Meta Advantage+ placements: exclude the weak ones or let the algorithm decide?

Say you run a furniture store on Meta with a decent but not huge budget. Advantage+ placements is on by default, and the breakdown shows a chunk of spend going to Audience Network and Right Column, where CTR looks fine but purchases are thin. The usual move is to exclude those by hand and keep Feed and Reels. I'm not sure that's right for everyone. Excluding placements shrinks the auction, so on a smaller budget you can end up paying more for the same Feed impressions. I'd compare cost per purchase by placement over at least two weeks, and only cut one that's clearly losing once it has enough conversions to judge. Has anyone had a placement that looked cheap on clicks but actually converted?

I'd disagree on the auction point, at least for a furniture store. Feed and Reels carry most of the volume anyway, so cutting Audience Network and Right Column doesn't leave you bidding into a thin pool. The real problem is the one you raised: CTR on those placements gets inflated by accidental taps, which is why it looks fine while purchases stay thin. I agree on judging by cost per purchase. If the numbers are close, a small CPM increase on Feed usually costs less than the spend going to clicks that never turn into sales.

Sasha's right that accidental taps inflate CTR on those placements. I still wouldn't call a higher Feed CPM cheap until you see the numbers, and it comes down to purchase volume. If the account gets dozens of purchases a week, cost per purchase by placement settles within a couple of weeks and the cut is easy to make. If it's only a handful, every placement's cost per purchase is mostly noise, so excluding the weak ones just swaps one guess for another. In that case I'd leave Advantage+ on for a few weeks and see whether Feed alone is actually buying more per dollar before cutting anything.

Purchase volume does decide it, but before anyone compares placements there's a measurement trap. Cost per purchase by placement depends on your attribution setting, and view-through credit can flatter Audience Network a lot. If 1-day view is being counted, that placement can look much better or worse than it really is. Pick one window, usually 7-day click, and judge every placement on that same window.

Reporting is the other problem. If you cut two placements, blended CPA can climb for a few weeks even when Feed is more efficient, because volume shifts and the learning phase resets. If the client isn't warned, they'll read that rise as the cut failing.

Which attribution window are you judging placements on?

Look at the attribution default before you trust the placement table. Plenty of accounts still have 1-day view included, and switching to 7-day click can reshuffle the ranking a lot. With furniture especially, a view credited to Audience Network is often just someone who saw the sofa there and ended up buying later through a Feed ad.

Agreed on view-through. With furniture the path is often phone browsing, then a desktop purchase a week later, so 1-day view credit to Audience Network is mostly noise.

Before you exclude the whole placement, though, break Audience Network down by publisher app or site. In most accounts a couple of apps carry the purchases and the rest just soak up impressions. If the losers are a short, clear list, block those at the account level and keep the good ones running. A blunt cut throws away whatever was actually working.

Run that test for at least two weeks on the same 7-day click window, and warn the client that blended CPA may bounce while the budget shifts. That dip is the reallocation settling, not the cut failing.

Splitting Audience Network by app as the first step is the wrong move, I think, and Theo's volume point matters even more here. On a modest furniture budget, Audience Network spreads across dozens of apps and sites, and most of them see only a purchase or two a month. One sale can make an app look like a winner, so a short "good apps" list could easily be noise. Check purchase counts per app before you block anything. If none of them have enough conversions to judge, a placement-level cut on the same 7-day click window is the more honest test.