Meta creative testing: ABO with fixed budgets, or CBO from day one?

People get surprised when one ad takes 90% of the spend within two days of a CBO launch with five new ads, but that's just how it works. For testing, I'd run ABO with a fixed budget per ad set, because each concept gets a fair share before the algorithm picks a favourite. CBO makes more sense once you already know your winners and want to scale them.

The usual objection is that ABO keeps spending on weak ads for longer. That's true, but the fix is to kill losers faster, not to let one ad hog the budget. At what daily spend do you stop testing in ABO and move to CBO?

ABO doesn't actually give each concept a fair share of anything. It caps each ad set, and if you put one creative in each, you're running five separate tests with no shared read. The bigger issue is that "kill losers faster" only works if the losers have enough conversions to judge. Meta's guidance is roughly 50 optimization events per ad set per week to exit learning. At two purchases a day, a "loser" on day three has about six conversions, so killing it is mostly noise. On a low-volume account, test against a higher-volume event like add to cart or landing page view, or accept longer runs, before trusting any kill call.

As for a daily spend number, I wouldn't use one. Weekly conversions per ad set is the better yardstick. Once a concept has enough volume to read, that's when to move it into CBO to scale.

Agreed on the volume problem. Here's a rule you can actually apply: don't kill a creative until it has about 15 of your target event and has run at least three to four full days, so one good afternoon doesn't decide it for you. At two purchases a day, that works out to roughly a week per creative, which is why I'd test on add to cart or landing page view if purchases are that thin. Tying the call to event counts instead of a calendar date or a spend number means a small account and a big one follow the same logic.

The 15-event floor makes sense, and the counting side trips people up too. Meta's conversion numbers keep filling in for a day or two after the click, especially on 7-day click attribution, so a creative that launched Tuesday looks worse on Wednesday than it will by Friday. Compare creatives at the same age, or only read ones that have had seven full days. Otherwise the newest concept gets killed just for being new.

Same-age comparison is the right fix, and I'd also put the kill rule in writing before launch. Something like "kill when a creative has 15 events and its cost per event is above 1.5x target CPA." Without it, every pause turns into an argument with the client, who remembers the one ad that almost worked. A fixed rule is easy to defend in the report, and it keeps you from killing a creative at 14 events just because it looked bad on day four.

Iris, the 1.5x rule makes sense, but 15 events is too small a sample for it. If a creative's true CPA is right on target, you'd expect about 15 events from your spend. Seeing 10 or fewer, which is what a 1.5x CPA looks like, happens about 12% of the time. So roughly one in eight decent creatives gets killed by the rule alone.

Raising the floor to 30 events drops that to about 4%, and it's still small enough to test in a week or so on a mid-volume account. If your volume can't reach 30, loosen the threshold to 2x target CPA instead of tightening the event count. Either way, put the math in the written rule so the client can see why a kill happened, not just that it did.

Your math holds up, but I'd tie the floor to how much budget the creative is eating. A 4% false-kill rate hurts more when the ad is carrying a small ad set. If a creative gets 5% of spend while three others run the account, a wrong kill costs little, and waiting for 30 events just means paying for a weak ad longer. So I'd set the floor by share of spend rather than one fixed number for every test.

Tying the floor to spend share doesn't hold up, because in CBO that share is Meta's call. It starves the ads it already thinks are weak, so your rule would loosen exactly when the data is thinnest. I'd keep the event floor fixed and let the budget decide how long you can afford to wait.

Agreed that spend share can't be the trigger in CBO. I'd also put a ceiling on the wait, measured in money. Rough rule: spend about 30x target CPA on a creative. If it hasn't hit 30 events by then, don't leave it running on hope. Mark it inconclusive, pause it, and log why. That covers both failure modes: a weak ad doesn't sit in CBO for three weeks, and a decent ad doesn't get killed at 14 events.

It works because the event floor stays fixed, so the only open question is how much you're willing to pay for an answer. Writing the cap into the test plan helps with the client too. Nobody has to argue about the one ad that almost worked, because the cost of an inconclusive result was agreed before launch.