From Sabermetrics to Moneyball and now optical player tracking, few domains have harnessed data in decision-making as deeply as sports.
“You hear coaches and GMs asked how they use data and artificial intelligence all the time,” says Ryan Resch, former assistant general manager and vice president of basketball operations for the Phoenix Suns. During Resch’s nine years with the team, he led data science, salary strategy, scouting, and other key functions of managing a basketball franchise.
Harvard Business School Professor Boris Groysberg first met him while researching a 2024 case about Shore Capital Partners, where Resch works on data and AI enablement across the firm. They recently discussed how technology—particularly agentic AI—might change professional basketball. Whether it’s a team or a business, Resch argues that competitive advantage ultimately comes from proprietary data and the ability to act on it.
During the past two years, Groysberg, the Richard P. Chapman Professor of Business Administration, has been exploring how firms can manage their talent as AI fundamentally reshapes work. Here’s what we learned from their discussion.
Machine learning data is omnipresent in basketball
Resch: Statistics in basketball have been tracked for decades, arguably since [James] Naismith invented the game. In the early 2010s, a company called SportVU came around. I believe they started as an Israeli missile technology company. They installed cameras into the rafters of arenas, and those cameras produced the first optical tracking data within the NBA.
This was a huge revolution. If you could build SQL databases, you were able to write a lot of code to answer basic questions, such as: What is the shot quality based on the player’s location, how close their defender is, and how quickly they’re running at them?
A company called Second Spectrum came online a few years later. They took all the optical tracking data, applied their machine learning algorithm, and created a summarized dataset of these actions. We could measure points per possession or expected points per possession. We could say, “OK, when my point guard and my center are together in an action, they produce 1 point per possession. But if I change out the center to this guy, I can get 1.2 points per possession.”
Early adopter advantage can fade quickly
Resch: For the first time, you were able to effectively have data-driven decision-making at scale. The handful of teams that adopted Second Spectrum early on were at a significant advantage. Every year, adoption grew across the league. And then Second Spectrum partnered with ESPN, so they were saturating the league with the same information.
Groysberg: So, the first-mover advantage was there early on, and then basically when everybody starts copying and using the same vendor, it became less?
Resch: Yes. Everybody was using the same data. You can also make a very strong argument that access to this data led to a homogenized style of play in the same sense that your Facebook feed, for example, becomes the same thing. Once the algorithm figures out what you want, once you figure out how to win the game of basketball, and everyone else has that same information, you’re going to end up with very similar overlapping styles.
Go beyond dashboards to gain advantage
Resch: When Second Spectrum came out with a dashboard, we now had to do less work. It cost less to get this information because I didn't have to code it. I could just go onto the dashboard and click through the filters. And they had an unbelievable filtering system.
We got away from optical tracking data, so we got away from the atomic level of measuring things in basketball. The teams that stopped digging into that optical tracking data were not able to answer the deep strategic questions that would lead to differentiation in this more homogenized style of play.
Groysberg: And just to be clear, the data was still there. Teams that wanted to get more details, they could have used it.
Resch: They could have, yes. It would have been a strategic differentiator to lean into the optical tracking data even deeper while everyone was turning to dashboards in Second Spectrum.
Why did we do this? A handful of reasons:
If you were an organization that was not data-forward, you weren't going to invest the money to get the people who had the technical coding chops to write a thousand lines of code to produce this information.
When you are playing 82 games in a year, you simply don't have the time to write all of that code. And the dashboard style does help because it's effectively available for you right away.
And people just forgot about it. We just kind of stopped talking about what optical tracking data can do.
Dashboards show you averages, so you can see nuance if you know what you’re looking for.
Protect your proprietary data
Groysberg: It seems like maybe there’s a lesson there. The great example for me would be Domino’s, right? Some data is completely their own and they will never outsource it to the market. Patrick Doyle, the CEO of Domino’s, doesn't believe you can outsource proprietary data to third parties because eventually they'll sell it to others.
Your particular case is very prescient. You have a provider that becomes best practice, right? And then it becomes harder to generate insights that can be easily generated by a dashboard.
