Targeting is not dying, it is moving
Ad platforms may now choose who sees an ad better than you can, which leaves marketers with positioning and with checking the platform's own reported results.
“The importance of targeting is diminishing over time.” I teach targeting, and I think the person who said it is half right.
The quote is from Matt Steiner, Meta’s VP of ads infrastructure, ranking and AI foundations, speaking to AdExchanger. Here is where I agree with him, where I do not, and what the disagreement implies for a marketing team.
What changed
Meta said its ad system now scores every eligible ad for every impression. It also added a new LLM-based model that reads each ad’s text and images to work out what it is selling.
If that is accurate, the job of choosing who sees an ad is increasingly done by the machine. The machine sees more signals than a media planner can, and it does so for every impression. On that narrow task, I would not bet against it.
Where Steiner is right, and where “targeting” means two things
“Targeting” is used for two different jobs, and the argument turns on separating them.
The first is delivery: given this ad, which person should see it right now? This is the job the platform is getting better at.
The second is strategy: which customers are we for, and how do we position ourselves to them? That is segmentation and positioning, the part of marketing that decides what the ad should say and to whom the firm is promising something. No ad system answers it for you. The platform can find people likely to respond to an ad. It cannot decide which customers you want to be the best choice for.
So targeting is not dying. The first job is moving to the machine. The second stays with you, and the quality of what you feed the machine depends on it.
The job that grows: checking the machine
There is a third job, and it grows as the first one is automated. Someone has to check the machine’s work.
Every gain Meta cites is its own number. It reports 2.4% more Facebook conversions from Andromeda since 2025, and 1.4% credited to it in Q3, and Meta has not reconciled the two. Average price per ad rose 14% in the same quarter. I am not claiming these figures are wrong. I am pointing out that the platform chooses the audience, runs the auction and reports the result, so nobody outside it has confirmed any of them.
An example I use in class
Ad spend and “revenue from ads” rise together, so leadership wants to double the budget. But that revenue is counted by the ad platform itself.
The problem is not dishonesty. It is attribution. A platform that is good at finding people who were already going to buy will show a lot of purchases after an ad, whether or not the ad caused them. A rise in both numbers is what you would see if the ads work. It is also what you would see if the platform simply got better at showing ads to likely buyers.
The fix is boring, and it works
Hold some audiences or cities back, and compare.
Here is the reasoning with deliberately simple, made-up numbers. Suppose you run ads in forty cities and withhold them from ten comparable ones. If sales rise by the same proportion in all fifty, the ads did not drive the rise, whatever the dashboard says. If sales rise only in the forty, the gap between the groups is a fair estimate of what the ads added. The numbers are illustrative. The logic is the point.
A few conditions make this work:
- Choose the held-back groups at random, or match them carefully. If you hold back the cities that were weakest anyway, the comparison is spoiled before it starts.
- Decide the measure in advance. Pick the outcome, such as sales or new customers, and the period before you see results.
- Hold back for long enough. Effects of advertising can be delayed, so a few days may not be enough.
- Count the outcome in your own systems. Use your sales records, not the platform’s conversion report.
This is a standard experimental design, and it does not require trusting anyone’s model. That is why I like it.
What this means for a team
If the platform picks the audience and also reports the results, the team’s role changes in three ways.
- Spend more thought on who you are for. The positioning, the offer and the creative are the inputs the platform cannot invent.
- Own the measurement. Someone in the firm should be accountable for holdout tests, and for reconciling the platform’s numbers with your own sales.
- Treat platform gains as claims. A reported lift is a hypothesis until a comparison supports it.
None of this requires distrust of Meta or any other platform. It requires the same discipline you would apply to any supplier who marks their own work.
The question I would ask
When the platform picks the audience and also reports the results, who checks its work in your team? If the answer is nobody, then that is the first thing to fix, before the next budget conversation, and well before anyone agrees to double the spend.