--- name: diagnose-creative-fatigue description: >- Reads whether a previously-winning Meta creative has fatigued, rising frequency (roughly >2-3 in a short window), declining CTR/hook rate, and rising CPM/CPA, all on the SAME creative and trending gradually, by pulling the day-by-day trend via mads_run_insights (daily time_increment) rather than judging a single bad day. Rules out account-wide auction pressure or seasonality producing the same symptoms before concluding fatigue, then prescribes a real refresh (new concept/hook) or a wider audience over a cosmetic edit that won't reset genuine wear-out. Reach for it when a winner's cost per result has been climbing for more than a few days, or as a standing cadence check on long-running creative. It ends at a refreshed, live slate plus a rotation cadence, it does not source the replacement creative itself (run-creative-testing-framework supplies validated winners) or diagnose a creative that never hooked well to begin with (improve-hook-rate-and-hold). --- # Diagnose and Fix Creative Fatigue ## Purpose Every winning creative eventually decays: the same audience sees it enough times that it stops working, which shows up as climbing frequency, sliding CTR/hook rate, and rising CPM/CPA on a creative that used to perform. This skill reads that trend correctly, rules out the other things that produce the same symptoms, auction pressure, seasonality, and prescribes a real refresh instead of a cosmetic one. ## When to run - A previously-winning creative's cost per result has been climbing for more than a few days. - Frequency on an ad set is climbing without a corresponding increase in audience size. - A scheduled cadence check on creative that has been live for an extended run, a standing maintenance pass, not just a reaction to a complaint. ## When NOT to run - The creative never worked well to begin with, that's a hook/hold problem, not fatigue; see `improve-hook-rate-and-hold`. - No winner exists yet to fatigue, run `run-creative-testing-framework` first. - Performance dropped account-wide, on the same day, across unrelated creative, check auction and seasonality causes (this skill's own Decision rules) before assuming fatigue. ## Prerequisites - A creative with enough runtime/history to show a trend, not just a few days of data. - Access to day-by-day (not just aggregate) spend, frequency, CTR, CPM, and CPA for the ad(s) in question. - Awareness of any account-wide events in the window, budget changes, competitor activity, seasonal demand shifts, that could explain the same symptoms without fatigue. ## Procedure 1. **Pull the day-by-day trend.** `mads_run_insights` (poll with `mads_get_insights`) at the ad level, `time_increment=1`, for frequency, CTR, hook rate where available, CPM, CPA, and spend, over several weeks, not 3 days. 2. **Plot the signals against time**, not just today-vs-30-day-average, fatigue shows as a gradual slide, not a cliff. 3. **Check frequency.** IF frequency has climbed past roughly 2-3 within the recent window on a stable audience size, THEN the same people are seeing this creative repeatedly, a leading fatigue indicator. 4. **Check CTR/hook rate.** IF click-through or hook rate is declining over the same window on the same creative, THEN interest is genuinely wearing out, not just a one-day dip. 5. **Check CPM/CPA.** IF cost metrics are climbing on this creative specifically, and not account-wide, THEN the auction is paying more to reach an increasingly saturated audience, confirms fatigue rather than a broader cost shift. 6. **Rule out non-fatigue causes before concluding fatigue.** Check whether CPM/CPA rose account-wide on the same days (auction pressure across all advertisers) or whether the calendar explains it (seasonal demand, competitor spend spikes, holidays). IF the rise is account-wide or calendar-correlated, THEN it isn't creative fatigue, don't refresh creative to fix a market-wide cost shift. 7. **Choose the fix.** A real refresh, new concept or hook, per Decision rules, beats a cosmetic edit (recolor, retrim the same footage); cosmetic swaps don't reset genuine audience wear-out. 8. **Ship the refresh.** `mads_create_creative` for the new concept, `mads_create_ad` (status PAUSED) alongside the fatigued ad rather than a blind replacement, `mads_preview_creative`, plain-English summary, explicit approval, then commit. 9. **Pause the fatigued creative** (`mads_update_ad`, status=PAUSED) once the refresh is live and has enough data to stand on its own, don't run both indefinitely once the old one is confirmed spent. 10. **Set a rotation cadence.** A standing check every few weeks against the frequency/CTR/CPM signals so the next fatigue cycle is caught on trend, not after cost per result has already climbed for weeks. 11. **Consider widening the audience** alongside or instead of a creative refresh when frequency is the dominant signal and the current audience is unnecessarily narrow, more reach lowers frequency without needing new creative at all. ## Decision rules - **Fatigue signature:** rising frequency (roughly >2-3 in a short window), declining CTR/hook rate, and rising CPM/CPA, all on the SAME creative, trending gradually. The combination matters more than any single metric. - **Not fatigue:** cost or frequency moves that are account-wide or calendar-correlated (holiday, competitor surge, broad auction pressure), the fix is bid, budget, or timing, not creative. - **Real refresh vs. cosmetic swap:** a new concept, angle, or hook counts; a color/logo/music swap on the same footage does not, audiences fatigue on the underlying idea, not the exact pixels. - **Frequency-dominant fatigue on a narrow audience →** widen the audience first. **Frequency-fine but CTR-declining fatigue →** the creative itself needs replacing regardless of audience size. - **Refresh slate comes from `run-creative-testing-framework`**, not from guessing, route the next concept through a proper test rather than shipping an unvalidated replacement at scale. - **Done means:** the day-by-day trend confirmed genuine fatigue (not auction or seasonality), a real refresh shipped and is live, the fatigued creative is paused once the refresh has data, and a rotation cadence is set. ## Common failure modes - **Refreshing creative to fix an account-wide auction cost spike** that had nothing to do with the creative itself. - **Cosmetic edits mistaken for a real refresh**, fatigue returns almost immediately. - **Reacting to a single bad day** instead of confirming a trend across the full window. - **No forward cadence set**, so the next fatigue cycle is only caught after cost per result has already climbed for weeks. - **Killing the fatigued creative before the replacement has enough data** to confirm it's actually better. ## Related skills - Before: `run-creative-testing-framework` (the source of any validated replacement creative). - Related: `improve-hook-rate-and-hold` (a declining hook rate is one of this skill's signals, but a creative that never hooked well in the first place is a different diagnosis). - Related: `select-formats-and-placements` (widening delivery via an added placement or format is one lever against rising frequency).