Meta · Published 2026-04-29 · 3 min read
Why switching off a creative breaks the campaign
The case
A buyer launches an ad set with five creatives. A week later he looks at the results
- Creative 1 — Spend: $10k, Predicted ROAS 50%
- Creative 2 — Spend: $2k, Predicted ROAS 150%
- Creative 3 — Spend: $3k, Predicted ROAS 105%
- Creative 4 — Spend: $1.5k, Predicted ROAS 110%
- Creative 5 — Spend: $1k, Predicted ROAS 35%
The decision: the buyer switches off creatives 1 and 5 because they do not make money. The logic is to push the budget onto creatives 2, 3 and 4, which have positive Predicted ROAS.
📉 What happened. After creatives 1 and 5 were switched off, the ad set's ROAS dropped noticeably. CPA went up. Overall performance sagged. The buyer expected the budget to move to the best creative and the campaign to improve. The opposite happened.
🧪 Simulation. Here is a simulation of the effect. You can play with it and watch the numbers move if you would rather not read.
🧠 Why it works this way: two models. The buyer's model. The buyer assumes every creative competes for one pool of audience. Creative 1 takes budget away from creative 2. Switch off the bad one and the good one gets more and earns more. In this picture the creatives work the same audience and convert it into installs.
How Meta actually works. Each creative gets its OWN audience. That audience has a size of its own — creative 2 can bring users at $2,000 of spend with ROAS 150%. It cannot supply $4,000 of spend. Creative 2 showed ROAS 150% not because it is “better”, but because Meta found it a small, high-quality audience worth $2,000. That is its ceiling at that ROAS. When creative 1 was switched off, Meta redirected the budget to creative 2 — but its quality audience was already used up. Meta widens the targeting by showing the creative to less loyal users. ROAS falls as spend rises.
✅ The right approach
- Do not switch creatives off. Meta distributes budget between creatives inside an ad set on its own. If one creative got $1,000 and another $200, that is not an algorithm error. It reflects the volume of audience available.
- Make variations of the good creative. If creative 2 showed ROAS 150%, make 5–10 similar ones. Each variation gets its own audience. That widens reach across quality segments without force-widening the targeting of a single creative.
- If the task is to improve the ad set's ROAS at any cost, then in this example cutting spend will certainly help. Less spend — higher ROAS.
📝 NOTE. I have simplified things for clarity, of course.
- At small spend, ROAS of 150% on a creative is often just noise.
- Sometimes META really does miss a good creative, and every buyer has a story about the top performer META overlooked.
Both of those are fair, and yet I still think that actively hand-managing an ad set cannot be a productive long-term strategy.