The Levers Are Disappearing. Is Your Brand Average?

In July, Meta's API version 26 removed Instagram Explore Feed and Messenger Stories as surfaces an advertiser can select. The change arrived with a number attached, the claim that Advantage+ placements deliver 11.7 percent lower cost per action than manual settings. Every removal of a control in the last two years has come with a number like that, and every number has been true, on average, across an advertiser base that includes a dropshipper in Manila and a hospital system in Ohio. The number cannot answer the only question that matters to you, which is whether your brand is average.

Precision matters here, because the story is easy to tell lazily. Meta still lets you advertise well. What it no longer lets you do is advertise differently from the system's estimate of what is best for a typical account. Placements collapse into one automated set. Audiences collapse into a broad signal the model steers. Creative rotation, budget allocation across ad sets, even the choice of which objective the account optimizes toward: each has moved, or is moving, from a dial you set to an outcome you accept. The company's generative model for ads is now producing the variations that a creative team used to produce, and its business is growing faster than Google Search, fast enough that analysts expect Meta to become the largest digital ad platform on earth by the end of this year.

Automation of that quality wins for most advertisers most of the time, which is exactly why it is dangerous for the advertisers it does not fit. The system optimizes toward the cheapest available version of the outcome you declared. If your outcome is a form fill, it will find the cheapest form fills, and cheap form fills are frequently the wrong customer wearing the right shoes. If your outcome is a purchase, it will find the customers most likely to buy anyway, because they convert cheapest, and it will report that as efficiency. A luxury brand whose entire strategy depends on which surfaces it appears on, a regulated advertiser whose legal team has opinions about Messenger, a B2B company whose buyer is one person in ten thousand: none of these are average, and the 11.7 percent was measured on a population that mostly is.

So the operator's job changes shape. When you cannot set the dials, you have to measure the machine. The only instrument that answers whether Advantage+ is helping your account, rather than the average account, is a holdout: a geography, a matched audience, or a time-split where the automated setup runs against the most controlled version Meta still permits, read on the same calendar, judged on business outcomes rather than the platform's own attribution. I have written the recipe for this before, and the reason it matters more now is that the platform's report is no longer a neutral witness. It is the defense presenting its own evidence.

Two practical moves follow from that. First, move the signal you feed the system as far downstream as your data allows. If Advantage+ is going to chase the cheapest version of your objective, make the objective expensive to fake: qualified pipeline, second purchase, margin-weighted revenue, offline conversions uploaded on a schedule. The automation is only as good as the definition of success it was handed, and most accounts handed it the definition that was easiest to implement in 2022. Second, treat brand safety as a performance metric rather than an exception you file. When placements were manual, the brand's rules lived in the setup. Now they have to live in the exclusion lists, the inventory filters, and the audit cadence, because the default is whatever the model prefers this quarter, and an impression on a surface your brand should never touch is not a cheap impression, whatever the CPA says.

There is a version of this argument that ends in nostalgia for manual buying, and I want no part of it. The old setup was slower, more expensive, and mostly worse. The point is narrower. A system tuned to the average will keep getting better for the average, and the advertisers it quietly underserves will be the ones with the most distinctive strategies, because distinctiveness is what an average cannot see. More levers would not help those advertisers. Proof would, run on their own account, of what the automation is actually doing to them.

Meta will keep publishing numbers, and the numbers will keep being true. Run the holdout anyway. It is the only average that includes you.