AgenticAdvertising.org Pushes AI Media Buying; Scientists Warn of Hidden Risk

Two industry camps are now openly fighting about what AI media-buying agents should optimize toward, and the outcome will determine whether brands keep reaching new buyers or quietly stop trying.

A new group, AgenticAdvertising.org, sent an open letter to the incoming CEO of the Association of National Advertisers arguing that brands grow through deep consumer relationships. The Ehrenberg-Bass Institute, an academic marketing-science body, pushed back through scientists Dr. Bill Page and Dr. Erica Riebe: brand love is rare, relationship-first advertising has no demonstrated track record of growing brands, and an AI agent chasing a sales target will default to targeting people who already buy from you, because those are the cheapest conversions to claim. Ehrenberg-Bass has decades of empirical work behind that position. AgenticAdvertising.org has an open letter.

The practical problem is invisible on current dashboards. A ROAS-optimized agent drains budget from prospecting and pours it onto people who were going to buy anyway. CPAs fall. The CFO loves the chart. Six months later, reach is flat and new-buyer acquisition has stalled. The machine hit its target by selling more to existing customers and called it growth. No alert fires.

This is not entirely new. ROAS-optimized campaigns and retargeting pixels have faced this critique for fifteen years, and brands that cared about reach always had to force it manually. The open question is whether agentic buying makes the drift worse than rule-based systems did, because no human is watching the trade. That case is plausible but not yet demonstrated.

What makes the current moment different is that defaults are hardening. These agents are moving from pilot to production over the next 12 to 18 months. Whoever writes the objective function inside agentic buying controls brand health at scale. The Trade Desk (the largest independent ad-buying platform), Google's DV360, and Amazon's DSP all let advertisers set the optimization goal, but defaults win because most buyers never touch them. The platform that credibly bakes new-buyer rate and reach breadth into the agent as native goals, not bolt-on post-campaign reports, owns a durable advantage with the brands that already believe reach drives growth, which is a large segment of the ANA membership this letter was aimed at.

That same gap raises the value of identity infrastructure like LiveRamp (a data connectivity platform), because you cannot optimize for new buyers unless you reliably know who your current customers are. It also opens a door for measurement firms like iSpot and VideoAmp: pipe reach-quality signals into the agent's bidding loop as hard constraints, not optional filters.

The de-risking move is straightforward. Run any agent pilot with a new-buyer-acquisition metric reported alongside ROAS. Set a hard reach floor the agent cannot optimize away. If new-buyer rate holds steady, the skeptics were right and nothing was lost. If it sags while CPAs improve, you caught it before it scaled.

Our call: By end of 2027, at least one of The Trade Desk, Google DV360, or Amazon will ship an advertiser-facing optimization mode that targets new-buyer or incremental reach as a native agent goal. The product brief is obvious, the demand is public, and all three are competing hard on agent features right now. Leaving a clearly articulated advertiser need unserved while racing each other on this exact surface is not how any of them have behaved in prior feature cycles.

Two industry camps are now openly fighting about what AI media-buying agents should optimize toward, and the fight matters because the default setting on those agents will decide whether brands keep reaching new buyers or quietly stop. A new group, AgenticAdvertising.org, sent an open letter to the incoming CEO of the Association of National Advertisers arguing brands grow through deep consumer relationships. The Ehrenberg-Bass Institute, an academic marketing-science body, hit back through its scientists Dr. Bill Page and Dr. Erica Riebe: brand love is rare, relationship-first advertising hasn't been shown to grow brands, and an AI agent chasing a sales target will default to targeting people who already buy from you, because those are the cheapest sales to claim.

This is easy to undo at the level of any one campaign. It is very hard to undo as an industry habit once the defaults are set and nobody edits them. What's actually being decided here is who writes the objective function inside agentic buying, and whether reach survives as a constraint when a machine is holding the wheel. No deadline is forced, but the window is the next 12 to 18 months while these agents move from pilot to production and their default targets harden.

The Operator

If you run a trading desk or programmatic ops, this is happening on your screens now, not in theory. Any AI bidding agent in pilot is almost certainly optimizing to ROAS or conversion rate (return on ad spend, or how cheaply it books a sale). That means it is already draining budget away from prospecting inventory and pouring it onto people who were going to buy anyway. The break is silent. No alert fires. Your CPAs look gorgeous, the CFO loves the dashboard, and six months later your brand tracker shows reach flattening and new-buyer acquisition stalled. In plain terms: the machine hits its target by selling more to your existing customers and calls it growth. Hard-code reach and frequency floors as constraints the agent cannot violate. Make "exclude recent purchasers" a first-class input, not a checkbox someone forgets.

