With automatic placements, Meta spends where inventory is cheapest to fill rather than where your audience converts best, so one surface routinely takes the large majority of a budget without anyone deciding that — check the placement breakdown for concentration, and split the ad set only when a placement is both large and performing differently from the rest.
A campaign is set up across four placements. A week later it has spent ninety per cent of its budget on one of them. Nobody decided that, nobody was told, and the campaign settings still say four placements — because they are still four placements. Meta simply filled the cheapest one.
This is not a malfunction. Automatic placements are doing exactly what they are designed to do: find the least expensive inventory that will produce the optimisation event you asked for. The problem is that the cheapest surface to fill and the surface where your audience converts best are different questions, and only one of them is being asked.
Why the budget concentrates
Meta runs one auction across every surface a campaign is eligible for. Some surfaces have far more inventory than demand — short-form video in particular — so impressions there clear cheaply. If a campaign optimises for a shallow event, such as a form open or a landing page view, the cheap surface will usually win that comparison, because shallow events are easy to produce anywhere.
Concentration is then self-reinforcing. The surface that wins early gets more of the learning data, which sharpens delivery there, which wins more of the auction. Within a few days a campaign that was set up across four surfaces is effectively a single-placement campaign with three placements listed in its settings.
None of that is wrong when the cheap surface also converts. It is wrong silently when it does not, and the reporting you look at every morning is built to hide the difference.
Why your averages hide it
A campaign-level cost per result is a weighted average, and once one placement holds most of the spend, it is that placement’s cost per result wearing the campaign’s name. The other surfaces are still in the report, still producing numbers, and contributing almost nothing to the figure you read.
So the campaign does not look broken. It looks like whatever the dominant surface is doing. If that surface produces high volume at low quality, the campaign looks cheap and improving — which is the most dangerous thing an underperforming campaign can look like.
The question is not whether one placement is winning. It is whether anyone chose the one that is.
How to check it
In Ads Manager, open the campaign and use Breakdown → By Delivery → Placement. Set the date range to the last seven or fourteen days rather than the campaign lifetime, because concentration builds over time and a lifetime view averages the drift away. Then read three things in order.
- Share of spend. Sort by amount spent and look at the top row as a percentage of the total. One surface holding more than about three quarters of the budget is a decision, and if nobody made it deliberately then the auction made it.
- Cost per result, per placement. Not impressions, not clicks — the result you actually care about. A placement taking most of the budget at a worse cost per result than the others is the whole finding in one row.
- Result quality, where you can see it. Cost per result is only half the question. If a placement produces leads that never answer the phone, its cheap cost per lead is the most expensive number in the account.
Do this per ad set rather than per campaign. Campaign-level breakdowns pool ad sets with different audiences, and a placement can look fine in aggregate while being the problem inside one ad set.
The naming trap in the breakdown
Meta’s placement breakdown is two fields, not one: the platform (Facebook, Instagram, Messenger, Audience Network) and the position within it (feed, story, reels, marketplace and so on). Reading only one of them is how two different surfaces end up looking like the same row.
The position values are not consistent either. Some repeat the platform name and some do not, so the same export can contain a position called “instagram_reels” under the platform Instagram alongside a position called simply “story”. And Meta’s internal name for the Facebook feed is “stream”, which is not what anyone calls it. If you are pulling this into a sheet, normalise the pair before you group by it, or you will split one placement across two rows and conclude neither is large.
What to do about it
The reflex is to exclude the dominant placement. That is often wrong. If it is taking most of the budget and still producing your best cost per result, it is not a problem, it is the campaign working — and turning it off will raise your costs.
Two conditions have to hold together before acting. The placement has to be large enough to matter, and it has to be performing differently from the rest. Either one alone is not a finding.
- If it is large and performing worse, split it into its own ad set rather than excluding it outright. A separate ad set gets its own budget and its own measurement, which turns an assumption into something you can read next week.
- If it is large and performing well, leave it alone and note it. The campaign has effectively told you where this audience is, which is worth knowing before the next launch.
- If it is large and you cannot tell, the answer is usually that your optimisation event is too shallow to separate the surfaces. Optimising for a deeper event costs more per result and buys you a comparison that means something.
Splitting has a real cost, so it is not a default. Each new ad set re-enters the learning phase and needs enough weekly events to leave it, which is why splitting a small budget across four placements usually produces four ad sets that never learn anything. Split the one placement worth measuring, not all of them.
How often to look
Weekly is enough for a stable account, and after any change that alters delivery — a new creative format, a budget increase, a change of optimisation event. Concentration rarely appears overnight; it appears over the week after something else changed, which is exactly when nobody is looking at placements because they are looking at the thing they changed.
The reason this one is worth a recurring check rather than a one-off audit is that nothing in Ads Manager will raise it for you. There is no alert for “your budget moved”, because from Meta’s side nothing went wrong. The campaign is delivering, within its settings, at the price the auction set.