Welcome to The Flux, where we hear stories from people who have asked what if questions to better understand the world and talk about how data can help tell stories that impact decisions and create an intentional impact on the future. This is your host, John Cordier, CEO at Epistemix. In a world where the flux capacitor from back to the future does not yet exist, people have to make difficult decisions without always knowing how the future will play out.

Our guests are people who've taken risks, made decisions when uncertainty was high, and who have assisted decision makers by using data and models. We hope you can turn lessons from our podcast into foresight, so you or your organization can make better decisions and create an intentional impact for others.

Hey there. Welcome to The Flux, where we hear stories from people who have asked what-if questions to better understand the world and learn how data can help tell stories that impact decisions to create an intentional impact on the future. This is your host, John Cordier. Today we're on with Les Craig, who graduated West Point in 2003, spent 30 months deployed in uniform and in government service with his applied math and computer science background.

He applied that to help the warfighter, started a company, got venture backing, and is now at Next Frontier Capital, making decisions on what companies to invest in and who's going to impact the future. Les, great to have you on today,

Les: John, it's such a pleasure to be on. That's the most concise and focused intro I've ever had. Well done, sir.

John: I had some help from somebody on the podcast to get it done, so appreciate that. As we were talking a little bit beforehand, you mentioned a story that kind of got you into applied data science, which at the time was not called data science. When at West Point you were doing math and engineering, you had a conversation with a professor in the fall of 2000 being like, Hey, I'm doing math, but with my engineering stuff, I might want to do science. How did that conversation go and what came afterwards?

Les: It's pretty funny to go back. I'm starting to feel pretty old, especially since now my son is looking at service academies. I'm very old, but yeah, 25 years ago I was declaring a major and decided I wanted to pursue math. At West Point, at that time (and I think it's still the case), you have to get an engineering minor or field of study. I was talking to my academic advisor, and he was trying to convince me to choose an engineering track where I could double-count my electives with the math degree. I chose the one engineering track that had no double-counts, and that was computer science. I remember him telling me, “Wait, Les, why would you want to do that? Computer science, that's like for building webpages and things. You're a math guy.” And I said, “Well, it seems kind of cool, and I'm interested in exploring intersections of math and computer science.”

Lo and behold, at the time they called that ORSA (Operations Research and Statistical Analysis). It's a terrible name for a field it almost reminds me of a killer whale or something. But ORSA became sexy. It became data science. So I feel pretty fortunate. Graduating in '03 from West Point with applied math and computer science, it was pretty fun to have that kind of behind me and part of the fabric of who I would be for the rest of my career.

John: Nice. So after that, you're deployed as a Ranger, you're on the ground. What type of data and I think this is probably a good theme for our podcast today was being used by decision makers impacting on-the-ground actions? Let's start with what type of information, what type of data were you getting fed, and what was frustrating about that? What opportunities did you see?

Les: Keep in mind, my first deployment was in '05 to Afghanistan. The deployment where I really started to think a lot about data was in Baghdad in the summer of 2006. I was probably one of the only Ranger platoon leaders with a data science background back then a weird combo. The IED fight was starting to peak. I was in a task force doing high ops tempo. We probably went out on 75 or 80 missions over the course of 90 days. These weren’t just presence patrols. These were literally kill or capture missions going after high-value targets.

What was frustrating for me was, with all the assets I could see, all the collection, all the special programs, when we went out, I’d go with my platoon supporting some of our nation’s greatest assets, tier-one forces. What was our hit rate on those ops? Maybe five or six times out of those 80 missions over 90 days. That’s when we got the right person and it was a success. And I thought, how is it so bad?

It wasn't that we weren't effective when the target was there we always got them. It was more, what is the fundamental breakdown in terms of the data being collected, how it's being analyzed, and how decisions are being made? Back then, it was very qualitative. Analysts wrote reports, decision makers and planners read those reports, synthesized them usually into PowerPoints. Some technical means were being used, but the actual go/no-go decision was made qualitatively based on theories and priorities. If you do this mission, you can't do that mission. So the choice was, what are we going to get tonight? Who are we going to wrap up? It was very frustrating. I thought, we've got to be able to do better.

I imagine that happens in so many Fortune 500 companies. It happens in government still.

John: So after that experience, you're kind of in more of a decision-making role, using data to drive insights. What were you able to work on in an applied sense, and how did that shift how those guys on the ground were working?

Les: I decided to get out. I had a lot of life stuff going on too my son was a newborn, I was a new husband. That was part of it, but I also wanted to do something that could impact the warfighter through more quantitative approaches. I had a really cool opportunity for about a year at Johns Hopkins Applied Physics Lab before I got scooped up by the CIA. I was supporting a team focused on the counter-IED fight.

The original mission of the Joint IED Defeat Organization was basically: 1) defeat the device (electronic countermeasures), 2) protect the warfighter (armor, jammers), and 3) defeat the network. I got involved in the third pillar defeating the network. That approach satisfies the first two goals. If you can stop bombs from being made or stop the people who place them, you're protecting the warfighter and defeating the device.

Understanding that network meant understanding a complex human system. There’s a supply chain, bomb makers, facilitators, people placing IEDs. It's one of the most complex systems imaginable, but not too dissimilar from the way corporations function minus the whole trying-to-kill-US-service-members thing.

The goal was to use unique data sets to build models and make more quantitative decisions that warfighters could operationalize. We really wanted to unravel the network.

John: And there were some pretty amazing results from that program, right? After it was implemented?