The Skeptic

Let's not pretend this is new. Every retargeting pixel and ROAS-optimized campaign for the last fifteen years has faced exactly this critique: machines chase the cheap conversion, which is the person already in your funnel. Brands that cared about reach always had to force it manually. AgenticAdvertising.org is a brand-new group with no named scientists and an open letter as its whole body of work. Ehrenberg-Bass is the credentialed voice here, so the "debate" is a bit lopsided, maybe a bit manufactured. The genuinely open question is whether agentic systems are worse at this than rule-based retargeting, and nobody has shown that. In plain terms: the risk is real, but it is the same risk we've managed for years, now with a more frightening name attached to it.

The Strategist / Market Analyst

Whoever controls the default optimization target in agentic buying controls brand health at scale. That's the prize. The Trade Desk, Google's DV360, and Amazon's DSP all let advertisers set the goal, but defaults win because most buyers never touch them. The moat goes to any platform that can credibly bake long-run growth metrics into the agent itself: new-buyer rate, reach breadth, not just bolt-on reports after the fact. Ehrenberg-Bass just wrote the product brief for a "brand growth mode," free of charge. This also raises the value of identity infrastructure like LiveRamp, because you cannot optimize for new buyers unless you reliably know who is already a customer. And it opens a selling door for measurement firms like iSpot and VideoAmp: pipe reach-quality signals straight into the agent's bidding loop as constraints. In plain terms: the companies that teach the robot what "growth" means will own a piece of every brand's future.

The CFO

Here is where my patience with the clean dashboard runs out. Agentic buying sells on efficiency, and efficiency is the exact thing that makes this dangerous. A falling CPA looks like money saved. What it often means is you stopped paying to acquire new customers and started paying to harvest the ones you had. That is spending your future growth and booking it as a win this quarter. The payback math on an agent that shrinks your buyer base is negative, you just don't see it for two to four quarters. Any pilot needs a new-buyer-acquisition line sitting right next to the ROAS line, or the number you are optimizing is lying to you.

Tensions

Three real disagreements. The Skeptic says this is fifteen-year-old retargeting dynamics with a scary new label, while the Operator says the automation and the invisibility of the break make it materially worse because no human is watching the trade. The Strategist sees a product opportunity worth building toward, while the CFO sees a measurement trap that every current dashboard actively hides. And underneath it all: is brand reach a genuine growth engine, as Ehrenberg-Bass argues, or a comfortable belief that performance marketers have been right to ignore? The whole case for forcing reach constraints collapses if reach doesn't actually drive growth.

Synthesis

This hinges on two beliefs. One, that broad reach genuinely drives long-term growth, which Ehrenberg-Bass has decades of empirical work behind and the relationship camp does not. Two, that handing optimization to an autonomous agent makes the drift toward existing customers worse than rule-based targeting did, which is plausible but undemonstrated here. The council leans toward the Operator and the CFO: the mechanism is sound and the damage is invisible on current dashboards, which is precisely what makes it worth fixing before it scales. What to de-risk: run any agent pilot with a new-buyer-acquisition metric reported alongside ROAS, and set a hard reach floor the agent cannot optimize away. If new-buyer rate holds steady under the agent, the Skeptic was right and you've lost nothing. If it sags while CPAs improve, you caught it early.

Prediction: By the end of 2027, at least one of The Trade Desk, Google (DV360), or Amazon will ship an advertiser-facing optimization mode that targets new-buyer reach or incremental reach as a native agent goal, not a post-campaign report.

Confidence: Medium. The product brief is obvious and the big three compete on agent features, but timing could slip past 2027.

Why: The Ehrenberg-Bass warning names a specific defect in how agents optimize: chase a sales target and the machine defaults to existing customers because they convert cheapest. That is a concrete product gap, and all three major DSPs are racing to differentiate their agentic buying layers right now. Whoever offers advertisers a credible "grow my buyer base" toggle wins the brands that already believe reach drives growth, which is a large and vocal segment of the ANA membership this letter was aimed at. The opposite outcome, that none of them build it, would require the biggest platforms to leave a visible, publicly-articulated advertiser demand unserved while competing hard on exactly this surface, which runs against how they've behaved on every prior feature race.

Revisit by 2027-12-31: We're right if The Trade Desk, Google DV360, or Amazon DSP publicly launches an AI/agent optimization target built around new-buyer or incremental reach by December 31, 2027. We're wrong if all three still offer only conversion/ROAS-style goals with reach available merely as reporting or manual constraints.