“Media buying is dead”: what Meta has automated, and what is still in your hands

I think it is right, on one condition: understanding where it comes from. It follows directly from the way Meta makes money. Once you understand that model, you see why Meta automated targeting and bidding, why you no longer really get to choose your audience, and where your performance now plays out.

My conviction: Meta has automated targeting and bidding, and your performance now plays out in the creative and the post-click. That is what this article sets out to show, in three parts: Meta’s business model, what it has automated, and then what it has left to you.

Understanding how Meta makes money

When you run paid media on Meta, you need to know how Meta makes money, so you can align your interests with its own. In the end, Meta always wins. You might as well know on what terms you win with it.

Meta’s revenue has followed an exponential curve for twenty years. It reached $201 billion in fiscal 2025, $196 billion of it from advertising, up 22% year over year (Meta, annual results published in January 2026).

That revenue comes from you, the advertisers. By default, Meta charges you per impression: you pay it to show your ads on your audience’s phones and computers.

To grow that revenue, Meta therefore has two variables.

First variable: the number of ads

The more ads there are, the more impressions there are, and the more revenue Meta makes.

This variable has a natural limit. We go on Instagram and Facebook every day to see content from the people close to us and the accounts we follow, and advertising comes on top of that. If Meta pushes the number of ads in circulation up too sharply, our interest in these platforms drops, we spend less time on them, and Meta ends up losing money. So there is a glass ceiling on how many ads users are willing to see.

Second variable: the price of each ad

The second variable is the price per ad, which paid media jargon calls the CPM, the cost per thousand: the price Meta charges you to show one of your ads a thousand times.

This variable has a limit too. If Meta raises its CPM too much, the cost of delivery goes up, advertisers’ ads stop being profitable, and advertisers go looking for other platforms. Meta loses money, once again.

To be precise: in 2025, both variables rose together. The number of ad impressions grew 12% and the average price per ad 9% (Meta, 2025 annual results). The ceiling on the number of ads can’t be measured directly from the outside; what is certain is that neither variable can rise without limit.

Why Meta has only one path: your performance

These two limits lead to a simple conclusion. Meta can only raise its CPM on one condition: raising the performance of your ads at the same time.

If your ads cost you more but keep bringing in more revenue, you keep making money, so you keep investing. That is exactly the tunnel Meta wants to keep you in. Its goal is to keep improving the performance of the algorithm that delivers your ads, so that it can raise its CPM, and with it its revenue.

In my view, Meta’s goal is to keep you around an ROI of 3, meaning three dollars of revenue for one dollar of ad budget (what many teams call ROAS). That is, as I see it, the global average performance on Meta. In other words, when you are in that zone, Meta captures about a third of the revenue its ads bring you.

What this model means for you: Meta needs your ads to stay profitable so it can raise its price. Your interests and its own therefore meet on performance.

The four elements that drive an ad’s performance

To see where Meta went looking for those performance gains, you have to break down, in chronological order, what makes an ad produce a sale. There aren’t a thousand elements. There are four.

1. 1The creative. The ad itself: the image, the video or the carousel.
2. 2The targeting. The audience, the group of people the creative is shown to.
3. 3The auction. You are never the only brand going after a given audience. An auction system inside Meta decides which ad wins the right to be shown to a person at a given moment.
4. 4The post-click. Everything that happens off the platform, once the person has seen the ad and clicked: the landing page, the site’s organic search, the user experience. All of these decide whether the person buys.

Seen from this angle, Meta’s strategy becomes clear. To raise the performance of its ads, it started by automating, with artificial intelligence, the two elements that sit on its platform: targeting and bidding. That is what, for about ten years, we called media buying.

On the bidding side, the strategy offered by default for most performance objectives, “Highest volume,” already lets Meta set the bid itself in each auction (Jon Loomer, guide to Meta bid strategies). On the targeting side, the automation is more recent and goes deeper. That is the one I suggest we look at in detail.

What changed: the creative became the targeting

In 2024, a phrase attributed to Meta hit the profession like a bombshell: “Creative is the new targeting.” In other words, the creative becomes the targeting. On its own, it sums up the consequence of automation.

Before Andromeda

Before Andromeda, Meta’s system that preselects, out of tens of millions of ads, the ones that can be shown to each person, you defined your audiences yourself. For example: “I want to target women aged 30 to 45 who are passionate about matcha, and men aged 40 to 50 who work out every morning. And to both of these audiences, I want you to show my creative number 1.”

You chose both the targeting and the creatives to show to that targeting.

After Andromeda

Today, it is the other way around. You give Meta your creatives, and that is it. Meta automatically determines the audiences it considers most relevant for each one.

