The Meta campaign structure working for us in October 2026

Every few months someone asks me what the right Meta structure is, as if there were one answer and the rest of us were hiding it. What follows is the structure running in one of our accounts right now, a consumer business in India that sells bookings online, with its numbers for the 30 days from 16 September to 15 October 2026. The last few days of that window will still move a little as Meta finishes attributing late purchases.

What does the structure actually look like?

Two sales campaigns, each with a single ad set, both optimised to the purchase event on the pixel and both sending people to the same landing page.

The first is the scaling campaign. It holds ads that have already proven they can sell. Over the 30 days it spent ₹33,782, bought 154 purchases at ₹219 each, and returned ₹3.32 lakh in purchase value, a ROAS of 9.84. Frequency sat at 2.08, which is comfortable for a month of steady spend.

The second is the testing campaign. Its only job is to find the ad that replaces the current winner before that winner tires. It spent ₹15,839, bought 19 purchases at ₹834 each, and returned a ROAS of 2.70. On its own that looks weak. It is not meant to be judged on its own.

Taken together the two campaigns spent ₹49,621 for 173 purchases, a blended cost of ₹287 per purchase and a ROAS of 7.56.

Beside them sits a video-view campaign optimised for ThruPlay. It spent ₹2,990 for roughly 64,000 ThruPlays, about five paise each, and its viewers become a warm audience the sales campaigns can find again.

CBO or ABO: does it still matter?

Less than the argument suggests. When a campaign holds one ad set, the campaign budget and the ad set budget are the same pot of money. Turning on Advantage campaign budget moves nothing, because there is nowhere else for it to go.

The allocation that decides results happens one level down, between ads. The scaling ad set held 19 ads. In this window one video took every rupee of spend. The other 18, including six statics and four carousels, received nothing. That is Meta choosing, and it chose well: that one ad bought all 154 purchases.

So the useful separation is between campaigns with different jobs, not between budget settings. Keeping scaling and testing apart means a new ad cannot quietly drain money from a proven one, and a proven ad cannot starve every new idea of delivery.

How do we run the testing campaign, and how many ads per ad set?

The testing ad set is crowded on purpose. It put money behind 32 ads in 30 days, roughly ₹530 a day in total. Spend was lopsided, as it always is: the top ad took ₹4,864 and half of the ads got under ₹230 each.

We read it in three passes. First, which ads Meta keeps choosing to spend on, because that is its own forecast of what will convert. Second, cost per add to cart and cost per checkout, which give a signal long before purchases do. Third, purchases, but only once an ad has spent enough for a purchase to mean something.

A few ads look promising this month on small spend. One bought four purchases on ₹560 at a ROAS of 6.39. Another bought two on ₹226. Neither is proven. Both are now candidates to move into the scaling campaign as fresh ads, and that move is how the structure renews itself.

The striking result is format. Across both campaigns, every static ad received zero spend in this window. Short vertical video won the auction every time. We still make statics, because one month is not a law, but we no longer make them first.

Does Advantage+ audience work for sales?

For this account, yes. Both sales campaigns run on Advantage+ audience, using interests as suggestions rather than hard limits, with location and age as the only real constraints. Meta treats the suggestions as a starting point and goes beyond them when it finds buyers elsewhere.

The creative does the targeting now. An ad about a team outing finds people who organise team outings. An ad about a discount code finds people who respond to discount codes. If your creative all says the same thing, a broad audience will find the same people over and over.

Do you still need a separate retargeting campaign?

We tested one. An ad set aimed only at people who had added to cart or started checkout spent ₹257 and bought nothing. The broad sales campaigns were already reaching those people, and at that audience size there was not enough room to deliver.

What does work is filling the pool cheaply at the top. The video-view campaign costs a few thousand rupees a month and gives the sales campaigns a large, recently engaged audience to find. That is a better use of money than a narrow retargeting ad set fighting the main campaign for the same few hundred people.

If you sell bookings where the conversion happens later and offline, the measurement side matters as much as the structure. I have written separately about measuring ads when the booking happens at the venue, and about what AI search changes for experience brands.

What would make us change it?

Two signals. If the scaling ad’s frequency climbs past about 3 while cost per purchase rises, the winner is tiring and the best testing candidate gets promoted early. If the testing campaign goes a full month without producing a candidate, the problem is the creative brief, not the structure, and we go back to the angles before we touch the settings.

Nothing here is permanent. It is what one account is doing in October 2026, written down with the numbers, so you can test it against your own.