CASE STUDY / VERIFIED
Flat for the best part of a week. The platform said everything was fine.
We repaired the attribution signal and ran agents over the creative and the budget. Under two sales a day became over nine across the paid run, on lower spend, a 6.3x return verified on Meta's own API and reconciled to the bank. Meta delivery ran nine days, 12 to 20 May 2026.
Book a callWHAT YOU CAN CHECK
The claims on this page, and how they were checked
Return on the paid run, checked against Meta's own API and reconciled to money that landed in the bank. Treat it as a floor: attribution was only repaired in the last day or two of the run.
Meta's own API, reconciled to the bank · 12 to 20 May 2026Sales a day across the paid run, up from under 2 before it, on lower spend. The closing four days peaked past fourteen, which is a peak and not the run rate.
Pledge record, reconciled to the bank · 12 to 20 May 2026Meta delivery ran nine days. The longer twelve-day figure you may see elsewhere is the reconciled window running to the campaign close, which is a different thing.
Meta delivery · 12 to 20 May 2026Pledges in the paid window settled, none refunded.
Pledge record, reconciled window · 12 to 23 May 2026What is live, what is roadmap and what we have not built is set out in the facts table.
SALES A DAY
Under two a day before the run. Over nine across it.
The paid run lifted the daily rate and held it, on lower spend. The last four days ran hotter again, and that peak is drawn apart from the run rate so nobody reads one as the other.
The stall.
The campaign had been flat for the best part of a week, money going out and very little coming back. And the platform's own reporting said everything was fine. That gap between what Meta claimed and what the bank account said is the whole disease we treat.
The look.
We brought the accounts into the warehouse and did what a dashboard can't: reconciled the platform's numbers against real sales. The attribution was broken at the timing level. Conversions were landing but reporting late and mis-stamped, so the algorithm was flying blind, and the reporting layer was papering over it.
The fix.
We repaired the server-side signal so the platform could finally see what was actually converting. Then we ran agents recursively over the creative and the budget, reviewing the work and moving the money day after day, at a pace no weekly meeting can match.
The scoreboard.
Two a day became over nine a day across the paid run, on lower spend, and the closing four days peaked past fourteen. A 6.3x return, and here's the bit that matters: that figure is verified against Meta's own API and checked against money in the bank, not taken from the platform's scorecard. A scorecard reports what the platform can see. We report what happened, and when the two disagree we can show you exactly where and why.
THE KICKER
“Nobody needed a cleverer ad. Once the machine could see what was selling, the decisions were obvious.”
WHAT THIS MEANS FOR YOU
The villain in this story is in your clients' accounts too. Somewhere, a platform's scorecard and the bank disagree, and the disagreement is costing money nobody can see. The warehouse that found it here is the one Kist puts under your clients' reporting. We ran it on our own business first, and we still run it there every day. Nothing gets shipped to a client that we haven't run on ourselves.
Every read is scoped to one client before it runs. A query that reaches outside that client is refused, not quietly trimmed.