Yeah

... area, I would contend that it is faster, cheaper, and higher quality to go through us than to try to do it yourself.

So we know all the data sources or a lot of them.

We understand the linkages between them.

We understand what's the process by which it needs to be curated.

We understand the process by which how you think about de-identification.

There is certainly some risk buffering, too, where we are taking on certain risks around de-identification that they don't need to.

What have you learned from Travis?

Can you explain that a little bit more in detail? Like when is synthetic data useful? When do you need real world data?

Yeah.

So Synthetic data, like the way I think about it is there are like three buckets.

People would describe synthetic data differently, but this is the way I conceptualize it.

You have synthetic data, which is data that's created by models to train models.

You have manufactured data, which is data basically created by people to train models.

Then you have real world data, which is data that is an exhaust from people living their lives and companies doing their things.

And healthcare was a developed data market mainly because of pharma.

And so do you have technology in here or is it very much just relationships like I connect this company to that company? It's a combination.

So we have about 150 data partners at this point across different parts of the data world.

We work in healthcare, we work in media.

So sort of a lot of the killer features, so to speak, is the data liquidity that we've built.

The technology is also a significant component of what we do.

Because the raw data is often critical, but it needs to have additional work on top of that.

In video, for instance, we've seen the desire for large amounts of content be consistent, but then there's also desire for smaller bits of content that have... differential characteristics like the wind blowing through trees or things like that.

Bobby, say more about that.