To do this, it analyzes the content of your creative (what it shows, what it says, what you hear in it) and cross-references it with everything it knows about its users. Andromeda sorts, out of tens of millions of candidate ads, the few thousand that are relevant to a given person, before other models choose the ones that will finally be shown to them. It runs, among other things, on NVIDIA Grace Hopper chips (Meta Engineering, December 2024).

Meta claims more than 3.5 billion people use at least one of its apps every day (Meta, December 2025). And the amount of information it has to describe them is huge: back in 2016, ProPublica counted more than 52,000 attributes Facebook used to classify its users.

To measure the gap, take my own case. If I tell you I am a 36-year-old man, the founder of a software product for digital paid media, and that I post videos on YouTube, you have four pieces of information about me. That is already a lot for a first meeting. Meta, for its part, can describe me with a catalog of tens of thousands of attributes: who I am, what I do, what I like, what I don’t like. It is by cross-referencing the content of your creatives with that knowledge that it raises performance.

The main consequence fits in one sentence: you no longer define the audiences, Meta takes care of it, and your job is to produce the creatives that guide it.

Why you no longer really get to choose your audience

The old targeting methods still exist. I advise you against using them, and Meta has made them much less visible in Ads Manager: Advantage+ audience applies by default, and the targeting details you add (interests, maximum age, lookalike audiences) are treated as mere suggestions, which the algorithm can go beyond (Jon Loomer, “What is Advantage+ Audience”).

To understand why this is good news, you have to remember how targeting used to work. To keep it simple, there were two ways of doing it.

Interest targeting

You told Meta: I want to target women interested in matcha and running.

This targeting had two flaws. The first is obvious: with a few pieces of information, even a few dozen, you will never match the precision of what Meta knows about each person.

The second is more troubling: this targeting could produce the opposite of the intended effect. It was semantic. It picked up words without capturing the user’s intent. By selecting “running,” you could reach people passionate about running, but also those who had commented on a post to say “I don’t like running.” That was how things worked in the early days, and Meta can’t be blamed for it. But you were adding people to your audiences who ran against your goal.

Lookalike audiences

The second method was to give Meta a list of customers, with as much information as possible (email address, first name, last name, phone number), and ask it to find people who resemble them.

The flaw in this method is subtler. You were giving Meta people who had already bought your product, so people closely tied to it. Meta went looking for an audience in the area closest to those customers, based on the thousands of data points it already had on them. That wasn’t enough: you stayed locked around your current customer base.

With Andromeda, Meta still searches around your customers, but it also goes looking for audiences that look nothing like them, and that can buy all the same.

What this asks of the media buyer

For these two reasons, I encourage you to be 100% convinced that Meta, with Andromeda, will bring you much better performance than back when you chose your audiences. This matters a great deal for what follows: a media buyer who keeps looking for performance in audience settings is fighting on ground the algorithm already holds.

One caveat is in order. This conviction holds for your campaigns as a whole, and Meta’s business model reinforces it: Meta needs you to make money so that it can raise its price. A single campaign, taken on its own, can disappoint. In that case, look first at the creative and the post-click, before touching the audience settings.

What is still in your hands: the creative and the post-click

Meta has left the advertiser everything that sits off its platform: producing the creative, and the post-click.

That is the meaning of the phrase “media buying is dead, long live creative strategy.” It holds if you give the word creative its full meaning: the image or the video, and also the link between the click and every step of the post-click. A creative promises something to a specific person; the landing page and the site have to keep that promise all the way to the purchase.

In practice, this shifts three things in your work.

1. 1Your creative does your targeting. What it shows, what it says and who it speaks to tell Andromeda which people to go after. A vague creative produces a vague audience.
2. 2Your lever on performance sits upstream. What you know about your customers, their problems and their situations, which you used to put into audience settings, now has to show up in the content of your creatives.
3. 3Your responsibility extends past the click. The landing page, the site’s organic search and the user experience are part of the creative’s promise. A creative that brings the right person to a page that doesn’t speak to them is still a creative that doesn’t sell.

The new goal of the game

If each creative automatically goes after its own pocket of audience, the market you are targeting looks like a large surface. A creative C1 reaches an audience A1, a creative C2 reaches an audience A2 that partly overlaps the first, a creative C3 reaches a smaller audience A3.

The goal of the game on Meta becomes covering as much of your target market as possible with your creatives. To cover that market, you need creatives in volume, diverse and specific.

What this shift doesn’t solve: the quality of what you give the algorithm. Andromeda chooses the audience from your creatives. It can’t invent a message you haven’t written, or a landing page that convinces.

Three questions to ask yourself

1. 1Where do you spend most of your time today: in audience and bidding settings, or in designing creatives and the pages that follow them?
2. 2If you showed your current creatives to a stranger, could they tell who each one speaks to and what problem it solves?
3. 3Does each of your creatives have a landing page that carries its promise forward?

If you answer no to any of them, start there.

Meta has taken over targeting and bidding. What it delivers now depends on you.