Episode 195: Scott Messer Answers the Question, Is There Crying in Ad Tech?

Full analysis

Scott Messer went on Marketecture with Ari Paparo and Eric Franchi and made a big claim: agentic direct buying, AI agents placing reservation buys straight with publishers, could be the biggest structural change to programmatic since real-time bidding. His math is that 51% of media dollars get lost to fraud, unverified inventory, and fees, with DSP and SSP fees alone eating roughly 30%. Kill the middle, hand that 30% back to working media. The same episode covered the open-web traffic collapse, the Google antitrust opening for Prebid, OpenAI's behavioral ad plans, and Apple's Safari blocklist.

What's being decided, and how hard to undo: Nothing is being decided this week. This is a read on where the programmatic plumbing is headed over the next two years. The direction of travel is easy to undo in any one quarter and very hard to undo over five years. No deadline, but the Google remedy and the Prebid Summit energy set a real clock on the auction-layer piece.

Three or four personas actually move this. Here they are.

The Market Analyst

The 30% fee number is the whole pitch, and it rests on a slide Scott Messer won't source. The ANA transparency work found somewhere between 25 and 50 cents of every dollar missing working media, depending on how you define a fee. That is a range wide enough to drive a truck through, and Messer is a consultant who sells the fix. Discount accordingly. The more interesting signal is the Prebid auction flip. If Google Ad Manager is forced to bid into Prebid Server rather than the other way around, the unified auction layer moves from Google's house to open-source infrastructure the SSPs control. That is a genuine power shift, and it is the first time the DOJ case produces something an operator can build on rather than wait on.

The Skeptic

Agentic direct removes the DSP and the SSP, so who does brand safety, who does frequency capping, who reconciles the buy, and who eats the fraud when there is no verification layer in the middle? Those fees aren't pure rent. Some of that 30% pays for work that still has to happen. Scott Messer concedes the point himself by cabining his thesis to CTV and audio first and banner display "last or never." The easy wins are the premium, low-volume, already-direct-ish channels, and the long tail stays exactly where it is. Agentic direct in CTV is automating a reservation process that salespeople already run. Calling that the biggest shift since RTB is a stretch.

The Operator

Try to run an agentic direct buy Tuesday morning. The protocols, MCP and ADCP, are early. The publisher has to expose inventory an agent can read, price, and reserve. Most publishers can barely keep their ads.txt clean. What breaks first is reconciliation: the agent books a reservation, the campaign underdelivers, and now there is no SSP log to argue over. At 90 days the second-order problem shows up. Every publisher that opens an agent endpoint is also opening a new way to get scraped and gamed, right as they are fighting AI bots on the content side. The traffic collapse is the real Tuesday-morning problem. Publishers down 20 to 70% on open-web traffic do not have the engineering bandwidth to go build agent endpoints. They are busy standing up newsletters and events.

The Customer / End User

Two very different customers here. The agency buyer wants that 30% back and will happily route around the DSP if the tooling works and the reporting reconciles. The publisher wants loyal audiences and direct revenue, and agentic direct sounds great until it means exposing inventory to an agent that negotiates price down. Then there is the actual consumer, who is about to meet OpenAI running Meta's playbook inside something that feels like search. Ari Paparo's framing is that OpenAI is showing you car ads three weeks after you researched cars, using LiveRamp for identity, which is behavioral retargeting in a product people trust because it answers questions. Google kept search context and behavioral data apart for years. OpenAI is not promising to.

The CFO

The fee math is seductive and the payback is not. Building agent endpoints, new reconciliation, new fraud controls outside the SSP, all of that is real headcount and real time, spent by publishers whose top line is down double digits. The near-term money is elsewhere. Apple's Safari blocklist is cutting eCPMs about 10% for updated users, but only 8% of users have updated, so the hit is about 1% today. Scott Howe at LiveRamp called the impact "absolutely nothing." That is a man talking his book, and Ari Paparo and Eric Franchi treated it that way. The quieter CFO lesson sits in Character.AI: a company valued at a billion, backed by a16z with $193 million raised, filed Chapter 11, with Disney opening at $20 million for a non-exclusive license after Google took the team for $2.5 billion. Any ad-tech startup that concentrated inventory in one AI chat app just watched the cautionary tale play out in real time.

Where the thoughtful people split

The first disagreement is whether the 30% in fees is rent or wages. Scott Messer treats it as rent to be reclaimed. The Skeptic and the Operator say a chunk of it pays for verification, safety, and reconciliation that do not vanish when you remove the box, they just move. Settle that and you know whether agentic direct is a margin windfall or a cost reshuffle.

The second is where the real power moves. The Market Analyst says the Prebid flip is the structural story because it relocates the auction layer out of Google's control. Scott Messer says agentic direct is the structural story. Both can't be the headline. The Prebid flip is closer, more concrete, and does not depend on protocols that barely exist yet.

The third is OpenAI. Is behavioral retargeting inside ChatGPT a new front door for advertisers, or a trust problem waiting to blow up when consumers realize the answer engine is also an ad-targeting engine?

What it comes down to

Agentic direct hinges on one belief: that fees are reclaimable rent rather than payment for work that still has to be done somewhere. The council leans toward "somewhere." The money comes back only in channels that were already close to direct, which is why Scott Messer lands on CTV and audio and quietly writes off display. The near-term, buildable shift in this episode is the Prebid auction flip. It is contingent on Google integrating through Prebid Server, the server-side version that actually delivers a unified auction, rather than the lightweight client-side Prebid.js. Before anyone reorganizes around Messer's thesis, verify two things: whether Google Ad Manager actually bids into Prebid Server under the remedy, and whether any publisher can run an agentic reservation end to end with clean reconciliation. Both are checkable. Neither is proven.

Prediction: Google will not integrate Google Ad Manager as a bidder into Prebid Server by the Prebid Summit in autumn 2027, leaving the "flip the auction" plan stuck on the client-side Prebid.js that doesn't deliver the unified auction publishers want.

Confidence: Medium. Google has every incentive to drag, and the remedy's technical terms are unsettled.

Why: Ari Paparo reported from the Prebid Summit that flipping the order, making Google bid into Prebid rather than Prebid bidding into Google, only works if Google integrates through Prebid Server, the server-side version, and AdExchanger's Anthony Vargas flagged that a Prebid.js integration does not count. Google's whole ad-server advantage comes from controlling the final auction, so handing the unified auction to open-source infrastructure the SSPs run cuts against its core economics, and legal remedies specify outcomes long before engineers wire up the deep integration. The opposite outcome, Google voluntarily doing the server-side work on the timeline publishers want, would require it to accelerate the dismantling of its own auction control, which no company under antitrust pressure does without being forced line by line.

Revisit by 2027-11-30: We're right if, by the autumn 2027 Prebid Summit, Google has not shipped a Prebid Server bidder integration for Google Ad Manager. We're wrong if Google Ad Manager is live as a bidder into Prebid Server by that date.

One more thing worth saying plainly. The traffic story is the one that actually moves money this year. Publishers down 20 to 70% on open-web traffic are not going to be saved by reclaiming fees on inventory that isn't getting the clicks in the first place.