Les: Yeah. Again, I'm not taking credit. I ran a small team, a small program. But there were so many incredible people working on this. Crime pattern analysis, multidimensional scaling, geospatial predictive analytics. I built products like density mapping and templates predicting where caches might be. Really cool stuff.

This data-centric approach started getting serious around 2008. By 2009, the IED fight in Iraq peaked. If you look at the slope of IEDs and their effectiveness from 2009 forward, it’s a decay curve. I’m not saying it was all us, but the "Moneyball" approach certainly contributed.

One of my favorite examples: we were predicting when and where IEDs would go off. We peaked at about 30% accuracy. If you tell a commander, "Tomorrow, between 2 and 4 p.m., on this street, there’s likely going to be an IED," that changes behavior especially after you’re right a few times. It starts to drive averting behavior. But the more commanders believe you're real, the more their actions change, and then your predictions become less accurate.

I briefed one colonel who said, "30%? I could do better flipping a coin." And I said, "You're right, sir. But these are actionable predictions. I could give you 100% prediction right now tomorrow in Baghdad, there will be an IED. But that doesn’t help you take action."

John: That’s the challenge we see all the time. The granularity of the insight determines whether it drives behavior. "There’s an IED in Baghdad" isn’t going to change much. But narrowing it down to an intersection, time window that’s powerful.

Now with investing, especially at the early stage, it's like that. We’ve got 100 companies in the portfolio, one is going to be a winner. How do you see venture capital and private equity grappling with the need for granular data in decision-making?

Les: It’s similar to what we just talked about. The earlier the company, the more qualitative the exercise. You're betting on the person, the team, maybe your relationship with them, and qualitative knowledge of the space. It’s hard to quantify early.

The key now is leveraging AI in the investment process. I try to ask myself: which parts of my thesis can I apply AI to? Does the data exist? Can I get it? Once I know that, I can ask questions or engineer prompts effectively to get answers or see things I’m not seeing.

For example, TAM (total addressable market). I know a founder working on characterizing government contract TAMs. That data’s public, so scrape it, clean it, structure it. But then querying it is another challenge. Even with structured data, things move fast. A TAM today is gone tomorrow. Categories change so quickly.

It’s a time to be slow and methodical in building investment theses. The advantage of that is you have time to dig into data and understand the category.

John: One of our favorite use cases at Epistemic right now is tracking those rapid changes in pharma NPV of a drug in-market or near-launch. Market access changes, public policy, sentiment, pricing all impact what pharma companies and investors care about: asset value. Totally changes how data-driven decisions can be made.

And even 30% accuracy, in the two examples you gave, is so much better than a 5 out of 80 success rate. That’s like early BDR numbers to SQL numbers, and now you're converting.

Les: One of my favorite principles in behavioral data modeling is this concept of multiple layers. First is the temporal layer time has natural structure. Then the geospatial layer location and movement in space. And the behavioral layer what people tend to do.

But behaviors are best understood not by what people will do, but by what they won’t do constraints. Constraints govern behavior. Like in the morning: you don’t shave after you shower or drink orange juice after brushing your teeth. If you were trying to guess someone’s routine, the constraints would guide you more than preferences. That applies to so many predictive models.

John: Love it. We’ve probably got time for a few more questions. One we always ask: what got you into math or using data and modeling in the first place?

Les: It's funny. It was the Nintendo Entertainment System. Super Mario Brothers, Legend of Zelda my all-time favorite game. I loved that tech. I sold my NES at a garage sale to buy a Packard Bell 286 computer. My favorite games were simulations football, baseball. I loved trading players, printing stats. My dad would yell at me for wasting ribbons on the dot matrix printer.

Even at West Point, one of my thesis projects was simulating the 1990 Pittsburgh Pirates should they have beaten the Braves and gone to the World Series? So yeah, I got into it through games and simulation.

And people. We shared a very special teacher Mrs. Mullen at Cathedral Prep. She taught AP math junior and senior year. That was probably the biggest inspiration and biggest leg up. When you pursue things you enjoy, you get good at them. I remember visiting West Point and getting a pop quiz in a calc class. Cadets were failing. I got a 100. That wasn't because of me. That was because of Mrs. Mullen.

John: She would absolutely expect that of you too, child of grace.

Les: Yes she would. Absolutely.

John: Final question. Models help make decisions. If you could go back to any point in your life or history, rewind the clock using the flux capacitor, and make a different decision using data what would that “what the flux” moment be?

Les: I’m going to go back to my decision to go to West Point. I made that decision based on incomplete data and assumptions. I went because it was the cheapest pre-med option. I wanted to be an orthopedic surgeon. After freshman year, I realized I didn’t want that. I had other great options Duke, UPenn, Notre Dame. It would be fascinating to simulate my life had I chosen a different path. I don’t regret it, but I’m curious who I’d be now.

John: So good to have you on. As your kids start their own journeys, what advice are you giving them about the future of data science and decision-making?

Les: Learn how to use AI safely and effectively. Use it to discover what you're passionate about, what you're good at, and where you have an unfair advantage. If you learn prompt engineering and how to use AI to teach yourself, you can do anything. We're still in the punch-card era of AI. Get ahead of it now. I wouldn't tell people to major in computer science or liberal arts. Just be curious. Learn the fundamentals reading, writing, arithmetic and use AI to unlock your superpower.

John: Awesome, Les. Thank you so much. This was a great episode.

Les: Thanks, John. It was a pleasure.