Meta Reels placement: exclude it from prospecting or judge it on purchases?

A lot of advertisers cut Reels from prospecting because the CPMs look cheap and the clicks look weak. Take a furniture store on $40 a day. Exclude Reels and Meta just pushes that spend into Feed and Stories, where CPMs usually climb to match. So the cheap impressions you were avoiding often reach the same people anyway, just at a higher price.

I'd leave Reels in and judge it on cost per purchase after a fair learning window, not CTR. Reels tends to produce softer clicks, so a click-based read punishes it unfairly. The case for excluding it is narrower: you're short on vertical creative, and Reels is eating budget on ads that don't suit your brand. Has anyone seen exclusions actually lower blended CPA on a small account?

Judging Reels on purchases is the right call, but be careful which purchases you count. Reels inflates view-through conversions more than most placements, since people watch a lot and then buy later without ever clicking. If you compare placements on 7-day click only, Reels stops looking better than it really is.

At $40 a day, though, volume is the bigger problem. A single placement might get two or three purchases a week, so one week's CPA is mostly noise. Adding the {{placement}} dynamic parameter to your URLs lets GA4 split Reels out, and then you can check Meta's numbers against it.

Agreed on the 7-day click window, but at $40 a day I'd be careful about how you isolate Reels. If you split it into its own ad set to measure it, you're probably creating a learning-phase problem. Two or three purchases a week won't get a small ad set out of learning, so you end up judging a campaign that never settled. Reading the placement breakdown inside one prospecting ad set avoids that, since nothing gets reset. The breakdown uses Meta's attribution though, so it carries the same view-through inflation you mentioned, which is where the GA4 split from {{placement}} earns its place as a cross-check. Read the two side by side and expect a rough answer over two weeks, not one. A client who wants a clean verdict on this budget will push you toward a conclusion the data can't support yet.

Are you splitting Reels out as its own ad set, or reading it from the breakdown inside one?

I'm not sure the breakdown is the safer option. It tells you what Meta spent on Reels and what it credited to Reels, but not what happens to blended CPA if Reels goes away. Inside a mixed ad set, Meta keeps shifting money between placements, so a Reels number there is partly Reels and partly whatever Feed picked up. The question in this thread is really about exclusion, and only an exclusion test answers that directly, even a noisy two-week one.

I don't think an exclusion test is the only way to answer this. At $40 a day you'd run it for two weeks on two or three purchases a week, so any shift in blended CPA is mostly noise. You can't tell a real drop from a bad week. Excluding Reels also doesn't isolate Reels, because the budget just moves into Feed and Stories, which means you're measuring a different placement mix, not Reels missing. The placement breakdown is biased, sure, but it comes from the same campaign and creative. For a budget this size I'd read the breakdown for direction and watch blended CPA over a longer window before paying for a test you can't read cleanly.

Iris has the sample problem right, and Mira's right that the breakdown can't answer the exclusion question. My rule of thumb: before you touch anything, work out how many purchases you'd need per arm to tell a 20% CPA change from noise. If you can't get there in about four weeks at your budget, skip the exclusion test. Put the effort into something you control, which is vertical creative for Reels, and use the placement breakdown for direction only. A test you can't read is just a more expensive way to guess.

Your power check is the right test, and it can be even simpler. Work out how many purchases a 20% CPA swing would take to detect, then see whether your budget actually produces that many in four weeks. For a $40 a day furniture store the answer is usually no, so the exclusion test is off the table before it starts.

Agreed, and it's worse than it sounds. If you treat purchases as Poisson counts, spotting a 20% CPA gap at 80% power takes roughly 400 purchases per arm. At $40 a day that's closer to a year of data than four weeks.

The 400 figure assumes you want to catch a 20% gap, and that's the part I'd revisit. Sample size drops fast as the gap grows: a 50% CPA difference needs about a third of those purchases, and a 2x difference needs about a tenth. So if Reels is sitting at double the blended CPA after a few weeks, you don't need a formal test to act on it. What you do need is a stop rule written down before you look at the numbers. What gap would actually change what you do with the budget?

Sample size scales with the square of the log ratio, not the raw gap, so the numbers come out a bit different. A 50% gap needs about a fifth of the 400 purchases per arm, not a third, and a 2x gap needs about a fourteenth, so roughly 30 per arm. Your stop rule still holds, and the 2x case is even easier to reach than you said.