The Agentic Media Race Is Really a First-Party Data Race
Agentic advertising promises to transform how media is planned, bought and optimized, but AI agents are only as effective as the data they can access. As agents become ubiquitous, the real competitive advantage will shift from the technology itself to the proprietary first-party signals that help those systems make smarter, more relevant business decisions.
This Signal & Noise exclusive is brought to you by AUDIENCES.
Year in, year out, there is always a handful of words that seem to take over Advertising Week. A few years ago it was identity. Then it was clean rooms. Last year, you could barely make it through a conversation without hearing about generative AI.
This year, the word was definitely agentic. I spent much of Advertising Week New York hearing about AI agents that can build media plans, discover audiences, negotiate with sellers, shift budgets, optimize campaigns, generate creative and increasingly make decisions that once required teams of people bouncing between spreadsheets, dashboards and buying platforms.
Hype aside, of which there was a lot, I don't think this is just another round of conference hyperbole. The hype is real because the technology is actually beginning to move from PowerPoint into actual media workflows. Buyers and sellers are experimenting with agents that can transact with one another. Platforms are embedding AI assistants into campaign management, and agencies are building their own agentic systems. The direction of travel is becoming increasingly clear: more of the decisions involved in planning, buying and optimizing media will eventually be made—or at least heavily influenced—by machines.
But amid all the excitement, I kept coming back to a much more basic question: What does the agent actually know? Because an AI agent can be extraordinarily good at optimizing toward the information it has been given, if that information is bad it will probably make a terrible business decision.
For example, a media platform may know that someone clicked an ad yesterday. It may know which sites they visit, which content they consume, or which audiences they resemble. It may also know that one impression is cheaper than another, or that a certain combination of creative and placement historically produces a higher conversion rate.
But does it know that this person became a customer three days ago? Does it know that another customer has purchased six times and is among the most profitable customers the company has?
Does it know someone is likely to churn, owns a particular product, has an unresolved service issue, isn't eligible for an offer, or has withdrawn permission for their data to be used in a particular way?
The reason for this gap is that, in most organizations, intelligence doesn't live in the ad platform.
It lives in the company's first-party data. This is why I think we're looking at agentic advertising slightly backwards. I get it, the industry is fascinated by agents. But as those agents become more widely available, the agent itself is unlikely to remain the competitive advantage for very long. Google will have agents, Amazon will have agents, and Meta will have agents. DSPs, agencies, publishers and every major marketing technology provider will also have agents.
What these systems won't all have is the same proprietary understanding of your customers and your business. The agentic media race may look like an AI race. Increasingly, it's going to be a first-party data race.
The Difference Between a Signal and a Customer
Advertising platforms are incredibly good at collecting signals. An impression, a click, a search, a video view, a visit to a product page, an inferred interest, or a resemblance to other people who converted. Marketers find these signals enormously valuable because they can use them to target advertising. In fact, modern digital advertising was built around collecting them, combining them, and using them to predict what someone might do next.
But a signal is not the same thing as understanding a customer. A retailer may see someone repeatedly browsing running shoes and reasonably conclude the person is a high-intent prospect. Its media agent might therefore decide to bid more aggressively to reach them. But the company's own data might tell a very different story. The same person may have bought the shoes yesterday, or may be a loyalty member who purchases every month regardless. They may have returned the product twice, or be a high-value customer who should be receiving an entirely different message than the acquisition ad the media platform is preparing to serve.
This distinction becomes much more important when machines are making decisions. When humans manage campaigns, incomplete information creates inefficiency. A media buyer might waste some impressions, target the wrong audience, or discover after the fact that a campaign had been optimized toward the wrong outcome.
An autonomous agent can make the same mistake at machine speed. Give an agent incomplete context and it doesn't become less capable. In some ways, it becomes more dangerous because it can act on that incomplete context faster, more consistently, and across thousands or millions of decisions. That means the next generation of advertising systems needs more than media intelligence. It needs customer context.
What I mean by customer context is purchase history, loyalty status, product ownership, customer lifetime value, propensity, churn risk, service interactions, eligibility, consent, preferences, margin, and so on. None of these data points are traditionally thought of as media signals—yet they may ultimately be far more important to an advertising agent than another click, impression, or inferred interest.
The companies that figure out how to make this information safely and intelligently available to agents will give those agents something their competitors cannot easily replicate: a proprietary understanding of the customer.
