Social Rocket
AI agents are meant to make digital ad bookings faster and cheaper. But for publishers, an automated deal alone is not yet a business model: what matters is whether more revenue remains after technology, review and sales costs. At Advertising Week New York in early October, publishing managers reported ongoing tests, not a proven, widespread ROI breakthrough.
What work AI agents are actually meant to take over
In so-called agentic advertising, software agents can take requirements from advertisers, find suitable inventory, communicate terms and prepare booking steps. The approach goes beyond automated bidding, because agents are also meant to coordinate planning and sales tasks. The IAB Tech Lab is developing open interfaces and reference implementations for this, rather than a new closed ad market.
Already published building blocks include buyer and seller agent SDKs, Agentic Direct based on OpenDirect, and extensions for AdCOM and the Deals API. An Agent Registry is intended to simplify the assignment of agents. However, a technical standard confirms neither higher revenue nor lower total costs.
The decisive metrics for publishers
For an economically sound pilot, publishers should consider four variables separately:
- Cost per completed deal: staff time for offers, coordination, review, follow-up work and technical integration.
- Net revenue after fees: booking volume minus platform, intermediary and operating costs.
- Error rate: wrongly offered environments, impermissible price reductions, budget overruns or subsequent corrections.
- Additional business: Has the agent reached new customers, or only automated a booking that was previously sold manually?
A sensible approach is to compare over several weeks with conventionally handled, as similar as possible, bookings. Teams should define in advance which negotiations the agent may conduct and at which points humans must give binding approval. A single successfully automated deal would not yet be proof of impact.
Why publishing houses are holding back
In Digiday reporting from 12 October, managers at The Economist, Newsweek and The New York Times described different stages of evaluating and experimenting. CNN, by contrast, is among the houses already testing its own path towards agent-supported sales. More on this is shown in our overview of CNN's planned agent ad sales from 2027.
The caution can also be explained with conventional programmatic advertising: according to the ANA Transparency Benchmark from May 2026, better-performing advertisers convert 54,0 percent of their programmatic spend into qualified impressions, weaker ones only 32,1 percent. This is not a measured effect of AI agents. However, it shows how important inventory quality and transparency of the cost chain already are without additional automation.
What German marketers can prepare now
Before a pilot project, available placements, minimum prices, exclusions, data protection rules and contract approvals should be available in machine-readable form. An audit log must document decisions, data used and human interventions in a traceable way. For personal user data, the relevant European data protection and consent requirements continue to apply; automation does not create a new legal basis.
Publishers that are not yet letting agents sell also benefit from this groundwork. They simultaneously need a clear strategy for high-quality inventory, as the discussion about Stagwell's acquisition of Freestar shows. And they should reduce their dependence on pure retrieval volume – see our analysis of publisher traffic, AI bots and more sustainable KPIs.
Conclusion: Agentic advertising has become technically more concrete. For the seller, however, the relevant question is not whether an agent can book, but whether it sells more efficiently in a verifiable way without worsening prices, brand quality or control.
Sources: IAB Tech Lab – Agentic Advertising, IAB Tech Lab – Roadmap, ANA – Programmatic Transparency Benchmark Q1 2026, Digiday – publisher interviews.