During the learning phase, Meta's delivery system explores different audience segments and placements to figure out who is most likely to convert, which typically produces less stable and less efficient results than once the ad set has exited learning. Meta generally recommends around 50 optimization events per week before performance stabilizes.
Significant edits — changing budget by a large percentage, editing creative, or adjusting targeting — reset the learning phase, which is why agencies avoid frequent tinkering with live ad sets and instead prefer to duplicate an ad set for testing changes, preserving the original's learned delivery.
We don't talk. We solve.
Private Meta infrastructure for buyers spending $100,000+ a month. Unlimited accounts, our cards, a 24/7 operator line.