Resch: And that's exactly the problem that teams face. If data is a moat, it comes back to how strong is that moat? If it's going to be third-party data, it's extremely weak because as soon as everybody buys it, then there's really no advantage outside of how you apply it. If it's internal proprietary data, it's going to be very strong—assuming that it's high volume, high quality.
Agentic AI could make proprietary data easier to analyze
Resch: What's so interesting now, and really in the last few months, is where the AI stuff comes in with the advent of these agentic coding tools. Between Claude Code and Codex from OpenAI, we can now return to the world where this optical tracking data is processed and analyzed at speed and at scale, and start generating insights just at a rate that we were not able to do before.
In theory, if you apply it correctly, you could start gleaning some of these strategic advantages that other teams don't have.
Groysberg: Can you imagine this scenario when it's going to be happening in real time in the game?
Resch: Oh, absolutely. There's governance concerns around that, just around league rules and how you can use technology during games. But capturing that data, ingesting it, and then running the model over it, that is no longer a technical problem. Before, you would've had to have somebody writing that code. Now, you would just have an agent assess that data as it comes in and spit out the answer.
AI enables faster experimentation, not just more analysis
Groysberg: You can probably now build a “digital twin” of a team with AI. You would build an agent based on a player and replicate what they do, so you can experiment without physically experimenting, right?
Resch: It's taking the sandbox idea to a completely different level.
If you think about the basketball version of that, it’s “OK, if I have a team of 15 guys and we're going to move three of them out and add three different guys in.” If we have agents that represent those players’ play styles, you can control for minutes and the allocation of the other resources.
Turning insights into action is critical
Resch: We're in an era where AI and data are in the hype cycle. It's extremely sexy and people want to apply it and they want to use it so that they can say that they're using it. But there really does have to be an understanding of applying action with the data.
Groysberg: So that's like having data versus having data insights, right?
Resch: Yes. What AI is doing now is it's compressing the problem from question to answer with data. It’s saying, “OK, we want to answer this question. We have all this data.”
Groysberg: But you still have to ask a good question and be able to take action.
Resch: Correct. Here's an example: You're in the NBA five years ago and you're trying to interview a head coach to hire. They're going to come in and say, “We're going to shoot a lot of threes. We want to be a high three-point shooting team.” Everyone's saying that at this point, right? There's nothing novel about that. What matters is the next question, “And how are you going to do that in a way that actually wins?”
That's where we're reaching with data and AI adoption. It's not going to be novel to just have it. As it is, you need to actually generate the insight and then take action on top of it.
So, when you're designing your AI strategy and your data strategy, everything has to be pointing toward that action bucket at the end of the day. Meaning, you literally need to structure your databases so that your agents can crawl over it. And then your agents can produce actionable insights rather than just fun facts.
Groysberg: So being data-driven or “AI-native” is really an operating model, not a label.
Resch: Right. You’re not AI-native because you use AI. You’re AI-native when your data, your questions, and your decisions are all set up so AI is doing something useful in each link of that chain.
Groysberg: That feels especially relevant right now, with the NBA Draft just behind us and free agency underway.
Resch: Exactly. This is when all the theory turns into action. The draft and free agency are not just events. They show you what a team learned about itself last season and what it believes it should be next.
You can see it in this year’s Finals, too. It was a great matchup.
The Knicks and Spurs were two very different team-building stories. The Knicks took a focused approach to building that roster: they were deliberate with their trades, the players they targeted, and the type of team they wanted around Jalen Brunson.
San Antonio followed a different path. That’s more of a homegrown, asset-management story: draft well, develop well, manage your picks and contracts, and build the kind of organization that can win without the advantages of a bigger market.
AI and data can help both of those situations, but only if you already know what you’re trying to build. AI can now help you process more information and test more scenarios. It can be as tailored to your specific situation as you’re willing to define. It can help you compare players, lineups, contracts, and trade-offs faster than before. But it cannot answer the first question for you: What kind of team are we trying to become?
Illustration created by Ariana Cohen-Halberstam with assets from AdobeStock.
