Most ad accounts we audit don't have a budget problem. They have a testing problem. They're A/B testing audiences in 2026 - on platforms that decided who to show ads to two years ago.
The lever moved. It's creative now. And if you test creative in the wrong order, you'll spend half your budget proving things the algorithm already knew.
This is the order we run on every account. We call it creative-first because that's literally the sequence: the ad comes before the media buy, not after.
The hierarchy: hooks, angles, formats, audiences
There are four things you can change about an ad. They are not equal, and they do not deserve equal budget. Test them in this order:
- Hooks - the first 3 seconds, the thumb-stop
- Angles - the argument the ad is making
- Formats - static, UGC, motion, the container
- Audiences - last, and barely
Each level only matters once the level above it is settled. A genius audience strategy can't save a hook nobody stops for. A perfect hook on the wrong angle sells the wrong thing to the right person.
Skip a level and you're optimising noise. The data looks like it means something. It doesn't.
Why audiences come last
Broad targeting plus a strong creative beats narrow targeting plus a weak one - consistently, across every account we run. Meta's Advantage+ and Google's PMax are built to find the buyer. Your job stopped being "who" and became "what makes them stop."
So we don't open with twelve interest stacks. We open with broad, and we let the creative do the targeting. The angle is the audience. A post-partum recovery angle finds new mothers more reliably than a "new mothers 25-40" interest layer ever did.
How we actually run it
Hooks first. Same body, same offer, five openings. We're not looking for a winner yet - we're looking for which openings earn attention at all. Cost per thumb-stop, hold rate, the cheap signals that show up in days, not weeks.
Then angles, on the winning hook. Is this a status play, a fear-of-missing-out play, a "here's the problem you didn't know you had" play? One angle per concept. Never two - you won't know which one worked.
Formats next. Take the winning hook-and-angle and change only the container. Sometimes the UGC version triples the click rate. Sometimes the clean static wins because the product is the story. You can't predict it, which is the whole point of testing it last-but-one.
Audiences dead last, and lightly. By the time we get here, the creative is doing the work and the platform is doing the targeting. We're nudging, not steering.
The kill rule that keeps it honest
A framework without a kill rule is a wish. Ours: nothing scales past 3x its target cost per result. If a creative can't get there, it dies - no sunk-cost rescue attempts, no "let's give it another week."
This is also where BAVai earns its keep. The accounts get scanned every morning at 7am, so a fatiguing winner or a quietly bleeding test gets flagged the day it turns, not in next month's report when the money's already gone. The human decides what to kill. The machine makes sure nothing slips through overnight.
The takeaway
If your testing roadmap is a list of audiences, you're running a 2022 playbook on 2026 platforms. Move the budget up the hierarchy. Settle the hook before you touch the angle, the angle before the format, and let the creative find the audience.
Is your testing actually moving up the hierarchy - or are you just shuffling audiences and calling it optimisation?