AI Only Sees What You Let It See: The Data Problem No One Wants to Admit

_{Estimated reading time: 4 minutes}

There’s a phrase that’s becoming increasingly common among marketing teams: “We’re going to use AI to optimize our campaigns.” It certainly sounds good… The problem is that most of the time, no one asks the following question: What data will that AI actually be analyzing?

Because AI isn’t magic—it’s a decision-making system that’s only as good (or as bad) as the information it has to work with.

Last week, Aravind Chandrasekharan, SVP of Engineering at The Trade Desk, published a reflection in The Current that deserves more attention than it has received. If the personal computer was the bicycle of the mind, AI is proving to be something more like a telescope for marketing: it doesn’t change the nature of the discipline, but rather how much of it we’re able to see.

The metaphor is good, but it overlooks an important detail: a telescope pointed in the wrong direction is useless.

The problem isn’t AI; it’s the data architecture.

AI is only as good as the data it has to work with. Within large-scale, objective platforms, AI can evaluate the full context of media buying (audiences, inventory, performance, and measurement) against an advertiser’s goals. Outside of that context, it operates with incomplete information.

To put it another way: if your AI only has access to data from a “walled garden”—whether it’s a social media platform, a retailer, or your own ad server—its decisions will be driven by the interests of that garden, not yours.

Closed systems are optimized for their own inventory and their own economic interests. AI does not eliminate bias; it amplifies the bias that already exists in the underlying system.

Here’s the real challenge for brands in 2025: it’s not about adopting AI, but about ensuring that AI has real insight into what matters.

From managing dashboards to leading agents

We are witnessing a paradigm shift in how industry professionals interact with technology. Instead of simply browsing dashboards, they are beginning to ask questions: Why isn’t this campaign meeting its targets? Which audiences are converting? What should be changed to improve performance?

This change isn’t just about efficiency; it’s a shift in role—from managing tools to driving results. Less time pressing buttons, more time making strategic decisions.

It’s already happening—it’s the present—and organizations that continue to organize their data into silos (a CRM here, a CDP there, a disconnected measurement layer) are going to be left behind.

The question you should ask yourself this week

Before you start thinking about which AI tool to adopt, there’s a more basic question to answer: What can your AI see today?

Do you have access to actual conversion data, or just intermediate metrics? Are you connected to first-party audience signals, or do you rely on third-party segments that are becoming increasingly unreliable? Can you cross-reference what’s happening in paid channels with what’s happening in owned channels?

Systems built on open and interoperable data create the conditions for better outcomes. Closed systems, by design, limit visibility and restrict what AI can truly understand or achieve.

In an environment where AI makes millions of decisions per second, data architecture is no longer a technical decision—it is a strategic one.

From MIO One

Alejandro Guerra, Digital Data Analyst, shares his perspective from MIO One: “In most of the projects we launch, the first hurdle isn’t technological—it’s the data. Not because there isn’t any, but because it’s fragmented, ungoverned, and disconnected from one another. Artificial Intelligence doesn’t solve that—it actually makes the problem worse. Before discussing models or platforms, we need to be clear on which signals are reliable, which ones actually predict behavior, and how they’ll be fed in real time. That conversation has to happen before any decision about tools is made.”

Communication MIO One