Campaign structure in Facebook Ads determines how much data the algorithm gets, how fast it exits the learning phase and how predictable scaling will be. In 2026 Meta has pushed automation to the point where manual targeting is barely needed, and the limits on the number of ads are starting to come off. Here is how to build campaigns now: ABO vs CBO, when to use Advantage+, how to test creatives without burning budget, and how to scale without a CPA collapse.
Three campaign levels and what is decided where
Campaign — objective (leads, sales, traffic) and budget distribution method. Ad set — audience, placements, GEO, schedule and, under ABO, budget. Ad — creative, copy, link. The algorithm learns at the ad set level: it needs around 50 target events in 7 days to exit the learning phase. Anything that prevents reaching those 50 — fragmented audiences, small budgets, frequent edits — keeps the ad set in learning and makes results unstable.
ABO or CBO: the difference and when to use each
ABO (Ad Set Budget Optimization) — budget is set on each ad set separately. You control how much each audience or creative spends. CBO (Campaign Budget Optimization, labelled Advantage Campaign Budget) — budget at the campaign level; the algorithm decides which ad set gets more.
ABO for testing. When you need a fair comparison of creatives or audiences, each needs an equal chance to spend. In CBO the algorithm picks a “favourite” within hours and starves the rest — not because they are worse, but because it had little data. The classic test structure: one ABO campaign, 3–5 ad sets varying one variable (creative or audience), budget per ad set = 1–2 expected CPAs per day.
CBO for scaling. Once winners are found, put them into a CBO campaign with a large budget. The algorithm reallocates money towards what works now, faster than you would. Minimum ad set budgets in CBO are insurance against the algorithm completely choking one of them.
Rule of thumb: test in ABO, scale in CBO. Mixing testing and scaling in one campaign is a reliable way to get neither.
Advantage+ in 2026: what it is and when to use it
Advantage+ is Meta’s umbrella for automation. The main pieces:
- Advantage+ Sales / Shopping (ASC) — a campaign without manual targeting: broad reach, the algorithm finds buyers itself. Requires a pixel or CAPI with event history. On accounts with accumulated data it often delivers CPA 15–30% below manual campaigns.
- Advantage+ Audience — you set the audience as a “suggestion”, not a constraint. The algorithm starts there and expands.
- Advantage+ Creative — automatic creative variations: placement fitting, image enhancement, background and text generation. Useful for formats, risky for precise offers — check the AI hasn’t rewritten your offer.
- Advantage+ Placements — all placements at once. Optimal for most offers; manual exclusions only make sense with data in hand.
Under the hood since 2025 runs the Andromeda system: it sifts through a huge pool of ad candidates and picks a creative for each user. The key consequence: the algorithm needs many different creatives. Not 3 but 10–20 per ad set, varied in format, angle and opening seconds. This also explains why Meta is testing the removal of the 250-ad cap: the old limit contradicts the new logic.
When Advantage+ doesn’t fit: a fresh pixel with no event history (nothing to learn from), a narrow offer with a precise profile (the AI expands the audience to where it doesn’t convert), a small budget (broad targeting needs more events).
Creative testing: a structure that doesn’t burn budget
Creative is the one variable that makes a multiple difference in 2026. The algorithm picks audiences, placements are automatic, so are bids. So the test structure is built around creatives.
- An ABO campaign with the same objective as your live ones (not “traffic” for cheap clicks).
- 3–5 ad sets, each with one angle (pain, benefit, social proof, demo) and 3–4 creatives of that angle in different formats: static, 9:16 video, carousel.
- Ad set budget sized to get at least 8–10 conversions in 3 days. At $20 CPA that is $60–70 a day.
- Don’t touch it for 72 hours. Any budget change over 20% or a creative swap restarts learning.
- Winner criterion — not CTR or CPC but CPA and click-to-conversion rate. Cheap clicks with poor conversion are the most common way to fool yourself.
- Winners (usually 1–2 out of 15) go to a CBO campaign for scale; losers get switched off; the winner gets 3–5 iterations (different hook, different first frame, different length).
The hook — the first 3 seconds of a video or the first visual element of a static — decides 70% of the result. Test hooks separately from the creative “body”: the same video with five different openings is five creatives.
Scaling: raising budget without a collapse
Vertical — increasing the same campaign’s budget. The 20% rule: no more than 20% per day, otherwise the ad set re-enters learning and CPA jumps. CBO with a large budget absorbs increases more smoothly than ABO.
Horizontal — duplicating the working funnel into new campaigns, audiences, GEOs. A duplicate doesn’t inherit learning, so launch it with the budget the original had at start, not the raised one.
Through creatives — the most reliable path in 2026. Instead of raising budget on one creative, add new variations of the winner to the same CBO campaign: the algorithm distributes the money and doesn’t burn out on one image. Creative fatigue shows as frequency above 2–2.5 and falling CTR at the same CPM.
What to monitor daily
- CPA and ROAS per ad set over 3 and 7 days — not today.
- Frequency — above 2.5 on a cold audience means it’s time to change the creative.
- CPM — a sharp rise with the same targeting signals auction competition or falling account quality.
- Learning status — “Learning limited” means not enough events: merge ad sets or raise budget.
- Gap with the tracker — Facebook attributes by its own model, the network pays on facts. Set up S2S postbacks and CAPI so the algorithm learns on real conversions.
FAQ
How many ads per ad set?
For testing — 3–4 creatives of one angle. For scaling in CBO — 10–20 variations: Andromeda works better with a large pool. Over 50 in one ad set is pointless — some simply won’t get impressions.
ABO or CBO for a beginner?
ABO. It is more transparent: you see where each dollar went and it is easier to understand why a funnel isn’t working. Switch to CBO when you have proven creatives and a budget from $100 a day.
Should I exclude placements?
By default, no. Exclude only with data: if Audience Network delivered 30% of clicks and 0 conversions in a week — turn it off. Without data, exclusions only narrow the auction and raise CPM.
How do I know the campaign exited learning?
The ad set status in Ads Manager changes from “Learning” to “Active”. Usually that is 50 optimisation events in 7 days. If “Learning limited” persists for over a week, the structure isn’t feeding the algorithm data.
Your structures, account screenshots and test results — in the Facebook section of the forum.
