The decision to kill or scale a Meta creative comes down to two things: whether you have enough data to trust the result, and which way the trend is moving. Kill a creative once it has spent enough to judge and its cost per acquisition is sitting above your target with no sign of recovery. Scale one that holds a CPA below target over several days with a strong, stable score. The mistake almost everyone makes is acting too early — judging a creative on a single day, or before its ad set has cleared Meta's learning phase, when the numbers are still noise.
Key takeaways
- Get enough data first. An ad set needs ~50 optimization events in a rolling 7 days to exit Meta's learning phase; judging before that is guessing.
- Kill when CPA stays above target after meaningful spend, the score is low and falling, or fatigue has set in without recovery.
- Scale when CPA holds below target for several days with a strong, stable score and there's audience headroom.
- Scale slowly — budget jumps above ~20% can reset learning. Increase incrementally or duplicate.
- Decide on store-reconciled true ROAS, not Meta's reported number, or you'll scale the wrong winners.
Give it enough data before you judge
The single biggest cause of bad kill/scale calls is impatience. A new ad set enters Meta's learning phase and needs roughly 50 optimization events within a rolling 7-day window before delivery stabilizes — and that pool is shared across all ads in the ad set, not per creative. Until you're past that, daily CPA swings wildly and means little. Jon Loomer has a good breakdown of how the learning phase works and why editing during it resets the clock. The practical rule: don't kill a creative on one bad day, and don't judge an ad set that hasn't exited learning.
When to kill a creative
Kill a creative when the evidence is clear and pointed the wrong way:
- CPA has stayed above your target after enough spend to be confident — typically once the ad set is out of learning and the creative has accumulated a meaningful number of conversions, not just impressions.
- Its creative score is low and still falling week over week, not just having an off day.
- It has clearly fatigued — rising frequency, falling CTR — and a refresh of the hook hasn't pulled it back. (More on the pattern in creative fatigue signals.)
- Negative feedback is climbing: hides, "see less," and a falling like-to-spend ratio.
Killing isn't failure — it's how you free budget for the creatives that are working. The cost of leaving a loser running is the spend it wastes plus the budget it starves from a winner.
When to scale a creative
Scale when a creative has earned it across several days, not one good afternoon:
- CPA holds at or below target consistently, post-learning, on a stable audience.
- The creative score is high and steady, not spiking and crashing.
- There's audience headroom — frequency is still reasonable, so more budget reaches new people rather than re-serving the same ones.
One nuance worth knowing: ad sets that consistently clear well past 50 events a week tend to outperform those scraping the minimum, because Meta's model has more confidence. So a winner that lets you push volume isn't just cheaper, it often gets more efficient as it scales — up to the point where you saturate the audience.
How to scale without resetting learning
Scaling clumsily can undo the very performance you're trying to grow. Big budget changes — roughly a 20%+ jump in daily spend — can re-trigger the learning phase and reset your progress. Two safer paths: raise the budget in modest increments every few days so delivery re-stabilizes between steps, or duplicate the winning ad set into a new campaign and scale the copy while the original keeps running. Avoid stacking edits; each change is another chance to reset learning.
The number that actually decides it: true ROAS
Here's the trap: if you make kill/scale calls on Meta's reported ROAS, you'll scale the creatives Meta says performed, which over-credits view-throughs and in-platform attribution. Decide on the number reconciled against your store instead. We cover why the two disagree in true ROAS vs Meta-reported ROAS, and reconciling it is exactly what True ROAS does. Scaling on a defensible number is the difference between growing profit and growing reported numbers.
Use a score to triage the decision
Across a dozen client accounts, you can't eyeball every creative daily. A single account-relative score with a 7-day delta lets you triage: protect the high-and-stable, watch the falling, and queue the low-and-declining for a kill or refresh decision. That's what Advino's creative scoring surfaces nightly, on the same flat $19–$299/mo plan — so the kill/scale call starts from a ranked list instead of a hunch.
Frequently asked questions
How long should I run a Meta ad before killing it?
Long enough to exit the learning phase and accumulate a meaningful number of conversions — not impressions. As a rule of thumb, wait until the ad set has cleared ~50 events in a 7-day window and the creative has had a fair share of that spend before deciding. Killing on one bad day is the most common mistake.
When should I scale a winning Facebook ad?
Once it holds a CPA at or below target for several consecutive days, post-learning, with a stable score and audience headroom. Then scale slowly — incremental budget increases or a duplicated ad set — to avoid resetting the learning phase.
Does increasing budget reset the learning phase?
A large change can. Budget swings of roughly 20% or more in daily spend often re-trigger learning. Smaller, spaced-out increases let delivery re-stabilize without a full reset.
Should I kill or just refresh a fatiguing creative?
If the concept still works and only the execution is tired, refresh the hook or rotate a variation rather than killing it outright. Kill when the creative is underperforming on a defensible CPA even after a refresh, or when its score is low and falling with no recovery.