I removed Meta’s top spenders from a campaign. To be precise, I duplicated the campaign without those ads and ran it in parallel.
The algorithm kept scaling them even though none reached target.
My tROAS campaign had 5 ads eating 27% of budget, and none of them were reaching my 30% D0 ROAS target. Day 0 is an early read, but in my own test day 0 ROAS predicted day 28 ROAS better than cost per purchase.
The setup
- Duplicated the tROAS campaign: same geo, same optimization goal, same total budget
- Removed the top spenders, the 5 ads Meta kept scaling despite low D0
- Ran both for 30 days, $18.3K on the original against $17.5K on the test
- Compared what each algorithm chose to scale
Would the algorithm find better D0 performers if its favorites were not available?
What each algorithm chose
Top five ads by spend in each campaign. iOS subscription app, Meta tROAS, 30 days, names randomized. The scale runs from 0% to 40% D0 ROAS, the dashed marker is the 30% target, and green bars reached it.
Not one of the original campaign’s top five reached target. Four of the test campaign’s did.
The test won 4 of the 5 weeks. More usefully, the algorithm found 6 new creatives above target that never got a chance in the original campaign.
The algorithm has favorites
Concentrated spend is normal well beyond my account, and in AppsFlyer’s 2025 creative optimization report the top 2% of gaming creatives pulled in 53% of total spend.
And its favorites are not always your best performers on D0. Removing them pushed spend to other ads, and several of those did better. I found the same habit by country in a $383K tROAS audit where failing geos kept their budget.
Run it on your own account
Copy the setup above and hold to two rules:
- Keep the budgets equal. If the test campaign spends less, you cannot tell whether the improvement came from the creative mix or from the spend level.
- Read the result as a strong signal, not a clean split test, because two campaigns running side by side share an audience.