Your AI-Built Marketing Mix Model Is Fast. But How Do You Know It's Right?

Inside the Stack

AI has cut the time it takes to build a marketing mix model from months to days, Bharadwaj says, which makes expert judgment more important, not less. Anyone can build a million models now. Someone still has to decide which one to believe.

His method is triangulation: run MMM and incrementality tests side by side and see where they agree. Incrementality tests run on real people, so the results are never clean. “If you get a very clean read, you probably got something wrong,” he says.

His answer is to start with a thesis about your go-to-market and wrap measurement around it, so the data can validate it or break it. He points to a beauty brand whose AI-built model recommended cutting $30 million from its Amazon ad spend. SKU-level modeling told a different story: each SKU behaved like its own market, and the first round of work found 5–7% top-line growth.

Also on this episode: Why finance should be able to read marketing results, what causal attribution means in practice, and why Bharadwaj thinks measurement is headed toward super cognition.