Why AI Labs Pay for Data They Could Find Themselves

Bobby Samuels, co-founder and CEO of Protege · Summation

Bobby Samuels is the co-founder and CEO of Protege, which licenses real-world data, mostly from healthcare and media, to companies building AI. Protege has raised $65 million from Andreessen Horowitz, CRV and Flex Capital. Before Protege, Samuels led partnerships at LiveRamp and ran Privacy Hub at Datavant. Protege is his third company working closely with Travis May, who co-founded LiveRamp and Datavant.

Summary

Samuels argues AI labs buy data through Protege because it is faster, cheaper and better than sourcing it themselves, and speed decides whether data makes a training run. He sees real-world data, from people living their lives and companies doing their work, as the most valuable kind, while bulk text loses value and synthetic data keeps falling short. Protege is pairing its data store with a research lab and moving into benchmarks and evaluations to widen its buyer base. He picks markets where he has a right to win, sells to researchers who spot nonsense instantly and hires people who care about the mission. These signals are from an interview that Auren Hoffman did with Bobby Samuels on Summation.

1. Buying Beats Sourcing. In healthcare, Protege's biggest area, Samuels says buying through Protege is faster, cheaper and higher quality than labs sourcing data themselves. Protege has hundreds of partners, many through existing relationships, and a team that includes former Hollywood licensing executives who have sourced data from studios and sports leagues. It knows how data sets link, how to curate them and how to de-identify them, and it absorbs some of that risk. Joined, longitudinal data covering many types is worth far more than its parts. ▶ 7:36