Having the Data Isn't Enough
Of course, simply having a lot of first-party data doesn't necessarily solve the problem. Most large companies already have enormous amounts of customer data. The problem is that it's scattered across CRM systems, CDPs, loyalty platforms, commerce systems, call centers, data warehouses, consent platforms, and dozens of other applications. Some of it is updated in real time, while some of it is days or weeks old. Different systems may disagree about who the customer is, what they bought, or whether they can even be contacted.
We have spent much of the last decade trying to connect this information for humans, albeit with limited success. The next challenge is making it usable by machines.
This requires more than giving an agent access to a database. An agent needs to understand which customer record to trust. It needs to know which data it is allowed to use for a particular purpose. It also needs to understand that a purchase made five minutes ago should change an acquisition decision, that an unresolved customer-service issue may change the message someone receives, or that a customer who has opted out of a particular use of their data should simply be excluded altogether.
In other words, agents don't just need data. They need governed context. This is where a lot of the work companies have already done around identity, customer data platforms, clean rooms, consent, and data infrastructure starts to look less like yesterday's MarTech plumbing and more like the foundation for agentic advertising.
The companies with the biggest advantage may not be the ones with the flashiest AI agent. They may be the ones that have spent years building the infrastructure required to tell that agent, accurately and responsibly, what is actually happening with the customer. Because in an agentic world, the quality of the decision will increasingly depend on the quality of the context behind it.
The Competitive Advantage Moves Upstream
This is also why I suspect the competitive advantage in agentic advertising will move upstream. Today, much of the industry's attention is focused on who has the best agent. Which platform can build the smartest media plan? Which agent can optimize fastest? Which system can negotiate with sellers, generate creative, or move budget across channels most efficiently?
These differences matter today because the technology is still relatively new. But over time, many of these capabilities are likely to become widely available. The largest advertising platforms will build them directly into their products. Agencies will develop their own. DSPs, publishers, data companies, and MarTech vendors will do the same. Open-source models and agent frameworks will continue to improve. What looks differentiated today may eventually become table stakes.
The harder thing to replicate is the information those agents are working with. Your competitors can buy access to many of the same models and use many of the same media platforms. They can hire agencies with similar technology, and they may even be able to purchase much of the same third-party data.
What they cannot easily reproduce is twenty years of customer relationships, transaction history, loyalty behavior, product ownership, service interactions, preferences, and the accumulated knowledge of how your business actually works. That is proprietary context. And once agents become capable of acting on that context in real time, first-party data starts to become more than something marketers use to build audiences or improve match rates. It becomes part of the decision-making infrastructure itself.
An agent deciding whether to spend another dollar against a customer should know not only whether that person is likely to click or convert, but whether the company actually wants to acquire, retain, upsell, suppress, or engage them at that moment.
This is a fundamentally richer problem than media optimization. And it means the companies that have spent years building strong first-party data foundations may discover that the payoff is much larger than they originally expected.
The Real Agentic Advantage
This brings me back to the question I kept asking myself throughout Advertising Week: What does the agent actually know? Over the next few years, marketers are going to spend enormous amounts of time evaluating agentic platforms, building agents, connecting them to media systems, and figuring out which decisions they are comfortable allowing machines to make.
I am arguing they should spend just as much time thinking about the information those agents will use to make those decisions. Which customer data can they access, and how current is it? Can they resolve identity accurately enough to know who they are dealing with, and do they understand consent and eligibility? Can they distinguish between a prospect, a new customer, a loyal customer, and someone the company should probably leave alone? And, perhaps most importantly, can they translate all of that information into context an agent can actually understand and act on?
These questions may end up mattering far more than which agent has the most impressive demo. Because agents will proliferate, models will improve, and features will be copied. Much of the technology that seems extraordinary today will eventually become widely available.
Your customer relationships won't. The companies that win in agentic advertising will not necessarily be the ones with the most sophisticated AI. Instead, they will be the ones that give AI the best understanding of their customers, their business, and the decisions they are trying to make. The agentic advertising race may be powered by AI, but it will increasingly be won with first-party data.
------------------------------------------------
Rio is an executive with 20+ years at the intersection of strategy consulting, AdTech, data, and media. He's a trusted advisor on customer experience, digital strategy, and marketing transformation. He's a partner at Credera, Omnicom's consulting arm. He's also a podcast host, writer, and public speaker focused on the future of advertising and AI-driven infrastructure.
Comments