half the replies here are fighting over a gate Meta never actually built. The learning phase isn't a checkpoint you pass through before performance unlocks. Meta's own help doc says an ad set exits learning as soon as it can deliver stably, which usually happens after about 50 results in the week following the last significant edit. that's a stability label, nothing more. They also straight up tell you not to chase avoiding the phase if your ad set lands in Learning Limited. so the folks saying it's no holy grail are right, and the folks calling it fake are wrong. it's just describing something a lot more boring.
Now the math. At $37.50/day you're working with roughly $262 a week. Four purchases on $431 spend comes out to about a $108 CPA. Hitting "about 50" purchases a week at that CPA would take around $5,400 in weekly spend, not $5k total. so no, Purchase optimization isn't getting this ad set out of learning at the current budget. And honestly that's fine, because exiting learning isn't the problem. A 0.55 ROAS is the problem. Pushing $100/day into a 0.55 ROAS is setting money on fire.
One more thing worth knowing: a big budget jump can itself reset learning. Meta's example is that $100 to $1,000 resets it, while $100 to $101 doesn't. Going $37 to $100 is a 2.7x move, so that hike would torch the little signal you have left, just to lose more. Watched that exact move bite someone on an account I touched recently, it was painful to sit through.
The wonky daily spend isn't a bug either. Meta treats a daily budget as a target across a calendar week, Sunday through Saturday. Their system can spend up to 75% over your daily budget on certain days and underspend on others, as long as the full week stays under 7x. An $18 day and a $9 day on a $20 budget sits comfortably inside that. judge it Saturday night, not Tuesday morning.
What I'd actually run with $37/day and a 0.55 ROAS: the people pushing add-to-cart optimization have the mechanism right (rule of thumb is optimize for the deepest event you can realistically get 50 of per week), but changing the optimization event counts as a significant edit and wipes everything, plus it drags in cheaper, lower-intent traffic. before touching that, verify the purchase event is actually firing against real orders. Four tracked purchases could easily be six real ones with a broken pixel, and that flips the entire picture.
