Meta Advantage+ placements: let them all run, or cut Audience Network and Reels?

Advantage+ placements default to everything: Feed, Stories, Reels, Audience Network, Messenger. Whether to cut any of them is a fair fight. Take a tutoring business running a lead form. Audience Network can bring cheap form fills that never pick up the phone, while Reels might be where the best leads actually come from. So cutting placements up front is guesswork.

My take is to let delivery settle for a couple of weeks, then read the placement breakdown on cost per qualified lead, not cost per form fill. Exclude Audience Network only if it's clearly worse on lead quality. Excluding Reels or Stories is rarely the right move, since that's often where the cheap reach sits.

Has anyone seen exclusions hold up once volume scales, or do they just push the spend somewhere worse?

A placement breakdown doesn't tell you much until each placement has enough qualified leads behind it. For a tutoring form that's usually 30 or more per placement, and Audience Network rarely gets there quickly. So I'd keep everything open until the sample is big enough, then exclude a placement only if its cost per qualified lead stays clearly worse over a few weeks. After that, judge the exclusion by blended cost per qualified lead across the whole ad set, not by how the excluded placement looks on its own. Pulling Audience Network can look like a win while the same budget pushes Feed costs up.

Agreed on the blended view. The thing to watch is that with Advantage+ the audience shifts as delivery moves, so Meta's placement breakdown often isn't comparing the same people in each placement. A placement can look cheap simply because it got the easier users. Comparing cost per qualified lead over a fixed window with the same creative usually tells you more than the raw split does.

I don't think a fixed window gets you there. Same creative over the same dates still doesn't hold the audience constant, because Advantage+ keeps re-deciding who sees the ad. The window controls for time, not for people.

If an exclusion is going to cost a client real budget, I'd run it as a split test instead: one ad set with the placement excluded, one left open, same audience settings and budget. Then compare cost per qualified lead. It's more setup, but you end up showing the client an actual comparison instead of a guess.

I'm not sure the two-ad-set setup holds up. If both ad sets target the same people, they bid against each other in the same auctions, so the open one can grab the cheapest users and make the excluded one look worse for reasons that have nothing to do with placement. Meta's own A/B test splits the audience at the user level, which gets you much closer to a clean comparison. Two separate ad sets with matching settings is only a rough approximation. Still, the client should see a real test rather than a guess.

The user-level split is the part that matters. Meta's experiment tool can use placements as the variable and tells you whether the gap is meaningful, which two ad sets can't do. The catch is that a new ad set restarts learning, so launch both on the same day and don't touch budgets or creative mid-test. Otherwise the ad set that got a few extra days of delivery skews the comparison before it even starts. I'm not sure what minimum budget the experiment tool requires, so check that in the interface first.

I'm not sure the experiment tool lets you vary placements while leaving the rest of the ad set alone. From what I know, once placements are the variable, each arm usually needs its placements set explicitly. That means the open arm stops being Advantage+ placements, so you'd be comparing two manual placement lists, not the setup the client is actually running. The result then says little about whether dropping Audience Network from the default helps. I could be wrong about how the current interface handles this, so check the setup screen before promising the client anything.

There's also a lag problem with a tutoring form. Qualified status often arrives days after the form fill, so a test read at the end of the platform's reporting window will undercount whichever placement converts slowly. Score the arms on leads qualified by a fixed cutoff date, not on whatever has come in by the time the test ends.

Agreed on the lag, and it hits tutoring harder than you'd expect. Qualified status often shows up days or weeks after the form fill, so a slow-converting placement gets scored as a loser just because its leads haven't been marked yet. Lock the cutoff date, then re-pull the qualified counts once the last cohort has had its full follow-up window.