[Column] Juan Baron: AI is only as good as its data

AI can make the media decision, but the data to support it sits elsewhere

Media teams are handing AI a growing share of the decisions that shape a campaign. Systems can adjust bids, reallocate budgets, refine audiences, and read performance signals at a speed no team could match by hand. At the same time, planners set the goals and guardrails those systems work within. The quality of each decision still depends on the information behind it.

Data sits in many hands

The data needed to make and measure a media decision rarely sits in one place. Brands hold first-party customer data. Agencies hold planning and buying data. Publishers hold audience and contextual signals, retailers hold purchase data, and measurement providers hold the results that show whether anything worked. An AI system that sees only one of these pieces is optimizing with a partial view.

The obvious fix is to pool everything in one location, and the industry's realities rule that out. Privacy regulation, contractual limits, competitive sensitivity, and internal governance rules all restrict what organizations can move or expose, and each party has a legitimate interest in keeping control of its own data. The industry now has to ask whether the infrastructure exists for AI to work across organizational boundaries, with permission.

Permission and control come first

Consider a brand that wants an AI system to optimize spend with a retailer. The brand holds customer data and the retailer holds purchase data, and neither can hand its data to the other. The system can only judge whether the campaign worked if it can see both, so the arrangement has to define in advance what each side contributes, what the system may compute, and what comes back out.

That infrastructure has to answer questions most current setups leave open. Who is allowed to use which data, and for what purpose? How can a party verify that its data was used only as agreed? When an AI agent acts on a combined view, how do the participants audit what happened? And can the output be trusted by everyone involved, including the parties that did not run the analysis?

Measurement needs a neutral place

Measurement raises the stakes further. If the system that makes the buying decision also grades its own results, independence disappears. Measurement across brands, agencies, publishers, and retailers needs a neutral place where each party can contribute data and still trust the outcome. This will matter more as AI takes on more of the decision-making, because the cost of an unverified result grows with the budget it controls.

Confidential computing offers a practical direction here. In a data clean room built on this technology, participants collaborate around data while each keeps control of its own, and computation is limited to what everyone agreed in advance. The same controls can apply when AI is part of the workflow, so an agent can work with a richer view without any party giving up ownership.

The organizations that settle questions of permission and control early will let their AI operate on better data. Those that leave them open will find their models limited by what they can see, however capable the models become.

Juan Baron is Head of International and Global Accounts at Decentriq

 Follow Marketing Report on LinkedIn

 Subscribe to the Marketing Report Newsletter

 Send us your press releases, columns, interviews…

 Join Marketing Report in our LinkedIn Groups