Media agencies that turned AI agents loose on media buying are now building audit trails to catch the agents overspending. Dept’s global EVP of technology, Jonathan Whiteside, told Digiday his staff have burned through 1.5 million tokens in a single day — the kind of run-up that pushed Dept, performance agency Rise, and others to instrument what their agents actually do, not just what they’re told to do.
01What the audit layer actually catches
- Token spend without a budget. Gartner’s April survey of 1,300 senior marketers found 56% of companies still implement AI tools without clear usage policies. Gartner projects 60% of AI-using organizations will face cost overruns from missing usage tracking.
- Model choice, not just volume. Gartner analyst Nicole Greene: “Rising token consumption costs are because people are literally not choosing the right model to meet the need of the activity.” Dept’s fix is an “AI gateway” that removes staffers’ ability to pick a model per task, routing them into a cost-appropriate default instead.
- The audit tool is coming from the sell side. Rise has been testing agentic media buying with a supermarket client since June using an audit-log feature built by PubMatic — an SSP, not a DSP or agency vendor — that flags when an agent is “drifting” outside pre-set parameters and time-stamps every change it makes, including how the change was made.
- Guardrails over punishment. Brainlabs CEO Daniel Gilbert frames the monitoring as enabling, not restrictive: “We want people to spend. We just want them to do it usefully” — and calls the visibility “not a nice-to-have… it’s existential.”
02What it means inside a GAM network
The detail worth sitting with: the accountability layer for agentic buying is being built by an SSP, not by the agencies spending the money or the DSPs routing their bids. PubMatic’s tool logs agent behavior — parameter drift, timestamped changes, the “why” behind an action — on the sell side of the transaction, for buy-side clients to consult. If that pattern holds as agentic buying scales into GAM/AdX auctions, publisher-side ops teams aren’t just watching yield; they may end up running the audit trail the agency side depends on to prove its own agents behaved. That is a new operational surface, not a hypothetical one — Rise’s group director, Klaudia Smykowska, described using the log to “reconstruct what happened and why” after an agent made an unexplained pricing change.
Nobody in this story is arguing agentic buying should slow down — Brainlabs' Gilbert is explicit that the goal is more spend, done usefully.
03What publishers should do about it
04The bottom line
Nobody in this story is arguing agentic buying should slow down — Brainlabs’ Gilbert is explicit that the goal is more spend, done usefully. But the fact that an SSP built the accountability tooling, not the agencies themselves, is the tell: as agentic systems start placing real bids inside real auctions, the audit trail is turning into sell-side infrastructure. A GAM operator that isn’t already asking what its own platform logs about agent-driven bidding is behind a question its agency partners are already answering.