Johni Broome bulldozed most opponents this past season at Auburn, but in the second round of the NCAA tournament against Creighton, the 6-foot-10 All-American struggled through one of his worst games. Auburn managed to win, but Broome scored only eight points on 4-for-13 shooting, missing all of his 3-point shots and free throws.
When asked about the performance in the postgame press conference, Broome gave the kind of stock answer you’d expect from a fifth-year senior. “This season is about winning,” he said. “I know my teammates got my back, and they know I have theirs. My shots weren’t falling, so I just had to do whatever I can to impact winning, and I let everybody else do the scoring.”
Broome’s answer isn’t very revealing—the clichéd response probably didn't make the game story in the local newspaper. But to Sean Farrell, the response speaks volumes about Broome’s chances at the next level. “If you’re focusing on the now … you’re more likely to make it into the NBA,” Farrell said. “If you’re ruminating on past mistakes you’ve made, you’re less likely to make it.”
By day, Farrell is the senior data scientist for the Queensland Fire Department near Brisbane, Australia. By night, the former astrophysicist makes “Moneyball models” for fun—including the models at the heart of a new research paper that analyzes how the language a player uses in interviews can predict NBA success. Farrell and his coauthors developed machine learning models that could analyze large sets of data, then plugged in nearly 26,000 transcripts from more than 1,500 college players going back decades.
With no other information about who the player was or how good they were in college, the models accurately predicted who would make it on an NBA roster 63 percent of the time. When factoring in context like age, height, stats, and college conference, the models got it right 87 percent of the time. The research also forecast which players were likely to last at least 250 games in the league (with 69 percent accuracy) and start more than 30 percent of those games (with 68 percent accuracy).
These models lend themselves to plenty of valuable use cases: choosing between two players with similar draft grades, finding a hidden gem in the second round, deciding which player to keep after their rookie contract. The results anticipated successful NBA careers for Kawhi Leonard (drafted 15th overall), Draymond Green (35th), and Danny Green (46th) but missed on Jimmy Butler (30th). Broome, ranked 41st in The Ringer’s NBA Draft Guide, rates as better than a coin flip to last at least 250 games in the NBA. Cooper Flagg, the presumptive no. 1 pick, falls in the middle of the pack based on the language analysis but still ranks first when factoring in his other measurables.
The models aren’t perfect. They work only with English, which ignores a crucial segment of international players. And postgame interviews aren’t a controlled setting where each player answers the same questions; reporters dictate what a player responds to, which influences the language they use.
Nevertheless, Farrell, who never followed the NBA before pursuing this research, may have discovered a new wrinkle in the draft scouting process. For instance, players who used words associated with being honest and non-defensive—such as “realize,” “believe,” and “understand”—were more likely to make it. Those who spoke in longer sentences and used more complex language were actually less likely to make it. The data, Farrell found, favored a clear perspective over articulation. So while athletes who take it one game at a time might sound boring at the podium, they may be on to something.
Farrell presented his findings at this year’s MIT Sloan Sports Analytics Conference, where he and his coauthors were finalists for the 2025 Research Paper Competition. Afterward, he told me that reps from the Miami Heat, Los Angeles Lakers, Phoenix Suns, and Cleveland Cavaliers lined up to talk to him. (Two MLB teams joined as well.)
Like most businesses across the country, the NBA has begun wading into the brave new world of artificial intelligence. The league office has had an open-door policy for new technology for years, and team front offices have evolved beyond advanced statistics to incorporate machine learning into player evaluation—including prep work for this week’s NBA draft.
Farrell hopes his research will be part of this new revolution. “To me,” he said, “psychology is like the last frontier in terms of analytics.” But AI figures to infiltrate all aspects of how your favorite NBA team is run—and in some ways, it already has.
The day after Farrell presented his research this past March, Philadelphia 76ers president Daryl Morey made headlines for revealing during a Sloan conference panel that the Sixers consult artificial intelligence—specifically large language models trained on their scouting notes and player tracking data—when evaluating players. “We absolutely use models as a vote in any decision,” Morey told the crowd. “We’ll treat them almost as one scout.”
The only surprising thing about Morey’s comment was that he said it out loud. Team executives keep secrets like spies if they think they have a competitive edge, but the league has been logging player tracking data in NBA arenas with expensive hardware for roughly 15 years. AI merely helps make sense of the collected data.
“They’ll try to use [AI] to potentially see information that’s coming out of their scouting reports that they have documented over the years,” said Dean Oliver, a coauthor on Farrell’s research paper who has also worked in analytics roles for the Washington Wizards, Sacramento Kings, Denver Nuggets, and Seattle SuperSonics. “Are there pieces of information in their scouts’ brains that they can use to make better predictions?”
