If you were looking to build the premier sport analytics platform for your university athletics department, you’d probably find no more perfect — or interesting — a candidate than Kirk Goldsberry.
Equal parts geography nerd and sports nerd, he is one of the lucky ones who managed to combine divergent passions into a successful and illustrious career: geography professor at Michigan State and Harvard, where his work applying mapping techniques to NBA shooting led to a writing gig at ESPN’s Grantland and FiveThirtyEight sites, which then landed him roles as VP for strategic research with the San Antonio Spurs and Team USA.
In 2018, Goldsberry returned to academia, to become a professor at the University of Texas and executive director of the school’s Business of Sports Institute. He also brought his analytical expertise to a few of the school’s nationally ranked teams, and, seeing the success they had, UT has decided to launch Bevo Lab. Built using Google Cloud services and Gemini AI, Bevo Lab aims to become the top analytics platform not only for college sports but all sports, period.
We recently sat down with Goldsberry to understand what that takes, what’s different between working in the pros and college, and how AI might bring more purity to athletic endeavors.
Pretty much any sports fan, and many Americans who aren’t, knows Bevo the longhorn steer. What is Bevo Lab, and will it become just as renowned?
Goldsberry: That’s the idea. The thesis is that the best analytics platform in sports shouldn’t be at a pro team — it should be at a university. We have 50,000 undergrads and thousands of PhDs who are experts in these fields, and one of the best computer science programs in the country. We have the tech industry on our front door here in Austin, I can literally see Google’s offices out my window.
And we have the best athletics department in all of college sports — that’s not a boast, we’ve won the Learfield Director’s Cup, which goes to the best overall athletic department in the country, five out of the last six years.
So with Bevo Lab, we want to build best practices for our teams and our student athletes that reflect how data and software and AI have changed professional sports, like the NBA, Major League Baseball, or Formula 1. We’re creating this giant sandbox where our students can collaborate and build AI models and resources and apps for our sports programs.
We’re also locking arms with industry leaders so that we know where the puck, or the ball — this is Texas — is going. We want both our athletes and our STEM students to benefit from the expertise of Google and our other partners.
We’ll get into the benefits for athletes, but what do you envision for those students working in the lab?
It’s a lot of opportunities for hands-on experience, jobs of the future, the stuff kids are clamoring for. It’s a job, like the college paper, we get dozens of applicants, they could be studying computer science, statistics, business, biology. Those we accept will work 10 to 20 hours a week in this cross-disciplinary environment.
Doing what?
We’ll start with building the biggest data lake and the best backend in college sports, those are the foundation for building the best intelligence platform. Right now, our data is in silos and disconnected, like a lot of universities, which is suboptimal.
Once we’ve addressed that, our students will leverage our data lake to build APIs to connect all of these applications and data sources we have across teams and the department, that’ll pull all the relevant athlete data into one centralized database.
This is where AI can do its thing. We'll build a bunch of models that make predictions on strategy, scouting, game planning, the transfer portal, recruiting. Of course, football is the star of the show at UT, so we’ll start there, and then expand to other sports, we’ll have dozens and dozens of models for all the teams.
We’ll also be building apps and front-facing intelligence services, those’ll be personalized. When an athlete uses an app, their experience should reflect their interests and their persona. If a coach logs on, they should have a completely different experience.
With the top athletics department in the country, leaders at Texas see no reason they shouldn’t also have the top analytics program in all of sports.
It sounds like enterprise tech, you’ve got your horizontal applications that are similar across the organization. What about your verticals, are there things for specific roles or teams?
The verticals are each sport, yeah, there’s some differentiation between those. And then there's applications that run across sports with some adjustments, like game planning, recruiting, the transfer portal, and general team management stuff. There's lots of factors, and AI can help us adjust those from sport to sport.
Medical is its own thing. We have a lot of opportunities for personalized fitness, performance, and recovery, but that data’s also very, very sensitive, and that's where we need Google Cloud to help us treat these data sets with the reverence and respect they demand.
Then there’s the Moneyball-type analytics stuff, which will probably be different between sports, like who should bat third for the softball team, and who should bat fourth, and who should pitch today.
Football is the star of the show at UT, so we’ll start there, and then expand to other sports, we’ll have dozens and dozens of models for all the teams.
What about cross-pollination between sports? You’re starting with football, maybe something feeds into the volleyball team, but then maybe volleyball finds some insights that feed back to the football team?
Kirk Goldsberry: In my class, we just wrapped up our baseball unit today, and I told them that as a basketball analyst, I constantly stole baseball’s ideas. The idea of disciplines informing other disciplines is as old as a university itself. So our great volleyball minds can certainly bring ideas to basketball and then to track and field, etc. It's this big intelligence tent where we're borrowing from each other. We’re developing best practices so that university athletics programs can move away from silos, which are the opposite of a virtuous loop.
Speaking of basketball, I understand you started working on projects like this not long after arriving on campus?
I was fortunate to work with Coach Popovich and R.C. Buford, the Spurs CEO, who set a great standard of excellence in sports operations. We were using AI back in the 2010s for a variety of purposes. When I came back to higher ed, to a great athletics department in one of the country’s hottest tech hubs, I was like, “Why don't we have something like this?”
So we started it, working on pilot projects the past four or five years, starting with basketball, and Bevo Lab became the logical extension of that. I worked with some students to build personnel models, we called it Transfer AI, that predict the performance of every college basketball player across all the essential skills — shooting, defending, playmaking, rim protection, and rebounding. And then we mapped that onto data from across the conference.
