First, confirm the concentration
Open the ad set at ad level and read each ad's share of spend and its results over complete days, with the start date of each ad visible. An ad that took most of the spend from launch, beside others launched at the same time, is a different story from an ad that has run for weeks beside newcomers added yesterday. Do the same at ad set level if the campaign uses Advantage+ campaign budget: there the question is which ad sets receive the money, not which ads.
Then look at cost per result for the ads that did receive delivery. If the dominant ad also has the lowest cost per result (CPACost of each conversion.) by a clear margin, concentration is the system working. If the starved ads have too little delivery to judge, the question is whether you need to know about them, and if you do, whether the current structure can ever tell you. How the main test methods compare is set out in the creative testing guide; this page only explains why spend is uneven.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Ads launched together, one took most spend within days | Ad-level spend share and early results | Delivery favouring the ad with the best early predicted results | Accept it, or run a controlled test if you need a fair comparison |
| New ads beside a long-running ad get almost nothing | Start dates; the established ad's history | New ads added where an established ad already wins | Launch new creative in its own ad set or test |
| One ad set in a campaign takes the budget, others get little | Campaign budget setting; ad set cost per result | Campaign budget concentrating spend on one ad set | Expected; use ad set limits or separate budgets where coverage matters |
| Spend concentrates while the ad set is in learning or learning limited | Delivery status; budget against the optimisation event | Ad set in learning with too little budget to explore | Consolidate, raise budget, or choose a more frequent event |
| Ads look nearly identical and one wins arbitrarily | Compare concept, hook and format across the ads | Ads differing too little for the system to tell apart | Vary the concept or hook, not small details |
Fix, verify and prevent each cause
Early predictions favouring one ad
Within an ad set, delivery shifts spend towards the ad it expects to produce the optimisation event most cheaply, and it forms that expectation early, from limited signals. Once one ad pulls ahead, the others rarely gather enough delivery to change the verdict.
Fix: if you are happy to run any of the ads, accept the result; if the comparison matters, move the ads into a controlled A/B TestControlled comparison of two variants. that splits the audience.
Verify: in a test, each variant receives comparable delivery and a winner is declared on the metric you chose.
Prevent: decide before launch whether an ad set is for running ads or for learning which is better, following the method comparison in the creative testing guide.
New ads beside an established winner
An incumbent arrives with a history; a newcomer arrives with none. Added to the same ad set, new creative often gets a trickle of impressions and is judged a failure without a fair hearing.
Fix: launch new concepts in their own ad set or test, then move proven ones into the main ad set.
Verify: the new ads gather enough delivery to show a cost per result you can compare.
Prevent: keep a regular route for new creative that never depends on displacing a winner, as the creative testing guide describes when it turns winners into a creative system.
Campaign budget concentrating on one ad set
Advantage+ campaign budget is documented as sending money to the ad sets with the best opportunities, not splitting evenly. One ad set taking most of it is the feature, not a fault.
Fix: if an ad set must receive spend, for example a market with its own target, give it a minimum ad set spend limit or move it to its own budget.
Verify: the protected ad set reaches its minimum and the campaign total stays within budget.
Prevent: put only comparable ad sets under one campaign budget, as the guide to campaign and ad set budgets recommends. If daily totals also look wrong, see the spending more than the daily budget page.
Learning with too little budget
While an ad set is in the learning phase, delivery is still exploring. If the budget cannot buy enough optimisation events, the system narrows early onto whatever looked best, and the ad set may be marked learning limited.
Fix: consolidate ad sets, raise the budget, or optimise for a more frequent event so learning can finish.
Verify: the ad set leaves learning, and spend across ads becomes more stable.
Prevent: size budgets against the optimisation event before launch, using the causes listed in the learning phase guide.
Ads too similar to tell apart
Five versions of one video with different background colours give the system little to distinguish, so it picks one and the rest look like losers. They were not tested; they were indistinguishable.
Fix: vary the concept, opening or format so each ad offers something different, and expect differences in CTRShare of impressions that result in a click. and cost per result to follow.
Verify: the ads show distinct early results and spend spreads where performance differs.
Prevent: brief variants at the level of the hook or concept, drawing on the patterns in the hooks and UGC guide.
Prevention: test structures that force a fair split
Concentration only hurts when you needed an answer the structure could not give. The prevention is to separate two jobs that often share one ad set: running ads that make money, and learning which ads deserve to. Running ads can live together in a consolidated ad set where delivery is free to favour whatever works, and nobody should complain when it does. Learning needs its own structure, either Meta's experiments tool, which splits the audience into groups that each see one variant, or a dedicated testing ad set with its own budget and a written question.
Write the question before launch: which hook earns cheaper purchases among new buyers, for example. Decide the metric, the minimum delivery each variant needs, and how long the test runs. Keep variants genuinely different, at the level of concept or opening. When the test ends, promote the winner into the running ad set and retire the losers, rather than leaving all of them in rotation where the result will simply be decided again by early signals. Record each test so the next brief starts from evidence. A team that runs the two jobs separately stops worrying about uneven spend, because it is only uneven where it should be.
An apparel brand that thought three ads were tested
An apparel brand launches three video ads in one ad set with £90 a day. After ten days the first ad has taken 87% of spend at £21 per purchase; the other two have spent £64 and £53 with two purchases between them. The team concludes the first ad is the winner and briefs more like it. A review shows the three videos share the same opening shot and differ only in their final card, and the first ad started a day earlier than the others. The team builds a test in the experiments tool with three new openings, each reaching a separate third of the audience at £30 a day for fourteen days. The second opening wins at £17 per purchase, a result the rotation could never have produced.
Common mistakes
Reading rotation inside one ad set as a test. Pausing the winner to force spend onto weaker ads. Adding new creative beside a long-running incumbent and judging it after a day. Splitting a small budget across many ad sets so none can leave learning. Producing variants that differ only in details nobody notices.
