I audited a £140k-a-month Meta account last month that had 47 active ad sets. The agency running it called this “comprehensive coverage.” Three of those ad sets were spending 71% of the budget. The other 44 were eating fees and confusing the algorithm. That account is not unusual. That account is the median.
Campaign structure isn’t dead. But almost everything most agencies think of as campaign structure is now overhead. The job has shifted from designing the account to designing the inputs the model gets to optimise against, and very few agencies have caught up.
The simplification trend isn’t a trend. It’s the destination
Advantage+, audience consolidation, broad targeting, lift in CBO over ABO. Meta has been pushing the same direction for four years. The official line is that the platform is “simplifying” for advertiser ease. The real reason is that simpler structures let the model (first the old ranking system, now Andromeda) actually do its job.
Every ad set you add is a wall the model has to work around. Budget can’t flow across ad sets in real time. Bids are pinned at the ad-set level. Audience signals collide with each other. A 12-ad-set campaign isn’t 12 chances to find a winner. It’s 12 mini-auctions, each starved of the signal density the model needs to optimise properly.
The brands seeing the biggest gains in 2026 are running one Advantage+ Shopping campaign and one Advantage+ App or Sales campaign, with a small handful of testing campaigns sitting alongside. Not three. Not seven. One or two production campaigns with a wide creative pool. That’s the destination.
Does segmentation still matter? Sometimes, and rarely where you think
Geo segmentation still matters. If you sell into the UK, US, and Australia, you almost certainly need separate campaigns. Currency, language, shipping reality, and demand patterns are different enough that the model can’t bridge them. I’d split by major market every time.
Funnel segmentation is where most agencies are still wasting structure. Separate prospecting and retargeting campaigns made sense when the system needed audience filters to decide who to show what. Andromeda models the funnel position of every user implicitly. It knows who’s been on your site, who’s converted, who’s lapsed. A separate “retargeting” ad set isn’t telling the system anything it doesn’t already know. It’s just constraining its budget allocation.
Audience-based segmentation (interests, lookalikes, demographics) is almost entirely overhead in 2026. The exception is a deliberate testing campaign to validate a new audience hypothesis. The rule is broad with no exclusions beyond suppression lists.
The simplest litmus test: if I removed an ad set from your account, would the model lose information it can’t recover? If no, the ad set is structure for the sake of structure. Most ad sets don’t pass that test.
Budget distribution strategies under AI optimisation
This is where the real money is being left behind. The traditional model (split budget evenly across ad sets, then optimise toward winners) concedes the optimisation job to the human. The Advantage+ model concedes it to the algorithm, which is better at it than any human team I’ve ever worked with.
In direct answer to the question most founders are asking (how should I split my Meta budget?), the honest 50-word version: put roughly 70–80% into a single Advantage+ Shopping campaign with a wide creative pool. Allocate 10–15% to a structured testing campaign for new creative and audience hypotheses. Reserve 5–10% for retargeting only where you can prove incrementality. Resist the urge to slice further.
The 70% Advantage+ allocation feels uncomfortable to most marketers, because it’s a lot of money handed to an opaque system. But the data on it is consistent across my portfolio. Accounts that push above 70% in a single Advantage+ campaign see better blended CPA than accounts that split across three or four “balanced” campaigns. Below 50% Advantage+, you’re paying a complexity tax with no offsetting benefit.
Testing vs scaling: not the same campaign anymore
The mistake I see most often is treating one Advantage+ campaign as both testbed and engine. Putting a new creative directly into the production campaign and “letting the system figure it out” is the lazy version of testing. The system can figure it out, but it does so by spending real money on unproven creative, usually at the cost of fatigued winners that get under-served.
A proper structure separates these two jobs. The scaling campaign is the one that wants the cleanest signal possible: well-tested creative, broad targeting, large budget pool, minimal change. The testing campaign is where you ship volume, accept worse short-term CPA, and look for the next winner. Graduating an ad from testing into scaling should be a deliberate decision, not a default.
I run testing at roughly 10–15% of total spend on most accounts. The CPA in the testing campaign is typically 30–40% worse than the scaling campaign. That gap is the cost of finding next month’s winners. Brands that try to make testing pay for itself in real time will discover, six months later, that they have no new winners to ship.
The other side: scaling is not the same as “spend more.” Most scaling failures I see are budget increases on accounts that didn’t have the creative pool to absorb the extra spend. The model spent the new budget on the same fatiguing creative, performance collapsed, the brand pulled back, blamed the platform. The platform wasn’t the problem. The pool was.
Where this lands
Campaign structure isn’t dead. It’s just been radically simplified, and the agencies billing you for the old version are billing you for theatre. One or two production campaigns. A small testing campaign. Geo splits where the markets demand it. Everything else is overhead.
If your account has more than ten active ad sets and you can’t explain what unique signal each one is contributing, your structure is your problem, not your creative, not your bid, not your audience. Strip it back. The model will thank you in CPA.
If you want a second pair of eyes on whether your account structure is fighting the system, I’m easy to reach.
