AI Buys Ads 13% Cheaper Than Human Traders. Almost No One Is Letting It.

An AI agent buying digital ads in May paid $6.13 for a thousand impressions. Every other kind of buyer paid $6.95. DataBeat published that comparison on June 22, in the first edition of its US Programmatic Trends Report to make AI-driven buying the featured theme.

The machines came in 13.4% cheaper. They also converted more of the bid opportunities they took, 0.204% against 0.183%, and they took remarkably few: agentic buyers sat out 86% of the auctions open to them.

If you buy advertising, those three numbers are the whole finding. The software is far more selective than a human team can afford to be, and the selectivity is what makes it cheaper.

DataBeat’s own read on the pattern was four words long: “Efficiency is becoming as important as scale.”

Then there’s the size of the thing. Michael Barrett, CEO of the sell-side platform Magnite, spent Cannes collecting 2027 forecasts for agentic ad spend and came back with a range that starts at nothing. The top of it, he told tipsheet.ai in an interview reported on August 2, is $600 million to $700 million.

“Even at the high end, that’s still relatively modest within the context of the overall programmatic market.”

So the technique is measurably cheaper, and next year almost nobody will use it.

That gap is the story. Most arguments about agentic buying have been about whether it works. The data says it does, at least on price, and the forecast says the industry has priced in roughly two years of nobody acting on that.

The results from inside the campaigns point the same direction. PubMatic reported in April that a fully autonomous campaign its agency partner Butler/Till ran for a beverage client cut buy-side fees fivefold, delivered 40% more impressions than planned, and landed a 30% lower effective CPM.

Gina Whelehan, Group Director of Strategy & Partnerships at Butler/Till, called it what it was:

“This was a clear proof point that agentic buying can drive real, measurable performance.”

Tom Leone, VP of Media Services at the agency Brkthru, described the change in his own team’s day rather than in the results:

“AgenticOS is allowing us to simplify execution across platforms, freeing our traders from manual work.”

The mechanism behind the cheaper price is the 86% figure, not the AI. A human trader working a plan has an incentive to spend the budget, so the bid list gets wide. An agent given a campaign objective bids only where the opportunity matches it. It declines most of the inventory it sees and concentrates the money on the rest. Fewer auctions, higher win rate inside them, lower average price. That’s selectivity showing up in the ledger. It’s not a smarter buyer so much as a buyer with no reason to look busy.

Worth naming the limit. A cheaper impression is not a better outcome. DataBeat’s report covers price, fill, and auction participation across a network tracking more than $55 million in monthly revenue and 35 billion monthly impressions from over 200 bidders. It doesn’t publish conversion data. A buyer paying $6.13 for inventory that converts worse than the $6.95 inventory has saved nothing. Run this comparison in your own account on cost per acquisition, not cost per thousand.

For a 20-person B2B SaaS team the practical read is narrower than the ad-tech coverage suggests. Most of this activity is happening between platforms and holding companies, and the tooling reaches you through your agency or whatever platform your team places bids from, not as something you install. What travels down to you is the pricing behavior. An automated buyer that skips most of the market and concentrates spend will look, in your reports, like a campaign that stopped bidding on most of your usual placements. That’s the intended behavior, and it will read as a bug the first time you see it.

The other thing that travels down is the job change. Leone’s phrasing is the honest version of what happens to the role, and it matches what we found when TikTok opened its ads console to agents: the execution work compresses and the brief-writing work becomes the whole job. Barrett said the same about his own company, telling the interviewer AI has absorbed contractor tasks internally without backfilling them. That’s the pattern we tracked in B2B teams shrinking without an announcement, showing up on an earnings-adjacent call instead of in a memo.

Two years of a technique being cheaper and unused points at procurement. Agency compensation is still built on hours and on percentage-of-spend. A buying method that cuts both has no natural sponsor inside the org chart that would have to adopt it. Barrett’s $700 million ceiling reads less as a judgment about whether agents can buy media and more as a judgment about how fast an industry adopts something that shrinks its own invoice.

Ask your agency for the CPM and CPA split between agentic and conventional line items on your last two quarters. If they can’t produce it, that’s the answer to whether they’ve tested it.