Agentic media buying went from a CES keynote topic in January to a client agenda item by March. By April, the questions arriving in agency inboxes were no longer coming from the trade press. They were coming from clients who already knew the word, had probably read the same Yahoo and PubMatic releases the agency had read, and wanted to know what the agency was actually planning to do about it.
The agencies that came back with "we're watching the space" lost ground in the same conversation. The agencies that came back with a roadmap deck lost slightly less ground, slightly more politely. Neither answered the question the client was actually asking.
The Three Questions Hiding Inside One
When a client asks about agentic, they are almost always asking three different things at once. The agencies that sort the three apart leave the meeting with a different relationship than the ones that don't.
The first is operational. *What changes about how my campaign runs?* This is the version most agency answers stop at. The version is real. Yahoo DSP, PubMatic, Amazon and others now have systems where AI agents set up, optimise and report on campaigns at speeds humans cannot match. PubMatic's early AgenticOS data showed an 87% reduction in campaign setup time. Butler/Till ran a live CTV buy for Clubtails in December where an Anthropic-powered agent handled the brief end to end. Money is moving through these systems today.
The second is commercial. *What does this do to my media efficiency?* This is the version most agencies are quieter about, because the honest answer is uncomfortable. Agents do not run on goodwill. They run on parameters, and the parameters route money. An agent optimising across a chain where roughly half of advertiser spend never reaches a publisher is not a more efficient agent. It is a faster way to lose the same proportion of the budget. The ISBA and PwC supply chain studies remain the cleanest summary of what that looks like. The 2020 finding was that approximately half of digital advertising spend reached publishers. The 2022 update lifted that figure to about 65%, though PwC themselves noted the result represented what the industry "should aspire to" rather than what generally happened. The long tail of programmatic, where most spend lives, has not closed the gap.
The third question sits underneath the other two. *Who is making the decisions about my budget now?* When an agent decides where to route a campaign, the question of who wrote the agent's logic becomes the question of who effectively controls the spend. If the agent runs on Yahoo DSP's default targeting, PubMatic's default routing, or any platform's default optimisation function, the answer is the platform. Not the agency. Not the brand. Platform defaults align with the platform's optimisation function, which is some combination of inventory clearance and platform revenue. The brand discovers which one when the post-campaign report arrives.
These three questions sit on top of each other. The operational answer is interesting. The commercial answer is uncomfortable. The structural answer is the one that will define which agencies clients keep on retainer in 2027.
The Arithmetic That Breaks the Awkwardness
Two pieces of data sit underneath any honest version of this conversation.
The first is what working media looks like in the existing supply path. Six to ten intermediaries clip a fee on every impression before it reaches the screen. That is the chain the agent will optimise across, whether or not anyone has agreed to look at it. The 2022 ISBA/PwC update narrowed the unattributable portion of spend to about 3% at the premium end, but the lift was conditional on cooperative vendors and controlled study conditions. The structural arithmetic underneath was not changed. It was photographed under better lighting.
The second is what happens when the chain is shortened instead of optimised. Across consumer electronics and financial services campaigns running through direct supply paths in H2 2025, working media delivery improved by 212% for the same spend. Cost per ad buy fell from $9.03 to $2.71. Platform fees compressed from the industry-standard 10% to 20% range to 3% for display, 5% for video and 8% for connected TV.
The downstream brand outcome is the bit worth pausing on. The Polestar connected-TV activation that ran through this stack produced a 34% lift in sales intent and a 24% lift in brand association with "100% electric." Nielsen measured both. Viewability landed at 99%, video completion at 96%. None of those numbers are infrastructure metrics. They are what infrastructure produces when more of the budget reaches the screen, which is the only metric that has ever mattered.
Neither set of numbers requires belief in any specific vendor's view of the future. They are arithmetic and they are auditable. The change in the agentic era is that the arithmetic stops being optional. An agent routes to the cleanest path it can audit, because that is what an agent is built to do. A media buyer can be talked out of it over a long lunch. An agent cannot.
What Agencies Should Bring Into the Next Meeting
The simplest version of the conversation is this: the fee conversation the agency has been deferring is about to be done for it, faster than it can intervene.
That sentence reframes agentic from a technology story to a procurement story. It also reframes the agency from a vendor at risk of being automated to the party that saw the shift coming and prepared the client for it. Both reframings matter.
Three actions follow.
Audit the current supply path. Not in the marketing sense. In the structural sense. How many intermediaries sit between the brand's budget and the publisher's inventory, what fee does each one take, and what proportion of every pound spent actually reaches working media. Most clients have never seen the stack mapped end to end. The mapping exercise is more uncomfortable than running the campaign through the stack. The agency that does it first wins the trust that comes with it.
Pilot a direct path. Not a rip-and-replace. A side-by-side: same campaign, same creative, same audience, run through the existing stack and through a more direct alternative for three to four weeks. Define the success metrics upfront. Working media percentage. Cost efficiency. Brand uplift. At the end of the pilot, the comparison is a number. The data does the convincing.
Ask the verification question of every agentic capability being pitched. *Can the agent independently verify what it bought, against a source of truth neither party controls?* If the answer is no, the agent is optimising against self-reported data. That is the same trust model the industry has always had, running faster. It is not progress. It is the same drift with better dashboards.
The Position That Wins
The agencies that handle the next twelve months well will not be the ones with the slickest agentic pitch deck. They will be the ones who arrive with the supply path already mapped, the fee structure already audited and a pilot proposal already costed. They will frame agentic not as a feature their tech stack supports but as the moment the questions the industry has been deferring become impossible to defer.
The agencies that get this wrong will reach for the comfortable answer: *yes, we have agentic capabilities, here is the roadmap.* The answer is true. It is also irrelevant. The client is not asking about the roadmap. The client is asking where the money goes. And the agent is already in the meeting.
---
Related reading from Alkimi Research
The verification layer that Programmatic needs! The verification infrastructure underneath the questions above, written for the programmatic veteran with no blockchain background.
Is your Brand ready for Agentic Advertising? The brand-side complement to the agency reposition above.
Run the numbers on a client's supply path
The Alkimi research team runs side-by-side pilots that compare a current supply path against a direct alternative across three to four weeks. Working media, cost efficiency, brand uplift and carbon footprint measured on the same campaign. Pilot proposals via lauren@alkimi.org or marco@alkimi.org.