Live in 14 days: a first performance campaign for FlapKap

The starting point

There was no performance channel to improve. The account started from zero on 5 April 2026 — no creative in market, no baseline to beat, and no way to know which of the things everyone assumed about the buyer were true.

That cuts both ways. Nothing to inherit also means nothing to unlearn, but it means the first weeks are the whole test: if the launch creative misses, there is no history to fall back on and no data to brief the next round from.

Same script, different cast

Two versions of one concept went out together — identical script, identical edit, different presenter. One of them brought qualified leads around 40% cheaper than the other.

Nothing about that was predictable from a brief. We did not guess which face MENA founders would trust; we shipped both and let the account answer. That answer then briefed everything cast afterwards.

25 concepts, 117 variants in four months

Not a campaign — a production line. Every week's winner briefs the next week's tests, which is why the concept count matters more than any single idea in it. A premise that works becomes a family of hooks; one that does not is closed out rather than defended.

What the AI pipeline actually bought

Two weeks from blank page to live is not a scheduling trick. It is what happens when production stops being the long pole — scripting, asset generation and voiceover run at once, and a senior director reviews output instead of waiting for it. A conventional model books those stages one after another, and the calendar alone would have eaten the fortnight before anything was shot.

The four-month output says the same thing from the other end. 25 concepts and 117 variants is not a shoot schedule. In a traditional pipeline each of those is separately scoped, booked, crewed and billed, and the number a client can afford is set by the production budget rather than by what the testing needs.

The cheapest thing it bought was the freedom to be wrong. Shooting the same script with a second presenter is a rounding error in a generated pipeline and a second shoot day in a conventional one. That single test is why we know the face was worth around 40% of the cost per qualified lead — and in a model where finding that out costs a crew, nobody runs it.

Measured against the creative already running

The cleanest read on this engagement is not against a benchmark, it is against the client's own in-house creative — same account, same weeks, same landing pages, same audiences. The only variable was the creative itself.

On that comparison our creative brought qualified leads 16% cheaper, at 3.6 times the outbound click-through. Over the period it carried roughly two-thirds of the account's spend and produced around 70% of its qualified leads.

Figures are drawn from the client's own Meta account, 5 April – 10 August 2026, and published as relative movements. Absolute spend, budgets, cost per lead and lead counts belong to the client and are not disclosed here.