Sample diagnosis · UK ecommerce · Meta
This is the analysis behind the paid diagnosis, run on a real account. Four hundred and two ads, every day of their lives, replayed against a set of kill rules to answer one question: what would have happened if the losers had been switched off when the data first said so.
Every account has ads that stopped working and nobody turned off. They rarely look dramatic. They spend a little, return a little, and the account average absorbs them, which is exactly why they survive for months.
The useful question is not which ads are bad today. It is how much money the account would have kept if there had been a rule, applied consistently, from the day each ad first failed it. That is a question you can answer with data you already have, and it is the core of the diagnosis.
Daily per-ad performance is pulled straight from the platform, then every ad is replayed forward through its own lifetime. On each day, each ad is tested against the rules in order. The first rule it fails is the day it would have been killed, and everything it spent after that day is counted as recoverable.
The order matters more than the thresholds. A shield rule that runs first is what stops a mechanical system from killing a winner on a bad Tuesday.
Of 402 ads, roughly a fifth survived every rule, half would have been killed, and the rest never gathered enough data to judge. That last group is not a failure of the method. Knowing which ads you genuinely cannot judge yet is a finding.
Nearly all of it from one rule: ads that had been running long enough to prove themselves and were still returning under 1.0. Not spectacular failures, just quiet ones nobody had a rule for.
This is the number that matters. Cutting spend is easy; cutting spend without cutting revenue is the whole point. Removing two fifths of the budget cost less than a twentieth of the sales.
From roughly 2.1 to roughly 3.2 across the same period, on the same creative, with no new ads and no new budget. The improvement came entirely from stopping things sooner.
The aggressive version of this is not the version I would run.
Tightening the evaluation window produces better headline numbers and also kills far more ads, including some that would have recovered. The run above is the aggressive one. A calmer setting on the same account returns a smaller improvement with a fraction of the interruption, and on a real account that trade matters more than the headline does.
So the deliverable is not a number. It is a set of thresholds tuned to your margins and your tolerance, with the evidence for why each one sits where it does.
Five working days from the moment access lands, then a call to walk you through it. $500 fixed, credited against your first invoice if we carry on.
See the rest of the method: the checklist, the QA routine and a full monthly report
Send me what you are spending a month and what you want paid acquisition to do next. If the honest answer is that ads are not your bottleneck, the document will say so.