SK Solomon Kahn On The Data Stack Show

“We still had a large operations team supporting us afterwards, which is one of the things that people don't appreciate about AI: the biggest challenge when you actually run a real business on AI is how you navigate the 5% of time that things are not right. And in some cases, you actually need a substantial effort in order to do that. It's still worth it, but you do need a substantial effort to manage that.”

The Data Stack Show · Hands-On Engineering Podcasts · November 2026

“We still had a large operations team supporting us afterwards, which is one of the things that people don't appreciate about AI: the biggest challenge when you actually run a real business on AI is how you navigate the 5% of time that things are not right. And in some cases, you actually need a substantial effort in order to do that. It's still worth it, but you do need a substantial effort to manage that.” — Solomon Kahn, The Data Stack Show

Context

Solomon is describing how his division at Nielsen moved from people manually tagging logos in sports broadcasts to a computer-vision system. He explains that the humans did not go away, in a product customers paid for and built their own businesses on.

Transcript

The Data Stack Show Around 18:43 into the episode
Solomon Kahn

But we shifted, and I know you have a lot of engineers here, so they'll appreciate this. We shifted our analysis from essentially manual. So we had an operations team that was a thousand people plus drawing boxes around logos, sports content in order to actually generate this data set. It's like

John Wessel

thousands and thousands of hours.

Solomon Kahn

Well, all of all of professional sports across the world, right? All the big ones. And we were only able to get, so we use this algorithm. It's like a, it's like a way of looking at brands to find them, like comparing images. So it's called SIFT. It's like, it's like a way to see, okay, I want an exact match to this image, even if it's bigger or smaller to find an exact match. But sports is not good at finding exact matches because the jersey patch, when you turn like this, even if it looks different, right? Right. And so we bought an AI company that was very good at computer vision and we acquired the company. And then we had that team work with us to build a computer vision version of, you know, and by the way, we still had a large operations team supporting us afterwards, which is one of the things that people don't appreciate about AI is you have to, the biggest challenge when you actually run a real business on AI is how you navigate the 5% of time that things are not right. And in some cases, you actually need a substantial effort in order to do that. It's still worth it, but you do need a substantial effort to manage that.

Eric Dodds

And for y'all, that was direct, that was actual dollars that are tied to managing that 5% in the long tail. Oh, because, okay, like a customer gets a response that's not super helpful from an AI agent. It's like, okay, I mean, that kind of stinks and happens enough. Maybe they're not a customer…

Solomon Kahn

Not only are we charging people, but like they're charging. It's like

John Wessel

supporting all of the exactly.

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.