The Next Iteration Podcast Ep. 82: William Steenbergen, CTO & Co-Founder, Federato

William Steenbergen
June 1, 2023

“Businesses don't have problems, people do. Oh, profits aren't high enough or, in insurance, loss ratios are too high? There's always someone in the company, some person who either doesn't have the tools or enough knowledge to make the right decision. That is something we truly took to heart when we started Federato. Who is the person that is making these decisions that get the company in this situation? And in this case, it clearly was the underwriters. So, that's where we started zooming in. Okay, what are the underwriters’ daily decisions, why are they making these decisions, and that's how we got to what we're building for insurers today.”

– William Steenbergen, CTO and Co-Founder, Federato. 

On this episode of The Next Iteration Podcast, William Steenbergen, CTO & Co-Founder of Federato joins co-hosts Fuad Ali and Damien Joseph to talk about building useful insurance software products with reinforcement learning and AI. William answers some really thoughtful questions and shares his journey from studying state-of-the-art algorithms and reinforcement learning at Stanford as the youngest researcher in his graduate program to co-founding Federato:

“I started out, like many people when they were going to college, not really knowing what I wanted to do,” says Steenbergen. “I loved math in high school, so I was doing a lot of that. But I also loved things like ethics… I realized that I loved the idea of modeling human problems with mathematics. And it wasn't so much about data necessarily. It wasn't so much about economics either or business for that matter. It was more just kind of trying to figure out how you could model, effectively, human problems with math. And that's how I started learning more about the field of statistics, which is really what I was passionate about.”

– William Steenbergen, CTO & Co-Founder, Federato

William dives into how his research at Stanford led directly to the development of Federato’s AI-powered RiskOps platform that helps insurance carriers and MGAs operationalize their data and optimize their risk portfolio:

“Our optimization algorithm is really the same underlying algorithm as the YouTube algorithm. If you're on YouTube, you are recommended a bunch of videos that are most likely to be watched by you, that you will find interesting. The reward function is something in the realm of getting you to stay on YouTube for a longer amount of time. Our algorithm is similar but, in our case, our reward function is some measure of a diverse and equitable portfolio. So what that means is we recommend those accounts to underwriters or those actions that they should focus on to get to that solid portfolio.” 

– William Steenbergen, CTO & Co-Founder, Federato

Steenbergen discusses how his unique educational experience in the Netherlands has shaped his ideas of learning:

“We didn't have any exams or any formal way of evaluating each other or ourselves. In normal educational programs, you get taught how to take tests, which is a completely useless skill in real life, right? We skipped all of that. We were like, No, absolutely not. Why do we even care about that? Let's focus on the thing that actually matters, which is how do you solve problems in real life." 

– William Steenbergen, CTO & Co-Founder, Federato

Steenbergen also shares what he looks for when hiring a software engineer: 

“When I hire people for Federato, the one thing I'm looking for is intrinsic motivation to grow yourself, to grow the business, to become a better engineer. That's one of the most valuable skills you can possibly have, just that intrinsic motivation and passion to actually do something well and become better at it. How do you find that? It's difficult. Usually, a really good proxy is curiosity. I ask a lot of people about their side projects and I love having conversations about that instead of, you know, can you solve this with a binary tree?”

– William Steenbergen, CTO & Co-Founder, Federato

Key Takeaways

  • The super interesting things going on right now in the field of reinforcement learning and AI such as ChatGPT 
  • How RiskOps is helping to eliminate bias in underwriting and make insurance more equitable.
  • The hardest part about starting a technology company is not thinking about a solution. It's really understanding the problem at a deep level.
  • How a popular course at Stanford called Lean LaunchPad, hosted by Bay Area billionaire Steve Blank, brought him and Federato co-founder Will Ross together in a version of startup speed dating.
“Then I met Will [Ross], and I remember his email very vividly. He was the only one that emailed where I was like, I think this is going to be the right fit because he wrote, ‘I am tired of doing ad optimization and doing anything that doesn't create any impact. I want to do something I actually care about.’”
“We love getting to the bottom of things,” Steenbergen says of co-founder Ross. “That's kind of one of the things that is core for us, getting to first principles. I had a first principle understanding of math. He had kind of a first principle understanding of climate modeling, VCs, and business. Actually, he did climate and atmospheric modeling, so he really got into climate math and modeling.”
“Did you guys actually spend 1,250 hours interviewing underwriters?” host Damien Joseph asks Steenbergen. “We actually did, yeah. That's legit. That's all on the Google Drive. We really wanted to understand what this market is about and, as I said, for us, it's all about first principles.”

Listen to the Full Episode

Tune in on Spotify, Apple Podcasts, or Soundcloud, or click the ‘Play’ button below to hear how William and the team at Federato think about solving problems for real users in the insurance industry.


(06:00) – With so many interesting problems to work on, why AI and Data?

(20:25) – Applying to Stanford for Graduate Studies 

(26:00) – The Federato Origin Story

(34:50) – Businesses Don’t Have Problems, People Do

(43:00) – Helping Kids Learn Math Through Machine Learning

(58:00) – Reinforcement Learning and the YouTube Algorithm


More on the topic

See more