Meta Ads Campaign Structure in 2026: Testing, Validation, Scaling

Most Meta accounts we take over have the same disease. There is one campaign. New creative gets uploaded into it, next to the ad that has been carrying the account for four months. Spend collapses onto that one ad, the new creative never gets a fair read, and the founder concludes "creative testing doesn't work for us."

It works. The structure was wrong.

This is the structure we run instead: three stages, and a creative has to earn its way through each one. It is deliberately boring. Boring is the point — the interesting decisions belong in the creative, not in the campaign settings.

Table of Contents

• The one-line version
• Stage 1 — Testing
• Stage 2 — Validation
• Stage 3 — Scaling
• Why this works: EAR
• What breaks it
• Which version of this you should run

Key Takeaways

The one-line version

Spend concentrates on creative we already know survives.

Everything below is machinery for making that sentence true. A creative that has never been tested cannot take budget from one that has been proven, because the two never sit in the same campaign.

Stage 1 — Testing

Small daily budget per ad set. ABO, with 2 to 4 creatives per ad set.

Three rules that surprise people:

Ads are never switched off. We let Meta sort them. If you kill an ad on day two because it hasn't converted, you are making a decision on a sample that cannot support it — and you are doing it on Meta's behalf, when Meta is better at it than you are.

Some die, some come back. We don't chase them. An ad that flatlined for a week and then started spending again is a normal event, not a mystery to solve. Chasing it — bumping budgets, editing copy, restarting — resets the learning that would have told you something.

The budget rule is mechanical. Beat target CPA, the ad set gets +10%. Miss target CPA, it gets −10%.

That last one is the whole discipline. It is small enough not to shock delivery, it compounds in the right direction, and it removes the daily argument about whether to give something "one more day." The number decides. You apply it.

Stage 2 — Validation

This is the stage almost nobody runs, and it is why most accounts have a scaling campaign full of creative that never really worked.

Validation gets roughly 20% of the scaling budget, in a CBO or ASC campaign at scaling-level competition.

The question it exists to answer: can this creative keep up with the creative already in Scaling?

A winner in Testing beat target CPA in a small, quiet ad set against low competition. That is not the same as beating it while sitting next to your best performer with real budget behind it. Validation is where you find out — and the cost of finding out is 20% of budget rather than your whole scaling campaign.

Ads are duplicated in — not post IDs. Bring the ad itself across rather than plugging the post ID into a new ad. Post IDs carry social proof forward, which sounds like an advantage and is exactly why it corrupts the read: you are no longer testing the creative, you are testing the creative plus its accumulated engagement.

Holds performance? Off here, duplicated into Scaling.

Stage 3 — Scaling

Simple CBO setup. Every proven winner lives here. Nothing enters that hasn't already earned it.

That is genuinely all there is to it, and the simplicity is load-bearing. The temptation is to make Scaling clever — bid caps, ad set splits, audience carve-outs. Resist it. Scaling's job is to spend efficiently on creative whose performance is no longer in question. Complexity here means you are still testing in the campaign that is not supposed to test anything.

Winners get renamed on the way through, so six weeks later you can still tell which creatives arrived by earning it.

Why this works: EAR

Meta's Estimated Action Rate decides where and to whom your ads serve. It is driven by three things:

• your bid
• your creative
• your URL destination

That third input is the one founders forget, and it is why we treat the landing page as part of the ad rather than as something the ad points at. Two identical creatives pointing at two different pages do not have the same EAR, and so do not get the same delivery. We covered how we build Meta ads landing pages with Claude separately; the short version is that a slow or off-message page suppresses delivery before a human ever judges your offer.

EAR is also why one ad takes all the budget and another takes none — inside a single ad set, Meta is not splitting spend fairly, it is backing its own estimate. The full mechanic is in what Estimated Action Rate actually is.

What breaks it

Testing and scaling in the same campaign. The proven ad wins the auction against your new creative every time, the new creative never gets enough delivery to be judged, and you conclude the creative is bad. It was starved, not bad.

Killing ads early. Covered above, and the most common self-inflicted wound in the account.

Editing a winner. Editing an ad in Scaling can reset what it has learned. Duplicate it and test the change in Validation instead.

Promoting on a good day. One strong day is noise. The promotion criterion is holding performance in Validation, not a screenshot of a good Tuesday.

Structure churn. Rebuilding the account every three weeks means nothing accumulates enough history to be judged. Pick a structure and let it run.

Which version of this you should run

Here is the honest part: the right setup depends on your spend, your creative volume, and how your account actually behaves. Three stages is the principle. The specific campaign layout that expresses it is not one-size-fits-all — we run four different ones depending on the account, and the choice is mostly determined by budget.

That breakdown is its own post. Meta ads structure by spend level covers all four, including the low-budget version where testing and scaling share a campaign, and the ABO-only setup that shouldn't work and sometimes does.

If you want the mechanics of the testing stage specifically — how much to put behind an ad set, and why we don't switch ads off — that is in the Meta ads testing rules we use.

And if you would rather not apply the ±10% rule by hand every morning, we run this whole structure with Claude — classification, promotion decisions and the weekly diagnosis, prompts included.

Branva runs paid acquisition for Shopify brands as part of one coordinated system with email, SEO and CRO — see Brooklyn Transfers for what that looked like over a quarter. Book a strategy call and we'll audit your current structure on the call.