IAB Europe: Agentic Ad Buying Expected to Scale, But Oversight Is Thin

IAB Europe: The Gap Between Agentic Hype and Agentic Reality

IAB Europe surveyed 50 ad-tech professionals on autonomous media buying, and the numbers contradict each other in a revealing way. 58% expect software buying with no human on the controls to be a regular or primary method within a year. 36 of 47 respondents run nothing like that today. They rated AI's current ad-ops performance 2.76 out of 5. The industry is planning to hand over the checkbook within twelve months to a system it currently grades below average.

The real decision isn't whether to buy agentic tools. It's whether to delegate spending authority before the controls to supervise it exist.

The oversight gap is concrete. Only nine of 38 companies that offer AI training teach people how to check an autonomous action or when to override it. When an agent picks a bad CPM floor or dumps spend on a brand-unsafe placement, the ops desk gets the call and there's no log to pull. That's not a future problem. That's the 90-day problem for any team that moves to production.

The contract problem is worse. A holding company is legally on the hook for every dollar it spends on a client's behalf. No procurement or legal team signs away that authority to a system it can't inspect, until there's indemnification language covering runaway spend. That language doesn't exist yet. 48% of respondents have advertising-specific AI rules. The other half are improvising. Until somebody writes down who pays when the agent misfires, this category stalls at the pilot stage regardless of vendor roadmaps.

Winners and losers by segment. Measurement and verification vendors (DoubleVerify, Integral Ad Science, iSpot, VideoAmp) and clean-room players (LiveRamp) have a real near-term opening. The supervision layer, audit logs, permission scoping, outcome proof, is the thing that makes agentic buying safe to run, and it doesn't exist yet. That's where the spend goes first. Pure-play DSPs (demand-side platforms) carry the real risk. If buyers decide autonomous execution is just a feature bundled into Google's and Amazon's stacks, the independents lose a pricing lever before they ever charge for it.

The vendor pitch will say "autonomous." The contract will say a trader signs off. That gap between the marketing and the clause is the whole game for the next year.

Our call: No major holding company (Omnicom, WPP, Publicis, Dentsu, or Havas) will publicly announce a client contract granting an AI agent unsupervised authority over live media budgets by the 2027 upfront negotiations in May 2027. The liability math moves slower than the vendor roadmaps, and the first runaway-spend incident, when it comes, will harden caution rather than loosen it.

For any team running pilots now: force parameter-level disclosure and an audit log into the contract, cap autonomous spend hard, and demand outcome-lift proof before granting production authority. Train the desk on when to intervene before you turn it on.

IAB Europe asked 50 ad-tech people about AI that buys media on its own, and the answers point in two directions at once. 58% think software buying with no human touching the controls will be a regular or primary method inside a year. But 36 of 47 said they run nothing like that today, or still keep a person in the chair. They graded AI's current ad-ops work a 2.76 out of 5. So the industry expects to hand over the checkbook within twelve months to a system it currently rates below average.

What's being decided: not whether to buy agentic tools, but whether to delegate spending authority to software before the controls to supervise it exist. How hard is this to undo? Easy at the pilot stage, hard once budgets, bid logic, and price floors are wired into an agent with no audit log and no sign-off rules. The deadline is set by the first public blowup, not the survey timeline.

---

The Skeptic. Fifty respondents is a conference room. A 2.76-out-of-5 rating with 36 of 47 not deployed means this measures hope, not practice. The 58% who expect agentic to be primary within a year are the same crowd that swore programmatic would kill the insertion order a decade ago. The real wall isn't the model. It's that the buyer is contractually on the hook for every dollar, and no legal or procurement team signs away that authority to a box it can't inspect. Until agency-client contracts carry language saying who eats a runaway spend, this stays in the sandbox. In plain terms: nobody delegates the credit card until they know who pays when it goes wrong.

The Operator. The thing that breaks first is the audit trail. An agent picks a bad CPM floor or dumps spend on a brand-unsafe placement, the ops desk gets the call, and there's no log to pull. That's the 90-day problem. Only nine of 38 companies that offer AI training teach people how to check an autonomous action or when to step in. So you'd be handing live spend to software and training almost no one to catch it. Trading desks should force the vendor to disclose exactly which parameters the agent touches on its own before signing anything. Anything less is a blank check. For a non-specialist: they're buying self-driving cars and not teaching anyone where the brake is.

