Bobby Samuels CEO at Protege | Unlocking Data Access for AI Development

“When I started at Datavant, which would have been like 2020, we were 18 months away from synthetic data having a huge moment. When I left four years later, we were 18 months away from synthetic data having a huge moment.”

Summation with Auren Hoffman · Startups & Venture · October 2026

Auren Hoffman asked why synthetic data was so hot a few years ago and why fewer people talk about it now. Samuels says it's compelling when data limits scaling, but nuanced tasks need real-world data. Data made up by people or by AI comes out stilted and full of bias.

Summation with Auren Hoffman · 2026-10-13 Listen to the episode → More from Bobby Samuels → More Startups & Venture quotes →

Transcript

Summation with Auren Hoffman Around 54:27 into the episode

Bobby Samuels

…a bunch of applications subject to the same rules that make any company good. And so if you're just doing another Me Too kind of company, you're going to, it's going to be hard. But I think your point around like the, you're always going to, things don't stay static. Yeah. Things don't stay static.

Auren Hoffman

Now, there was a point, maybe, maybe a couple of years ago where everyone was talking about synthetic data. And for whatever reason, people aren't talking about that as much anymore. Like, why was it so hot a few years ago? And why is it like less hot now?

Bobby Samuels

So in the healthcare world, when I started a data van, which would have been like 2020, we were 18 months away from synthetic data having a huge moment. When I left four years later, we were 18 months away from synthetic data having a huge moment. And it always feels like we're like right there. But look, I think that there is a, there's something super compelling about the idea of in the context of a scaling of scaling laws where data is a huge constraint. If you can scale up with synthetic data, that is like a very, very compelling concept. And I think the research community has sort of gone back and forth over time to your point on whether it is valid. I think that as data has gotten more focused on not even niche use cases, but things that are a little bit more nuanced, it is clear that real world data is where things need to go. And I mean, you can think about it, like if a company wants to build an agent to create a deck, PowerPoint deck, you want to have the deck itself, but you want to have all…

Auren Hoffman

guess you kind of like, if you're the Pentagon or something, okay, like you might need to create synthetic data around like a war with an adversary because like the, okay, there is no data about, well, you know, there is no data about that. So maybe somehow we need to create it. I don't even know…

Bobby Samuels

I think that there is a place for synthetic data. The ones that I, the use case I see is like most compelling is you basically like want to, if there's like highly sensitive data that you want to be able to look at before you license or can only be shared in very sensitive situations, you basically…

Speaker 3

Yeah. But I'm open to being proven wrong. Totally, me too.

Speaker names from our own diarization · position estimated from where the line sits in the episode

More from Summation with Auren Hoffman

“The best days for the salad concepts are the Mondays in January. And the worst days are the Fridays and Saturdays in November and December. It's a complete inverse relationship. … People like to save healthy favorites because they think, I'm so good, I ate this salad and I'm going to eat this salad all the time. And they don't like to save these unhealthy, guilty-pleasure, late-night comfort foods, but they reorder them more often than they do the favorites.” — Noah Glass, Summation with Auren Hoffman

“One of the jokes we have internally is the issue with buying Spirit Airlines data is you're going to teach the models to be really good at operating a failed airline.” — Bobby Samuels, Summation with Auren Hoffman