It is optimising signal, not structure
Meta ranks ads on the probability of a desired action and the bid you are willing to pay for it, then spends where that combination looks best. Early performance is noisy, so the algorithm needs volume before it can tell a winner from a slow start.
Which is why the tidiest account in the world — one ad set per audience, one ad each, ten campaigns — often underperforms a messier one where each creative gets enough spend to prove itself. You are not building a taxonomy for a human. You are feeding a model.
What good structure looks like at $50K+ a month
Consolidate until each campaign has enough budget to exit the learning phase, and no further. Usually that means a small number of broadly targeted campaigns split by funnel stage, each holding a batch of distinct creative concepts rather than one creative and twelve audiences.
Creative is the variable with the most headroom. Audiences have been consumed by the platform’s targeting; the ad itself is still yours to lose.
- Fewer campaigns, each with enough budget to learn.
- Broad targeting, with creative doing the segmentation.
- Budget increased in steps, not doubled overnight.
The campaigns people refuse to switch off
Retargeting that takes credit for traffic the brand created anyway, and brand-search campaigns that report enormous ROAS against demand that would have converted without them. Both look excellent in a report and both are frequently just paying for demand that already existed.
Test them honestly: switch one off for two weeks and measure total revenue, not platform-attributed revenue. The difference is what the campaign was actually buying.