The Market Analyst. The gap between expectation and deployment is the whole story, because it tells you where near-term money goes. The supervision layer doesn't exist yet: audit logs, permission scoping, outcome proof. That's the spend. Measurement and verification names, DoubleVerify, Integral Ad Science, iSpot, VideoAmp, have an opening to sell audit-trail and outcome-check tooling as the thing that makes agentic buying safe to run. LiveRamp and clean-room players win if agents need permissioned, traceable data access. Pure-play DSPs carry the real risk: if buyers decide autonomous execution is just a feature bundled into Google's and Amazon's stacks, the independents lose a pricing lever before they ever charge for it. For outsiders: the pick-and-shovel sellers do better than the gold miners here.

The Customer / End User. Put the advertiser in the chair. The study says any vendor selling an agentic product should prove it beats a human on advertiser outcomes and publisher net revenue, not on time saved. That line matters because every vendor pitch leads with efficiency. A CMO doesn't lose their job over hours saved. They lose it over a brand-safety fire or a quarter of wasted budget. So the buyer's real question is who signs off and who pays when the agent misfires, and the honest-broker answer right now is nobody has written it down. 48% have advertising-specific AI rules. The other half are improvising. In plain terms: the customer wants a receipt and a refund policy, and the market is selling neither yet.

The CFO. Time saved is not the number that justifies this. Delegated spend authority with no log is an unpriced liability sitting on the media budget. The payback case for an agent has to clear outcome performance and the cost of building supervision around it, which the survey shows almost nobody has budgeted. The cheap-looking tool carries an expensive tail: the first runaway campaign, the clawback fight with the client, the legal review that follows. Until a vendor can show a clean audit log and an outcome lift, the prudent line is pilots with hard spend caps, not production authority.

---

Where they split. The Market Analyst sees a real near-term market in supervision tooling. The Skeptic says the whole category stalls until contracts assign blame, so there's nothing to tool yet. That's the live disagreement: does money flow to audit-and-verify vendors in 2026, or does the legal logjam freeze the category first? Second split: the Strategist read (in the window) says whoever owns the execution-to-outcome loop gets the moat, and that's The Trade Desk, Google, and Amazon. The Market Analyst says the independents get commoditized in exactly that scenario. Both can be right, and that's the bad news for the pure-plays.

What it hinges on. One fact: who is liable when an autonomous agent misspends. Everything else, deployment pace, pricing, which vendor wins, follows from that. The council leans Skeptic on timing and Analyst on direction. The capability will arrive faster than the accountability. The gap between the two is where the spend sits, and it favors the measurement and clean-room layer over the DSPs.

To de-risk: force parameter-level disclosure and an audit log into any pilot contract, cap autonomous spend hard, and demand an outcome-lift proof before granting production authority. Train the desk on when to intervene before you turn the thing on, not after the first fire.

---

Prediction: No major holding company (Omnicom, WPP, Publicis, Dentsu, or Havas) will publicly announce a client contract granting an AI agent unsupervised authority over live media budgets by the 2027 upfront negotiations in May 2027.

Confidence: Medium — liability math moves slower than vendor roadmaps.

Why: The IAB Europe study shows 58% expect autonomous buying to be primary within a year, but 36 of 47 run nothing like it today and only 48% even have advertising-specific AI rules, so the supervision and contract scaffolding does not exist yet. A holding company is contractually on the hook for every dollar it spends on a client's behalf, and no procurement or legal team signs away that authority without indemnification language that currently isn't written. The opposite outcome, a holdco publicly putting its name on unsupervised spend authority inside eighteen months, would require that legal gap to close faster than any such gap has closed in ad-tech history, and the first runaway-spend incident will harden caution, not loosen it.

Revisit by 2027-05-31: We're right if no top-five holding company has announced a client contract delegating unsupervised live-budget authority to an AI agent by the 2027 upfront window. We're wrong if any of the five announces one, with human sign-off removed from live media spend.

The announcements through 2027 will all carry the word "autonomous" and a human still holding the pen. The pitch will say the agent runs the buy. The contract will say a trader signs off. That gap between the marketing and the clause is the whole game for the next year.