Building an agentic system and a learning loop around customer IP

Kantar has spent three decades building metrics and frameworks to calculate a signal for whether an ad works.

Turning that signal into a system that improves ads meant building graders, evals, and a learning loop—with the realization that every component inside the loop is its own optimization challenge.

The scenario: A long history of rich data and domain expertise Kantar is a leading brand intelligence company.

For more than three decades, it has helped brands answer a deceptively simple question: will this ad work?

Before AI, that kind of ad testing meant recruiting consumers, showing them the creative, running focus groups, and interpreting the results.

And no company did it—or still does it—better than Kantar.

Asking real people how an ad influenced them—and doing it at scale for 35 years—produced the LINK database of 35 million consumer interactions and a testing database of over 260,000 ads.

…