Find out why your Meta ads broke

When CPA spikes or leads dry up, the default guess — “the creative's fatigued, kill it and refresh” — is wrong often enough that acting on it can torch a winning ad while the real problem keeps bleeding the account.

A drop has a small number of realistic causes, and they demand opposite responses. Creative fatigue says refresh the ad. A delivery or budget change says restore the structure and stop editing. A traffic-quality break — clicks holding but conversions gone — usually points off-platform, at a landing page or a tracking failure. A demand or seasonality shift says wait rather than act. Guess wrong and you spend a week fixing the thing that was not broken.

This diagnostic removes the fatigue bias structurally. Four investigators each examine an isolated slice of your export: the fatigue investigator only sees frequency and CTR per ad — never spend; the delivery investigator only sees spend, impressions and CPM; the traffic investigator watches the click-to-conversion funnel; the demand investigator sees conversion volume alone. None of them knows the other theories exist.

Then an adversarial referee gets every finding plus the full dataset and tries to kill each theory. Fatigue gets pressure-tested hardest: flat frequency and holding CTR mark it dead, and a same-day cliff across all ads is never fatigue. What survives gets ranked — cause first, symptoms attributed to their cause — and you get the single most important action for this week.

The blinding is the point. An analyst who can see every column at once anchors on the first plausible story and then finds evidence for it. An investigator who can only see frequency and CTR has no way to reach for a budget explanation, so its theory has to stand on its own data before the referee ever sees it.

Export a daily report broken down by Day and by Ad, covering at least two weeks either side of the drop, as CSV and under 1MB. Include Day, Ad name, Amount spent, Impressions, Frequency, CPM, CTR (link), Link clicks, and Leads or Purchases. The more of the funnel you include, the sharper the traffic investigator's read. Nothing is stored — the file is parsed in memory and discarded.

A CSV cannot see placement mix, landing-page edits, pixel health or a competitor's promotion, so those come back as “needs manual check” rather than as invented causes. If the surviving theory is fatigue, confirm the timing on the specific ad with the fatigue forecast. If the drop looks structural rather than creative, the health grader scores the whole account across waste, budget efficiency and testing velocity in one pass.