aiming at, and the offer with its terms. Ours came in at 14:06 on a Thursday: Peru, lapsed 30 to 90 days, 40 free spins on a named slot, no deposit needed to claim.

A brief is rarely unworkable because of the offer. It is one of four things:

• The offer is not live in the platform yet, so a player who clicks lands on nothing.
• The segment is defined by a field nobody can resolve at send time, usually a marketing tag nobody maintains any more.
• The market has no registered sender identity, or the one it has sits with the wrong entity.
• The terms in the message and the terms in the platform disagree. That is a complaint, not a campaign.

We check those four first, because each one is unfixable at minute sixty and trivial at minute two. This brief had the offer live and the sender registered. Cleared at 14:11.

Minute five: cut the file into value tiers

The raw lapsed file was 152,000 rows. That is not a segment, that is a spreadsheet. The consent record, the weekly contact cap and the self-exclusion register took it to 128,400 sendable. Then it is cut on deposit history, not on a lifetime figure that flatters anyone who had one good month two years ago.

Three tiers came out: 6,100 high value, 34,800 mid, 87,500 low. The high tier does not go in this push. It gets a person, its own ladder and a manual review, and a 40-spin no-deposit offer is the wrong thing to put in front of someone who used to stake four figures a week. That leaves the mid tier plus the recent end of the low tier: 96,400 players.

Minute twenty: two variants that differ in exactly one thing

Both variants carry the same offer, the same terms, the same signed link, the same sender. The single difference is the opening clause. Variant A leads with the spins. Variant B leads with the named slot the player last played.

That discipline is not fussiness. Change two things and a win teaches you nothing you can use next time. One variable, or do not bother splitting. The deciding metric is written down before the send goes out, and on reactivation work it is always deposits, never clicks. A message that pulls clicks and no deposits has cost you money and taught the list to ignore you.

The split is drawn on a hash of the player id, so the two arms are one segment cut in half rather than two segments that look similar. Written and approved at 14:33.

Minute thirty-eight: the first wave

The first wave is a tenth of the segment: 9,640 players, 4,820 to each arm. It went at 14:44.

A tenth rather than a half, because the first wave is the only cheap mistake available. If the link is wrong, if the offer code is not attached to the account, you have burned 9,640 players and not 96,400. We have sent the wrong short domain before. Once. The first wave is the reason it was not a whole campaign.

Minute fifty-three: fifteen minutes of data, and what it is worth

By 14:59 the first wave had settled. 9,371 delivered, 97.2% on first attempt. Variant A had pulled 573 clicks against 380 for variant B. Deposits attributed so far: A 19, B 7.

Here is what those fifteen minutes tell you. The plumbing works. Links resolve, the offer attaches, deposit events are coming back from the platform and landing on the right arm. That is the entire job of the first quarter hour and it is worth a lot.

Here is what it does not tell you. It does not tell you which variant won. 4,820 a side is nowhere near our floor of 9,000 per arm, and deposits landing this early come from impatient players, not representative ones. A 19 to 7 gap on those volumes is a direction, not a result. We do not call a test on it, and we do not let an operator call it either.

Minute sixty-two: the point where a variant dies

Wave two went at 15:08 and doubled each arm to 9,660 a side, past the floor. By 15:31 the picture held. A had 41 attributed deposits, B had 14, on the same delivery. B was not converting the clicks it did get.

B was killed at 15:34. Not paused, not left to run out its remaining waves so the report looks symmetrical. A losing variant allowed to finish is money you have chosen to spend on a message you already know is worse. Roughly one test in three overturns the variant the operator expected to win. That is why the split exists, and why the beaten arm stops when it is beaten, not when it is convenient.

Minute ninety: the rest of the list, and the report

The remaining 77,100 went out as waves three and four on variant A, spaced across the evening against the player's own local clock.

What lands in the operator's inbox is one page, sent while the promotion is still open rather than the following Tuesday:

1. Sent, delivered, clicked and deposited, by wave and by variant.
2. The call. Which variant won, on what margin, at what sample size.
3. Attributed deposit count and value inside the 72-hour window, assisted counted separately.
4. Opt-out rate on the front page, not the back.
5. Anything that did not reconcile against the operator's own reporting, named.

On this one: 93,700 delivered, 466 attributed deposits at a median first deposit of EUR 68, so EUR 31,700 against EUR 4,150 of send spend. That is 7.6x, a little above our blended 6.4x, which is what a clean market and an untouched file will do for you. Opt-out came in at 0.38%.

None of that came from sending faster. It came from only risking a tenth of the list at 14:44, and being willing at 15:34 to stop something we had written ourselves.