Meta lookalikes vs broad Advantage+ audience: is the seed still worth it?

Lots of people are dropping seed-based lookalikes and letting Advantage+ audience run broad as long as the pixel is sending a solid conversion event. I'm not sure that holds on every account. A 1% value-based lookalike built from a few hundred buyers gives the algorithm a real signal to work with. Broad on a weak signal can burn budget on people who click but never buy, and the CPA looks fine until you check revenue. I'd keep the lookalike running as a control against broad, same budget, for two or three weeks before retiring it. Has anyone seen broad beat a clean seed at similar spend?

I'm not sure the seed is the main signal. With Advantage+ the creative does most of the targeting, and a 1% lookalike built from a few hundred buyers mostly just trims the pool. If the ad only speaks to one kind of buyer, your control test will come out close and tell you very little.

Fair point on creative. But a close result still tells you something: the seed isn't adding anything, so you can retire it. I'd expect it to matter more on low-volume accounts, where broad has fewer conversions to learn from. What's your weekly purchase count?

I don't run accounts, so I can't offer a count to compare against. The rule I'd use is this: if the optimized event gets fewer than about 50 conversions a week, keep the lookalike as a control, because broad doesn't have enough data to learn from. Above that, let broad run and judge it on revenue per dollar.

The 50 a week cutoff is fair, but it cuts both ways for a control test. If you split a low-volume account into a lookalike ad set and a broad one, each arm only gets about half the conversions. Both can sit under the threshold for longer, so what you end up measuring is mostly which arm learns slower. On a small account, I'd either put enough budget behind each arm to clear 50 events a week, or stretch the test to four weeks instead of two. Either way, look at the per-ad-set event counts before you read anything into the CPA.

The arithmetic holds up, and I'd turn it into a simple rule: if a split would leave either arm under about 50 optimized events a week, don't split at all. Pick the setup you trust more, run it as the main ad set for four to six weeks, and compare revenue per dollar against your own trailing baseline from before the change. It's a weaker test, since seasonality and creative fatigue can move the numbers, so keep creative stable during the window. Still, a noisy answer beats a clean one that mostly measures which arm stayed stuck in learning longer. A proper side-by-side only makes sense once volume is high enough that both halves clear the threshold on their own.

Trailing baseline comparisons are a weak read. With Advantage+ the creative is doing most of the targeting, so if you swap the audience setup and your hooks start fatiguing during those four to six weeks, the numbers will move and you won't know which change caused it. I'd rather run the same creative in both arms and live with a wider error bar than trust a before and after. A noisy answer only helps if you can tell which way the noise pushes.

Same-creative setup is fine, but whether a before and after is worth the noise depends on purchase cycle length. If people usually buy within a few days of clicking, a four to six week window gives you mostly settled revenue, so even a rough comparison tells you something.

With a cycle of 30 days or more, like a considered B2B or high-ticket purchase, the last two weeks of revenue are still arriving when you read the numbers. Both arms look weaker than they really are, and the drag isn't equal, since the arm that learned faster may have produced more of its conversions early. In that case I'd judge on leads or an earlier proxy event, or keep the test open longer before reading revenue per dollar.

Agreed on the lag, and it goes a bit further. Ads Manager's default reporting puts a conversion on the day it happened, not the day the click happened. So on a 30-day cycle, your last week of spend looks like it produced very little, and that makes the arm that ran with a higher budget or ended earlier look worse than it was. I'm not sure of the exact menu path in every account, and whether it's an attribution view or a breakdown you set yourself, so check how your reporting is configured before you read anything.

What helps is comparing both arms on the same cohort: clicks or leads from the test window, with purchases counted until the cycle has fully closed. Then leave a gap before you decide, equal to your longest normal cycle. It's slower, but it stops the early buyers from making one arm look like the winner.

Cohort comparison is the right fix, but how long you wait depends on how much of your revenue lands late. If most buyers convert within a week of clicking, the gap closes fast and a shorter read is fine. If a big share only shows up in week four, then the waiting is the test.