So it’s not just how will a player from Yale or Pitt perform at UT, but how will this player be as a rebounder against our SEC opponents next year, or, how great of a scorer would this freshman be in the SEC environment?
Another early example from our proto-Bevo work that we launched this year is our Football Scouting Intelligence. It automates what used to be the manual compiling of opponent data and video reports that Coach Sarkisian’s staff had to do each week. It’s saving them 15 hours a week, which they can instead spend on more valuable things like player development, game planning, and strategy.
As one of the premier tech hubs in the world, Austin is a natural place to incubate innovations in sports science and sports medicine with the help of AI.
What are some of the differences between what you can do with AI in college versus the pros?
I am jealous of my friends at the Spurs for one thing — they only have 18 athletes to worry about all year long, in one sport, with a huge budget. I have 500-plus athletes, and 21 different teams. So one big difference is just the scale.
At the end of the day, we have resources these pro teams would kill for. The best pharmaceutical or medical labs are attached to universities, and the best sports labs should be, too. It’s the culture of learning and innovation and trying new things that is university life. We can see more about what’s happening in robotics than the Spurs ever could, and maybe that can apply that somehow. So they have some perks, but we have more.
What’s something you've seen in the gen AI era that you wish you had when you were with the Spurs?
The way these systems digest boatloads of structured or unstructured data is unreal. I would have loved to plug in the entire shot log from the NBA last season to identify the best mid-range shooters. Or built a model that estimates who makes the hardest shots at the highest rates. Or figure out who are the best paint scorers.
Then there’s the natural language aspect. Taking reliable structured or unstructured data and querying it with natural language removes esoteric statistical software from the process and lets everybody just ask questions and get insights and make graphs. It really lowers the barrier to entry for people to get intelligence.
We have resources these pro teams would kill for. The best pharmaceutical or medical labs are attached to universities, and the best sports labs should be, too.
Maybe it’s already changing, but what’d it take before to get these usually very instinctual coaches and players to trust the technology?
One trick I have is to use data to empirically prove something a coach already believes. A great example is that Coach Popovich believed in corner threes even before he knew the efficiency down to the hundredth of a decimal place. He was just like, "This seems
like a good shot. It spaces out our offense. It makes defense hard to play." And I was able to say, "That shot is also worth 40% more than any other jump shot on the floor."
At the end of the day, sports analytics is a communications — we called it Transfer AI — problem and you have to be a good communicator to thrive.
We hear a lot about the purity of sports, and especially college sports. Do you think AI can unlock more of that pure athletic expression?
Sports are the ultimate human thing, right? Whoever wins wins. In the face of AI everything, you can't AI slop Victor Wembanyama. People want to see him play. But if we can sharpen the greatness, it just elevates the whole endeavor. This stuff’s going to help athletes and coaches become greater versions of themselves.
We've seen how analytics has already changed basketball and baseball at very high levels. Now, those changes are coming to football, volleyball, tennis, swimming, and so on.
I told my students how, for 150 years, before analytics took over baseball, we thought swinging flat was the right way. But with 3D modeling, we started to analyze outcomes, and now, everybody swings upwards in a way that their little league coach would have been, like, “What are you doing?!” Then you apply that to shooting in basketball, blocking in football, swinging a tennis racket, and so on.
AI and other cloud technologies are going to elevate best practices, which is going to elevate the game play, elevate the physical achievements, elevate that purity of sports — and elevate our enjoyment watching it all unfold.
Just as Bevo has been an icon for UT for the past century, the school hopes Bevo Lab will become a symbol of the school’s vision and success for the next hundred years.
There’s also a longstanding focus on student athletes, and then more recently the role of AI in education. Do you see a way that AI can help athletes also be better students?
I think about saving people time and reducing some of the tedious tasks that burden everybody, especially student athletes, but also their coaches and the executives. There's lots of opportunities for building little processes that save people time and give it back to them to study. The student athletes will teach us things about time management, and we'll learn how they want to use AI. Somebody will bring us a brilliant idea and we'll scale it to the whole athletic department.
This conversation has been all about AI pushing boundaries and breaking barriers — so why not time travel. Is there one game, season, event, catch, throw, something in Longhorns history that you wish you could have used Bevo Lab for?
I wish our women's basketball team would shoot more threes. We've made the Final Four the last two years, and come up just short. If we shot a few more threes, it’d be a couple more points a game here, a couple more points a half there. Could we have beaten UCLA if we had been more analytically aware in shot selection? I know the coach well, so I can say this. And he’d tell me why I’m wrong, too.
I think those small marginal gains are what we can expect with a project like Bevo Lab, and those small marginal gains are often the difference between a win and a loss.
Opening image created with the Gemini agent, using the prompt: A minimalist editorial vector illustration of a Texas Longhorn steer (Bevo), side profile body with his head turned forward looking directly at the viewer, showing iconic wide horizontal sweeping horns. Flat graphic art style, clean vector lines, high contrast. The steer's coat uses University of Texas burnt orange and crisp off-white. In place of organic hide spots, his markings are subtly composed of two things: 1) clean, wordless data visualizations: stylized scatter plots, geometric trend lines, radar charts, coordinate grids, and bar distributions; and 2) sports play diagrams for football and basketball with Xs and Os and arrows, and you can add some arrows around the body, all seamlessly mapped onto his body contours. Sophisticated sports analytics concept. Solid, clean warm off-white background, elegant graphic design, modern tech-editorial aesthetic, uncluttered composition, completely free of text, letters, or numbers. // Add a flat, vector-style soccer ball with white outlines to the orange spot near the steer's head.