The account was working. Then, over something between two weeks and two months, cost per acquisition climbed by half and nobody can point at the day it started. This is the most common problem in performance media and it is almost always misdiagnosed, because the first instinct — make new creative — is the expensive fix and only one of four possible causes.
Diagnose in this order, cheapest first.
Start with tracking, because a broken pixel produces exactly the symptom of a failing account and costs nothing to rule out. Compare platform-reported conversions against what your own back end recorded for the same days. If the platform is under-reporting, nothing is wrong with the ads — the algorithm has simply gone partially blind, and it optimises worse as a result. A consent banner change, an app update, a developer removing 'that script nobody uses' — all of these look like a marketing problem for weeks before anyone checks.
Then check the auction, not your account. Pull cost per thousand impressions over the same period. If CPM rose in line with your CPA, your ads did not get worse — the space got more expensive. Ramadan, back-to-school, a category competitor raising a round, a global advertiser expanding into your placements. There is no creative that fixes an auction, only a decision about whether the new price still clears your margin.
Then audience, which is where the real answer usually is. Look at frequency inside your best-performing segment. When a set of people has seen the same offer eleven times, the ones who were going to convert already have, and everyone left is being paid for repeatedly to be shown something they have declined. This is the failure that pretends to be creative fatigue and is actually arithmetic: the pool ran out. New creative shown to the same exhausted pool buys you two good weeks and the same problem again.
Only then creative, and specifically the part that fatigues. Not the whole ad — the hook. The first two seconds and the first line carry almost all of the decay, because they are what a returning viewer recognises. A new edit of the same concept with a different opening frequently recovers most of the performance, which is worth knowing before commissioning a shoot.
There is a fifth cause nobody wants to be told about and it is worth naming: the offer got worse relative to the market. A competitor started shipping faster, or undercut you, or a category leader made your price look unreasonable. Advertising is a magnifier — when the thing being magnified changes, the magnifier gets blamed first and is innocent.
The practical discipline that prevents all of this is unglamorous: know your baselines before you need them. Frequency, CPM, click-through, and back-end-confirmed conversions, recorded weekly. Diagnosing a decline is nearly impossible without a picture of what normal looked like — and the month it starts is the wrong month to begin measuring.