The signature is a cluster, not a single number. On a fatigued ad set, frequency and CPM drift up while CTR and CVR drift down, and CAC rises in step, with bid, budget, and audience held constant. A one-off CTR dip is noise; the co-movement names the condition.
Mechanically, the decay traces to audience saturation against a finite creative inventory, not a platform-side event. The un-fatigued slice shrinks with every impression on the same creative, and the marginal impression stops doing causal work even as the ad set keeps spending.
Smaller audiences burn faster than broad targeting because the reachable-user denominator is smaller, so retargeting fatigues sooner than prospecting at the same cadence. Privacy-era signal loss (ATT on iOS and ongoing web-tracking changes) adds directional pressure: the measurable retargeting universe has narrowed, magnitude varying by category and stack. Audience overlap compounds the problem: per-audience frequency caps don’t compose, so a user in three pools sees the creative roughly three times as often as any single cap suggests. The thresholds at which buyers intervene are conventions, not laws: modest on prospecting, materially higher on retargeting, shifted by category, creative, and audience.
A fatigued set also drifts from blended reads. Platform-reported ROAS can hold up on diminishing impressions while MER softens. The platform keeps crediting conversions those impressions no longer caused.
The operator response is creative-pipeline sizing, not bid or audience tweaks. Concept (a distinct hook or angle) and variant (a re-cut of the same concept) are different testing units, and on a fatigued account, volume of new concepts beats sophistication on the winner.