Future of Marketing Briefing: Agentic advertising is closer than you think and further than you hope

From his vantage point — Nvidia sits inside the infrastructure of the major open internet advertising platforms — the technical capability is there. The bigger hurdle is fiduciary. Any system making autonomous financial decisions needs guardrails that are deeply specific, not just bolted on, and right now most of what’s being called “agentic” in advertising doesn’t clear that bar. 

Allen was deliberate about that distinction. There’s a meaningful difference, he continued, between a well-structured automated workflow managed via conversational interaction and genuine agency — and the industry is conflating the two, much like it did with generative AI before it. Getting to real autonomy requires reinforcement learning and guardrail work that’s very specific to advertising’s compliance and decisioning requirements.

“There is that requirement to make sure that all the checks and balances are being put in place before those things are really highly activated, and that’s where that next level of consideration of how much agency do we give becomes very important,” said Allen.

Understanding why those checks are so hard to define in marketing specifically requires understanding two things: the first is decision complexity — most marketing problems operate across a smorgasbord of variables simultaneously, a scale at which both humans and most AI systems become unreliable without the right support beneath. The second is what happens when you deploy at scale: marketing exists to change behavior, which means the moment a system starts acting on its prediction — buying, targeting and optimizing — it shifts the very environment it was trained on. The past stops predicting the future. That’s not a fixable engineering problem. It’s structural to what marketing does, and it’s why the guardrail question is harder here than in almost any other industry.

“There has to be a huge amount of intelligence input to those systems that is very specific, which brings us back to some of that work that we’re seeing in the startup space around reinforcement learning and guardrails that are very, very specific,” said Allen. 

Where it’s moving fastest, Allen said, is in the startup ecosystem where companies are building from scratch, unencumbered by legacy systems, grounding agent workflows in accurate data through synthetic data generation, vector embedding and reinforcement learning rather than retrofitting new capabilities onto old infrastructure. That’s the work that needs to mature, and the companies doing it are the ones to watch.

None of that matters, though, if the standards don’t keep pace. The Agentic RTB Framework and Ad Context Protocol are early attempts to lay that groundwork — early being the operative word, with neither spec broadly adopted. Still, Nvidia is trying to move things along from its end too, having announced a coalition of model builders, labs and academic institutions earlier this year to define what secure agents look like, with the underlying work being open-sourced.

“The advertising and marketing industry has a very high value opportunity to get there very quickly,” said Allen. 

The economics make the case better than any keynote could. Every unnecessary hop in the programmatic supply chain costs money — in fees, in data fidelity and in targeting accuracy. Agents that can negotiate directly, route around intermediaries and optimize in real time don’t just do the job faster. They do it cheaper and with better results. Butler/Till saw it in its first live test: lower than expected CPMs, premium supply, rapid execution. The efficiency case, in other words, is no longer theoretical. 

“Yes, broadly speaking that reflects a lot of the conversations happening across the industry right now,” said James Chandler, chief strategy officer at the Interactive Advertising Bureau U.K. “There’s a lot of excitement in the industry about the opportunities agentic transactions could offer.  At the same time, it’s certainly true to say that the industry is still defining what ‘agentic’ really means in practice. A lot of what’s currently being described that way is still sophisticated automation rather than fully autonomous decision-making, and the distinction matters.” 

His view won’t surprise regular Digiday readers. Time and again ad execs have told us the same thing over the last several months: the tech is ready. The industry isn’t. 

Wes ter Haar, chief AI officer at S4 Capital’s Monks, saw that up close on a recent call with a client’s technology and security team. They wanted to know about end-to-end autonomous workflows. His answer: technology could do it. The marketing team wasn’t ready to sanction it. The most eager, he noted, tend to be in financial services — fintech clients, in particular, who are “most interested in just having the machine run itself.”

But it’s not just financial services at the front of that queue. It’s automotive and smaller players too — categories facing the kind of structural pressure that makes big swings not just acceptable but necessary. You don’t experiment cautiously when the alternative is getting left behind.

Publishers understand that dynamic better than anyone. They’ve been on the wrong end of enough technological shifts to know what happens when you wait. Which is why some of them aren’t waiting this time. News U.K. and CNN are already scoping out their own sells-side agents, getting the infrastructure in place ahead of the moment when advertisers and agencies arrive with their own.

For its part, Omnicom is further along than most — or at least louder about it. Its executive team has made no secret of the fact that they’re already running ad campaigns autonomously through agents with real clients and real money. Granted, they’re small and experimental for now but they won’t stay that way. The efficiency gains and economic upsides will make sure of that.

“In the tests that we’ve done, we’ve seen significant increases in effectiveness in the areas of the media buys that we’ve done,” said Omnicom’s chief technology officer Paolo Yuvienco at the the J.P. Morgan Global Technology, Media and Communications Conference earlier this week. “Now, again, it’s very early days, so it’s still kind of, we’re still laying the foundation for this all, but what we’re ensuring is that we are laying the path so that our clients can leverage that once it becomes a scaled offering.”

The same dynamic is playing out across the board, albeit at different speeds. Many marketers aren’t anywhere near ready to hand that kind of autonomy to an agent. Many others are actively testing how much they can and how much human oversight is actually required once they do. 

What’s becoming clearer through it all is that agentic trading, when it does inevitably scale, will look less like a replacement for programmatic and more like its next iteration — addressable, biddable and optimizable, just with different pipes and a shorter supply chain.

