creative fatigue on facebook ads: the 4 signals frequency won't show you.
by the time frequency spikes, the CPI damage is already booked. the four leading signals numu watches instead, the thresholds we use, and a 10-minute weekly audit.
frequency is where most teams look for creative fatigue on facebook ads. it's the worst place to look. frequency is a lagging indicator — by the time it's high enough to alarm you, the damage is already in the CPI and the algorithm has already started pulling delivery. numu watches four leading signals instead, and acts on the composite, not any one alone.
why frequency lies
frequency tells you how often the average person saw the ad. it says nothing about whether they still care. a creative can be fresh at frequency 5 in one audience and burned out at frequency 2 in another. worse, it moves slowly — it's an average over a window, so it smooths over exactly the sharp drop you needed to catch on tuesday. use it as context, never as the trigger.
the four signals that lead
| signal | what it catches | rough trigger |
|---|---|---|
| hook-rate decay (72h) | the thumbstop dying before the click does | down 15%+ over 72h |
| CTR-to-frequency slope | clicks falling faster than reach grows | negative and steepening |
| thumbstop delta vs ad-set median | this creative fading relative to its peers | below median two days running |
| CPI variance day-over-day | instability that precedes the real climb | a widening band, not just a spike |
each threshold gets tuned per genre — a hypercasual creative fatigues on a different clock than a mid-core one.
why the composite beats any single signal
any one of these throws false positives. hook rate wobbles on a bad traffic day; CPI variance jumps on a holiday. together, they're specific: on our production data, the four-signal composite pauses a genuinely fatigued creative without human review and is right about 94% of the time. one signal is a rumor. four agreeing is a decision.
a 10-minute weekly fatigue audit
- rank creatives by hook-rate change over the last 72h, not by lifetime CTR.
- flag anything below its ad-set thumbstop median two days running.
- check CPI variance, not just the average — a widening band is the early tell.
- only then glance at frequency, to confirm what the leading signals already told you.
- refresh the flagged winners before you pause them — a fatigued concept often has one more variant in it.
the version that runs itself
this is the exact playbook we encoded into numu — it watches all four signals across every creative, acts on the composite inside your guardrails, and briefs the refresh instead of just killing the ad. put it on your account.
