At Cannes 2026, the conversation shifted abruptly from marveling at AI's potential to demanding proof that it actually drives business outcomes. Marketing executives and agency leaders spent the week trying to bridge the gap between the capabilities of new technology and the measurable ROI their finance teams require. The pressure to justify these investments is mounting. We're moving past the experimental phase of artificial intelligence in advertising. The focus is now entirely on accountability.
The challenge is that an algorithm can't generate accountability on its own. It requires a foundation of reality.
The Amplification Engine
Modeling is amplification. When you amplify a clean signal based on ground-truth purchase behavior, the patterns you find are real. If you amplify a noisy signal based on inferred behavior, you just make the noise louder. You end up finding patterns in the artifacts. Both approaches amplify the underlying information, but they produce vastly different business outcomes.
Artificial intelligence accelerates this dynamic. It processes vast amounts of information and scales it rapidly across media plans, audience segments, and optimization strategies. When the foundation relies on proxy metrics or assumed intent, the AI will confidently optimize toward the wrong goal. It will find the most efficient way to reach people who look like they might buy, rather than people who actually do.
This is why the source of the data matters more than the sophistication of the model. A highly advanced algorithm fed on probabilistic guesses will still produce a guess. If a retail media team tries to understand whether a campaign drove incremental buyers using inferred data, the AI will simply scale the uncertainty.
The Cost of Inferred Data
Many current applications in our industry rely on assumptions about what a consumer might do next. They look at browsing patterns, location pings, or content consumption to build a profile of intent.
When an algorithm scales these guesses, the resulting audience or insight drifts further from reality. The model finds correlations that look impressive on a dashboard but fail to translate into actual revenue. This creates a persistent disconnect between the reported performance of a campaign and the actual sales figures a company records. We see this frequently when brands audit their legacy measurement frameworks. They discover that their models have been optimizing for engagement rather than conversion.
As I noted recently when discussing the SaaS market shakeout, the companies that survive this transition will be the ones grounded in reality. The market is losing patience with platforms that grade their own homework using disconnected metrics. Advertisers need to know that their media investment is driving incremental growth. They need proof, and proof requires a better input.
Ground Truth as the Standard
Accountability requires a different starting point. Ground truth means beginning with observed, verified purchases from permissioned consumers.
When you model from a large set of deterministic data, you hold the AI accountable to actual business outcomes. The algorithm learns from what people actually bought. It identifies the subtle behavioral shifts that precede a real transaction. It scales an audience based on verified reality. All data companies scale their data to provide a more complete view, but the quality of that scaled audience depends entirely on the quality of the seed data.
This approach changes what's possible for a marketing team. Instead of hoping an AI agent finds the right proxy metric, you can direct it to optimize for verified sales. You can connect media exposure directly to the outcomes that matter to your business. The technology becomes a tool for scaling certainty rather than scaling assumptions.
We're seeing this shift play out across the industry. As marketers return from Cannes 2026, the industry has stopped asking whether AI belongs and started asking what it's worth. They're auditing their data supply chains. They're asking hard questions about where the information originates and how it's validated. They want to know if the audience they're buying is based on a verified transaction or a modeled assumption.
The Output Reflects the Input
The marketing industry has the technology to process data faster than ever before. The opportunity now is to ensure that data is worth processing. We've reached a point where the limiting factor is no longer computing power. The limiting factor is the integrity of the consumer data feeding the system.
If you want your AI investments to deliver measurable ROI, start by examining the signal you're asking it to amplify. A model built on observed consumer behavior will always outperform one built on inference. The output will always reflect the input.
It's time to hold our algorithms to a higher standard of proof. Look closely at the foundation you're building on, and make sure it's grounded in reality.
.png)
