ABO vs. CBO in 2026: Why We Test Inside the CBO and Killed Force-Budgeting

Here's a confession that would have made my younger self wince: we stopped force-budgeting our ad tests, and our accounts got healthier almost overnight.

For years I did it the way everyone teaches it. New creatives go into a fresh ABO ad set, you force a fixed budget onto each one, you let them fight at $20 or $30 a day, and the winners "graduate" up to your scaling campaign. It feels disciplined. It feels like control. And for a long time I'd have argued it was the only sensible way to test.

I don't believe that anymore. So let me walk you through what changed, because the ABO vs CBO Facebook ads debate is usually framed as a religious war when it's really just a question of what you're trying to measure.

The habit I had to break

The old logic goes like this. ABO (ad set budget optimisation) lets you protect each test. You decide that creative A and creative B each get $25 a day, full stop. Nobody starves. Everybody gets a fair go. Then you crown a winner on test ROAS and carry it up to the big campaign.

I used to love that tidiness. The problem showed up the moment those "winners" hit real spend.

I'd find a creative doing a 4.0 at $25 a day. Genuinely promising. Then I'd lift it into a campaign spending $2k a day and it would just sit there. Flat. Sometimes it barely spent at all. I kept getting frustrated, kept blaming the algorithm, kept duplicating campaigns and re-uploading the same ad hoping it would suddenly catch. All I was doing was creating chaos in the account.

It took me embarrassingly long to understand why.

Why ABO winners die at scale

When an ad runs at $20 or $30 a day, it's spending a tiny slice of your account, maybe 1 to 5% of total budget. Meta isn't testing whether that creative can carry your business. It's just picking off the lowest-hanging fruit. The people who already know you. The ones who've seen your other ads five times and were halfway to buying anyway.

So your little test ad looks brilliant. It posts a beautiful ROAS because it's quietly harvesting demand that the rest of your account warmed up.

Then you hand it the big budget, and now it has to go find cold people who've never heard of you and convince them. That's a completely different job. The test never asked it to do that job, so the test never told you whether it could.

Here's the thing. A high test ROAS at low spend isn't a signal of a winner. It's a signal that the ad is good at eating cheap conversions. Those are not the same.

This is why I had to change my own definition of what "winning" even means.

What we actually count as a winning ad now

For a long time I thought a winning ad was simply one with a strong ROAS. I now think that belief held me back more than almost anything else.

A winning ad, in our book, is one that does two things:

• It takes the spend when you give it room to. It scales without falling over.
• It moves the account-level number in the right direction. Either it lets us spend more at the same blended CPA, or it pulls our cost per acquisition down across the whole account.

That's a much higher bar. We're no longer hunting for ads that nudge the account 5 or 10% better. We're hunting for the ones that change the shape of it, the creatives that let us spend meaningfully more without the CPA running away.

You don't find one of those every week. Honestly, for most clients it's maybe one a month. But when you do, it can lift the whole account in a way a hundred tidy little 4.0-at-$25 ads never will.

So if that's the target, you have to test in the environment where it'll actually live. Which is the CBO.

How we test inside the CBO

These days new creatives go straight into the campaign budget optimisation campaign that's already spending. We're not asking "what's the test ROAS." We're asking one question: will this ad set pull spend away from the incumbents?

If we drop a new ad in and the campaign decides to push real budget into it within a few days, that's the account telling us it sees something. If it sits there starved while the existing winners hoover up the budget, that's also the account telling us something. Both are useful. Both are honest in a way that a quarantined $25 test never is.

We give it three to seven days. If it earns spend, we keep building on it. If it doesn't, it gets cut and the next round goes in. No agonising. No emotional attachment to a creative we paid good money to produce.

I'd flag the trap here too, because it's the one that drains accounts quietly. When a new ad won't spend in the CBO, the temptation is to rescue it. Pull it into its own ABO ad set, force a budget onto it, prove the algorithm wrong. I've done it. Picture spending ~$600 on producing a batch, then another ~$300 a day forcing the dud to run for three days, and you're well over a thousand dollars deep on a creative Meta already told you it didn't want. Force-budgeting mostly just funds your own stubbornness.

There's a structural reason CBO testing is cleaner too. Run a pile of separate ABO ad sets and they start bidding against each other in the same auction. You end up paying to compete with yourself. A CBO moves money freely between ad sets instead, so the budget flows to whatever's genuinely working rather than getting trapped in a box you drew around it.

Where ABO still earns its place

I'm not religious about this, and I'd be lying if I said CBO is always the answer. There are two situations where I'll still reach for ABO, and both come down to forcing data when the account can't generate it on its own.

The first is a small or fresh account. If the total budget is genuinely tight, say you've only got room for a few hundred a day and the client wants five things tested, a CBO will simply ignore three of them. It'll back one or two concepts and the rest get no data at all. In that case I'll run ABO with a small fixed floor per ad set, just enough to get a read on each one. Not to scale them. Just to learn.

The second is a low hit rate. If a brand's creative is landing maybe one winner in ten, a CBO will keep dumping budget chasing ads that were never going to make it. There, forcing a controlled minimum per test at least caps the bleed while you figure out the bigger problem, which is usually the creative itself, not the structure.

But notice what both exceptions have in common. They're about generating signal in an account too immature or too thin to produce it naturally. The moment you've got real spend and a few proven winners carrying the load, that scaffolding comes off and you test where the ads will actually compete.

Where to from here

If you take one thing from this, let it be the redefinition, not the campaign type. A winning ad isn't the one with the prettiest test ROAS. It's the one that takes spend and improves the account when it's standing next to your big dogs. Once you judge creatives that way, most of the ABO-vs-CBO argument quietly dissolves, because you stop trusting numbers that were never earned at scale.

So a question worth sitting with: how many of the "winners" in your account right now actually proved themselves at real spend, and how many just got crowned at $25 a day and have been coasting on a reputation they earned in the easy auction?

If you're not sure, that's usually worth an outside look. We run a Signal/Noise Audit that maps your creative history against where the spend actually goes, and it tends to surface pretty fast which of your "winners" are carrying the account and which are just decorating it.