“Encouragingly, many of those conversations are already happening relatively early in the development cycle,” said Alex Kozloff, director of industry relations at the IAB U.K. “Continued collaboration between industry bodies, standards organizations and technology companies will be important to making sure innovation and trust evolve together.”

Rude health, stubborn budgets: the state of out-of-home advertising in the U.K.

Despite years of industry enthusiasm, out-of-home advertising has stubbornly refused to break through its 3-5% share of media budgets. But there are signs that may be changing — a genuine shift toward brand-building, smarter creative, and better measurement practice is building the case for the channel in ways that previous inflection points didn’t.

Bauer Media Outdoor’s research and insight director Lindsay Rapacchi, has spent nearly a decade in OOH and thinks the tipping point is closer than it’s ever been  if the industry can close the gap between the story it tells and the evidence it produces.

The following conversation has been lightly edited for clarity and brevity.

On the enthusiasm-to-spend gap 

“There’s a huge amount of love for the channel, and media owners have backed that with massive investment in digital. But it’s not translating into more money. We’re still sitting at around three to five percent. There’s this tipping point we need to reach where people’s enthusiasm is backed by their wallets.”

On what’s different in briefs now:

“What excites me most is seeing the fundamentals of marketing science in the briefs coming through. More people are talking about penetration, reaching category buyers en masse, category entry points. For a long time you’d get a brief saying ‘I want to build trust’ or ‘add awareness’ — all very well, but that’s not the same thing. Now I’m seeing people genuinely act on the science, and that makes it much easier to make the case for out-of-home.”

On macro headwinds:

“We’ve weathered recessions, we’ve weathered COVID. And one thing that helps is that whenever economic uncertainty hits, the rhetoric around maintaining brand presence comes through hard and fast. We revert to that story — because I believe in it, and because our performance through those periods proves we can weather the storm.”

On the measurement hierarchy:

“There’s a lack of knowledge about how robust different measurement methods actually are — not a lack of methods. MMMs are at the bottom of the hierarchy. If you talk to any econometrician, they’ll openly tell you it’s a statistical guess. Yet people are making very important decisions based on an ROI figure with little understanding of how it was produced. Above that you’ve got mobile proximity and brand lift studies, which are convenient but fraught — most of those screens are digital, and your test and control groups end up more similar than different. The most robust approach is a geographic holdout experiment. But for a channel that’s three to five percent of plan, most clients won’t commit to it. You can’t have it both ways.”

On AI as an indirect tailwind:

“There’s an intuitive sense that real-world experiences have value that’s harder to argue for things like AI. A physical screen that can’t be avoided is a valuable commodity. But we need to evidence that quickly — it’s one thing to tell the story, it’s another to show that out-of-home delivers effects that AI-generated environments aren’t going to replicate in the same way. That’s not to say they don’t do a job in themselves — but we need to isolate what out-of-home uniquely brings.”

What we’ve heard

“The opportunity here isn’t simply about automating existing processes faster. Done well, these technologies could help streamline workflows, improve efficiency and build on the strong foundations already in place across digital advertising.”

—Alex Kozloff, director of industry relations at the IAB U.K. 

Numbers to know

$1 trillion: The reported valuation OpenAI is aiming for when it IPOs

41%: Percentage of U.S. adults that still prefer shopping in store to a retail app

RM10 million ($2.5 million): The amount that Malaysia has threatened to fine TikTok over content moderation failures

8,000: The number of job cuts in Meta’s latest round of layoffs / company reorg

What we’ve covered

OpenAI gives ChatGPT ads a visual upgrade

Brands can now customize how their ads appear in ChatGPT, including format and CTA elements. The update gives advertisers more control over creative presentation in testing.

Pitch deck: X leans on AI and performance in a bid to win ad dollars

X is pitching advertisers on a familiar promise: “scale plus AI equals performance.” The deck highlights improved CTRs and conversions, alongside its revamped AI-driven ads manager, but relies heavily on internal performance claims. Overall, it frames X as a full-funnel, outcomes-focused platform in an effort to revive ad spend.

Inside The Trade Desk’s Claude-powered campaign agent

TTD is testing “Koa Agents,” an AI system powered in part by Anthropic’s Claude that can automate key parts of programmatic campaign setup and optimization. The agents can ingest media plans, build campaigns and troubleshoot issues, but still require human approval before execution.

‘Moved that drop dead date’: Omnicom accelerates LiveRamp exit after Publicis deal

Omnicom is speeding up its exit fromLiveRamp after Publicis Groupe’s Lotame deal reshaped the data landscape. The move reflects shifting data alliances across major holding companies.

What we’re reading

Before Mass Layoffs, Meta Reassigns 7,000 Workers to Focus on A.I.

The New York Times reports that Meta is reorganising around AI, reassigning around 7,000 employees into AI-focused teams while cutting thousands of other roles. The restructuring reflects CEO Mark Zuckerberg’s push to make AI the company’s central priority.

Elon Musk Loses Landmark Lawsuit Against OpenAI

Elon Musk lost his lawsuit against Sam Altman and OpenAI after a jury ruled the case was filed too late. The lawsuit accused OpenAI of abandoning its nonprofit mission as it commercialised AI, per Wired.

OpenAI Is Preparing to File for an IPO Very Soon

The Wall Street Journal reports that OpenAI is preparing to confidentially file for an IPO, potentially within days, with a listing possible later this year. The move follows a legal win against Elon Musk.

CapCut is partnering with Gemini to let you edit videos without editing videos

CapCut is integrating Google Gemini to enable AI-powered, prompt-based video editing. The update lets users make edits using natural language instead of manual tools, according to Adroid Authority.