PERFORMANCEPerformance · May 2026
← Journal
PerformanceMar 2026 · 6 min read
When to kill an ad: a practical guide to creative fatigue
Most teams kill ads too late. A small group kills them too early. Here are the specific signals we use to make the call — and the rule we use when the signals disagree.
PERFORMANCE⏤ Summarize with AI
Open this article in your favourite chatbot for a 3-bullet summary.
Killing ads too late is the most common cause of monthly ROAS slumps. Killing them too early is the most common cause of stalled accounts. Here’s the framework we use to make the call.
The signals, in order
Kill an ad on these signals, in order: frequency climbing past 2.5–3.5, hook-rate and CTR declining while frequency rises, CPM creeping up, and ROAS halving. Archive rather than delete — fatigued winners often revive after 4–8 weeks of rest.
1. Frequency
At frequency < 1.5, you’re still in expansion. At > 2.5, the same humans have seen this ad 3+ times. Past 3.5, you’re paying Meta to remind people they aren’t buying.
2. CPM trend
Compare today’s CPM to last week’s, on the same audience. If CPM is up > 25% and frequency is up, the audience pocket is full.
3. Hold rate / 3-second video views
The earliest fatigue indicator. If 3-second view rate falls 10% over 5 days while spend is held flat, your hook is tiring — even if ROAS still looks fine. This is the leading indicator. Most others are lagging.
4. CTR
CTR drops are the noisiest signal. Don’t use it alone. Combined with frequency and CPM, it confirms.
5. ROAS
Lagging. By the time ROAS visibly crashed, you lost a week of spend. Don’t wait for this one.
When NOT to kill
- First 3–5 days. Learning phase. Bad numbers don’t mean bad ad.
- Spend < ₹3K total. Statistical noise.
- Right after a budget jump. Meta re-learns. Wait 48 hours.
- During a category-wide spike (Black Friday, Diwali). Everyone’s CPMs are up; yours rising isn’t a fatigue signal.
Killing without losing the win
When you kill a winner that’s fatigued, archive — don’t delete. Most fatigued ads come back to life after 4–8 weeks of rest. Re-running an old winner with refreshed targeting is one of the cheapest tests in the playbook.
The replacement rule
Don’t replace 1 killed ad with 1 new ad. The variance bus you’re on is creative diversity, not 1:1 substitution. Kill 1, ship 3 — ideally in different formats. The algorithm will pick the replacement.
Frequently asked questions
What frequency is too high for a Meta ad in 2026?
Frequency above 2.5 is the warning band; above 3.5 you're paying Meta to remind people who already saw the ad that they aren't buying. Kill the ad when frequency crosses 3 and ROAS or hold rate is also slipping. Below 1.5 means you're still in expansion — don't touch the ad.
How early can I tell an ad is fatiguing?
The 3-second video view rate ("hold rate") is the leading indicator. If it drops 10%+ over 5 days while spend is held flat, the hook is tiring — even if ROAS still looks fine. CTR is a noisier signal; CPM is a confirming signal; ROAS is lagging. By the time ROAS visibly crashes, you've lost a week of spend.
When should I NOT kill an ad even if numbers look bad?
Four exceptions: (1) first 3–5 days of the learning phase — bad numbers don't mean a bad ad; (2) total spend under ₹3,000 — statistical noise; (3) right after a budget jump — Meta re-learns, wait 48 hours; (4) during a category-wide spike like Black Friday or Diwali when everyone's CPMs are up. Yours rising isn't a fatigue signal in those windows.
When two fatigue signals disagree, which wins?
Frequency. If frequency is above 3 and ROAS still looks decent, the ROAS is reporting old buyers — the ad has stopped finding new ones. Kill it. The reverse — low frequency, bad ROAS — usually means the ad is wrong, not fatigued, and the fix is creative not timing.
Should I delete a fatigued ad or archive it?
Archive, don't delete. Most fatigued ads come back to life after 4–8 weeks of rest, especially with refreshed targeting. Re-running an old winner is one of the cheapest tests in the playbook — you already know the creative converts, you're just waiting for the audience pool to refresh.
How many new ads should I ship when I kill one?
Don't replace 1 killed ad with 1 new ad. Kill 1, ship 3 — ideally in different formats. Creative diversity is the lever, not 1:1 substitution. Let the algorithm pick the replacement winner from your new batch; that's how the next 7-day winner gets discovered.
Where this fits
Fatigue is the most common reason a healthy account starts slipping. If your ROAS is dropping, fatigue is step 2 of our 6-step ROAS diagnostic — and it’s why CPMs keep rising. For the full picture, see our D2C marketing pillar guide.
⏤ Free audit
Get a free audit in 5 working days.
Read-only access to your Meta + Google Ads accounts. Full teardown of wasted spend, creative fatigue, and ROAS forecast. No obligation.
⏤ Keep reading
Related from the journal.
⏤ Working on something similar?
Get a free audit in 5 working days.
Read-only access to your Meta + Google Ads accounts. Full teardown of wasted spend, creative fatigue, and ROAS forecast. No obligation. Or jump straight to our performance marketing page.