Meta is spending your budget inside mobile games

Open one of your Meta ad sets, switch the breakdown to placement, and sort by spend. In most of the accounts we audit, a row called Audience Network sits near the top, and the owner has never heard of it. Audience Network is Meta's inventory on other companies' apps and mobile sites: puzzle games, utility apps, sites nobody would name out loud. Your money went there because one setting was on when the campaign was built, and nobody opened that menu again.

Part of what you pay is the auction, and you do not set the auction. The rest sits inside your own account, and in most of the accounts we open it is the bigger number. The five places we look are placements, ads killed too early, stacked interest targeting, overlapping ad sets, and three defaults buried in menus. All five take about ten minutes to check, and they are where we spend that first ten minutes before anyone rewrites an ad. Creative comes after.

Advantage+ placements, which Meta used to call automatic placements, lets the system put your ad anywhere it sells space. That includes Audience Network, which works the way Google's search partners work: it is the leftover inventory, sold to whoever bids lowest. Some accounts do fine there. Most of the ones we audit are paying for it and getting nothing back.

1. 1

   Open the ad set, not the campaign

   Placement data lives at the ad set level, and so does the setting that controls it.
2. 2

   Break the table down by placement

   Use the Breakdown menu, choose By Delivery, then Placement, and your spend splits across Facebook Feed, Instagram Feed, Reels, Stories, Audience Network and the rest.
3. 3

   Sort the table by amount spent

   You are looking for rows that take real money and hand back no results, so read the cost per result column beside them.
4. 4

   Switch to manual placements and cut what is not paying

   Deselecting even one placement moves the ad set off Advantage+ placements, and you should give the change a week before you judge it.

Operators who know the platform do this constantly, because they watch the account too closely. An ad spends $80, brings no purchases, and gets paused. The ad set was four days old and still in learning, which means Meta had not yet worked out who to show it to. You paid $80 for data and threw the data away.

Meta calls that period the learning phase, and it ends once an ad set has gathered enough of whatever event you told it to optimize for. The figure Meta has published for years is about 50 of that event in a seven-day window. Below it, the ad set sits in a state Ads Manager labels learning limited, where delivery is unstable and the cost per result swings hard enough to make any read on it meaningless.

So we use a number instead of a feeling. An ad gets three times your target cost per result before anyone judges it. If a purchase is worth a $20 target, the ad gets $60 of spend, and under $60 it is not evidence.

Then there is the part that costs the most. Meta lists pausing among the edits that can send an ad set back into learning, so the nervous manager who pauses on Tuesday and relaunches on Thursday buys the same learning twice. Do that across a quarter and you have paid a tax on money you already spent, and it never appears as a line item.

If you know the platform and Meta still feels expensive, count your changes from last week. Ads Manager keeps an edit history on every campaign, ad set and ad. Open it, look at the past seven days, and count. Most people who run that count find their answer in the number.

In most accounts, no. Stacking interests was a good idea in 2019 and it is a CPM machine now. Every layer you add shrinks the pool and tells Meta to compete harder for fewer people, and you pay for that in the auction before anyone has seen the ad.

We audited an account targeting women 25 to 34 who were interested in skin care, and clean beauty, and Sephora. It ran at about a $38 CPM. Broad targeting in the same account, with the same creative, ran at about $11. That is one account and one category, so treat it as a direction rather than a number to plan against.

The reason is not a feeling about the algorithm. Meta rebuilt the machine that finds people. In December 2024 its engineering team published Andromeda, the retrieval system that decides which ads are even considered for a given person, and reported a ten-thousandfold increase in model capacity along with a six percent improvement in recall. A system that large does not need your checkboxes. It needs a bigger pool to search and a piece of creative that tells it who to look for.

Your creative is the targeting now. An ad about retinol for sensitive skin finds people with sensitive skin, because those are the people who stop scrolling, and the stop is the signal Meta reads. The checkboxes used to describe your customer to the platform. The ad does that job now, over a far larger pool.

When narrowing still earns its place

• Legal or product constraints: alcohol, age-gated services, a license that covers one county.
• Real B2B, where the buyer is a rounding error in the general population.
• Retargeting, which is a list of people who already did something rather than a stack of guesses about who they are.

Outside those three, each extra layer is a bid on scarcity, and scarcity is what you pay for in CPM.

You do not bid against yourself in Meta's auction the way the folklore says. What you do is fragment your own delivery. Three ad sets aimed at roughly the same people force Meta to divide impressions between them, and each one collects its optimization events more slowly. Three ad sets stuck in permanent learning is a worse account than one ad set that left learning weeks ago.

We consolidated a client from 14 ad sets to 4. Same budget, same creative pool, nothing new written. The cost per lead dropped. We are not publishing the size of the drop, because one account over one quarter is an anecdote and the only part of it worth your time is the direction: delivery stopped being cut fourteen ways, and something was finally allowed to finish learning.

The arithmetic fits on a napkin. Divide your weekly budget by your number of ad sets, then divide that by your cost per result, and you have the events each ad set can expect in a week. A $2,000 weekly budget across five ad sets is $400 each; at a $25 cost per lead that is 16 leads per ad set per week, well short of the 50 an ad set needs to leave learning. Four of those five ad sets should not exist, and the fix is subtraction rather than a new headline.

Every default in Ads Manager was chosen by Meta, and most advertisers have never opened the menus where they live. Three are worth checking today.

Advantage+ audience

This is a different setting from Advantage+ placements, and the shared name causes trouble. Placements decide where the ad appears; Advantage+ audience decides who sees it, and it lets Meta treat your targeting as a suggestion and go looking outside it. Sometimes that finds buyers you would not have picked. When the product is age restricted, license restricted, or tied to one service area, the expansion spends on people who are not allowed to buy from you. The test is simple: if someone outside your targeting could not complete a purchase, turn the expansion off.

The attribution setting

Meta's default window counts a conversion when someone clicked your ad in the past seven days, and also when they only saw it in the past day. That view credit makes the dashboard read higher than a click-only window would, and if you have been scaling on the dashboard, you have been scaling on the difference. Find the setting, then compare a month of reported conversions against what your own sales records say happened, and scale on the comparison.

The account spending limit

Most accounts have none, which leaves one mistyped budget between you and a $5,000 day. An account spending limit is a lifetime cap that stops delivery once it is reached, it lives in billing settings, and it takes about a minute to set. It is the cheapest insurance on the platform, and the only reason it stays rare is that nobody enjoys planning for their own bad Tuesday.

Work through the list in order, write down what you find, and change nothing until you reach the bottom of it. The point of writing it down is that you will want to fix the first thing you see, and the first thing you see is rarely the expensive one.

Then change one thing and leave it alone long enough to read. Meta needs about a week and about 50 events before it can tell you anything, and checking more often does not shorten that.

The accounts that scale are dull to watch. Fewer changes, longer windows, decisions made on samples big enough to mean something. That is an unsatisfying answer for anyone who wants to feel busy in Ads Manager, and it is the one that shows up in the cost per result at the end of a quarter.

So start where we started. Open one ad set, break it down by placement, and read the spend column from the top. Whatever sits on that first line has been taking your money all year, and it has never once been asked to justify it.