Advantage+ audience: keep the detailed targeting suggestions or go fully broad?

Say you're running a Meta Advantage+ sales campaign for a mid-size apparel store, and the audience setup keeps nudging you to add detailed targeting suggestions and layer in interests. Some people swear by it because it gives the algorithm a starting point. My view is that for an account with a decent purchase history, the suggestions usually do more harm than good. Once the pixel has a few hundred purchasers to learn from, a broad setup tends to find the same buyers on its own, and interests mostly just shrink your reach and make it harder to tell which setting is doing the work. I'd run it broad for two weeks and only add suggestions if CPA stalls and you can say why. Has anyone seen the reverse, especially on smaller catalogs?

Agreed for stores with a few hundred buyers, but on smaller catalogs broad can sit in learning for weeks. If the ad set exits learning within about a week, stay broad. If it's stuck, the real problem is conversion volume, so a light seed of suggestions is a reasonable fix.

Elena, I'd push back on the last step. If an ad set is stuck in learning because it's short on conversions, a seed of interests doesn't add any. It just shrinks the pool the system can spend in, so you're asking for the same volume from fewer people, which usually makes the stall longer.

Say a store gets around 40 purchases a month. Broad with a purchase goal can struggle to exit learning at that volume, but I'd look at the event before touching audiences. Optimizing for add to cart, or raising the daily budget enough to clear learning, usually gives the algorithm more signal per dollar. Consolidating two or three similar ad sets into one helps too, since every split starves the others.

Your one-week test is a good one. A stuck ad set is a budget and event problem first, and only becomes an audience problem once those are sorted.

The 40-purchase case shows why the event matters most. Meta's learning guidance is roughly 50 optimization events per ad set in a week, and at 40 purchases a month you're closer to 10 a week. A single purchase-optimized ad set will rarely exit learning at that volume, however you tune the audience.

A simple rule: if an ad set can't plausibly hit about 50 of its target event a week, leave the audience alone. Either move the optimization event up the funnel or merge it into a bigger ad set. Audience changes don't create new events, they just spread the ones you already have across more or fewer buckets.

Agreed, and a lot of people miss that Meta's learning phase is counted per ad set. A store doing 40 purchases while running five ad sets is really running five starved tests. Merging them is often a bigger lever than any audience tweak, because the consolidated ad set can finally clear the threshold.

Here's the rule I'd use: only split into separate ad sets when each one can realistically hit about 50 events a week on its own. If it can't, run the variations as separate ads inside a single ad set. They share that ad set's learning, so you still get your creative comparison without starving any of them.

Keeping variations as ads in one ad set protects the learning, but it doesn't give you a fair creative test. Delivery tends to pile onto whichever ad gets early clicks, so after a few days one creative can have most of the spend while the other has barely run. That means the early CPA gap is partly just who got the impressions. If you need a clean answer on creative, the A/B test tool is built for that, though it needs its own budget. Otherwise, treat in-ad-set results as directional and only compare cost per purchase once each ad has a reasonable number of conversions.

Marco's right that in-ad-set results are directional, and it mostly comes down to how much spend each ad gets. Whether the early gap means anything depends on daily volume. At a store doing 40 purchases a month, neither ad will pick up enough conversions to separate them for weeks, so I'd hold off on reading anything into it.