The NBA has been ahead of other sports leagues on this front for decades. Since 2000, the league has held an annual tech summit during All-Star Weekend where guest speakers talk about innovation in sports. Since 2018, there’s been an invitation-only tech expo held during summer league, which gives tech companies a chance to demo their products in front of teams. And in 2022, the league even started its own tech incubator, NBA Launchpad, to invest directly in research and development that could improve on-court action or the fan experience. Many recent companies chosen by Launchpad use AI, with products ranging from smart basketballs that collect data to in-shoe sensors that track player movement to software that turns MRI scans into 3D imagery.
Some of the bets won’t work out, but these investments are more than just moon shots. After watching a presentation at the first NBA tech expo, the Orlando Magic got exclusive access to an AI platform called AutoStats that provides player tracking data from game broadcasts rather than in-arena sensors. As a result, the Magic saw unique stats and player movement trends on college basketball players that other teams couldn’t access.
“We have found that our predictions have become way more accurate,” said David Bencs, Orlando’s assistant general manager, in regard to the team’s draft scouting.
After the 2024 draft, the Magic switched to another AI platform called SkillCorner that collects similar data across college basketball, the G League, and some international leagues. These tools are revolutionizing the way scouts work. With SkillCorner, a scout can analyze information like shot release time, pass defender location, and how players react to screens. It makes old box scores look like kindergarten math. “I actually think that AI and the data we have can help create better scouts,” Bencs said. “They don’t have to focus on some of the more mundane parts of the job, and they can actually do a lot more studying.”
For example, in the past, if a Magic scout wanted to study how a prospect ran the pick-and-roll in college, they’d have to sift through hours of film to find those plays. Now, with AI, they can filter game footage for pick-and-rolls and get an edited compilation with just one click.
Are there pieces of information in their scouts’ brains that they can use to make better predictions?Dean Oliver
As Mavericks minority partner (and former majority owner) Mark Cuban told me over email, teams can also use that functionality to help current NBA players make in-game adjustments and scout upcoming opponents. “We can use it to create videos directly from game footage for players to review without waiting for the video room. And it can be taught to give feedback.”
While AI threatens to replace engineers, lawyers, and other corporate jobs, nobody I spoke to around the league worried that AI would make scouts obsolete. Oliver argued that there will still be important clues that you can’t collect data on to train AI, like a player’s offseason training habits. Yet teams may still try to save money by outsourcing jobs to AI—like other industries have—especially lower-level positions like video coordinators, for instance. Eliminating those entry-level jobs could make it harder for teams to find and nurture talented basketball minds. There’s an impressive list of former video coordinators who worked their way up the NBA ladder, including NBA champion coaches Erik Spoelstra and Frank Vogel and Thunder executive vice president Sam Presti, who was voted the 2025 Executive of the Year.
The other problem with AI is groupthink. As Oliver put it, “AI has been described as a great interpolator. It is great for that. It can be difficult for thinking outside the box.”
When Morey first became an NBA executive, he was an outsider who brought a unique analytical perspective to personnel decisions. Today, front offices are full of his disciples. They crunch a lot of the same numbers and consider the same tools. There aren’t as many inefficiencies to exploit. And the league is already saddled with the notion that teams play a similar style of basketball, hunting too many 3-pointers and foul shots. It’s possible AI leads to more like-mindedness.
“We are always open-minded and trying out new technologies to get a better advantage, but it’s a zero-sum game,” Bencs said. “I don’t have any illusions that other teams are not using or not starting to use tracking data now for the draft. And I think we’re going to have to find whatever that next edge is.”
In a surreal scene at the NBA All-Star Technology Summit, commissioner Adam Silver shared the stage with some unexpected special guests: physical AI robots named K.I.T., M.I.M.I.C., and B.E.B.E.
Silver introduced his new colleagues with a jokey video featuring the Golden State Warriors, who hosted All-Star Weekend. In the sketch, one robot flings passes to Steph Curry. Another encourages Draymond Green in the weight room. At one point, Warriors coach Steve Kerr yells, “Get the fuck out of here” to robots moving on a practice court. After the video ended, the robots flung T-shirts to onlookers in the crowd.
The presentation was so confusing to some in the room that Fox Sports’ Ric Bucher asked the Warriors whether they actually used the robots. They didn’t; the video was just conceptual. But with AI, it doesn’t take long for something conceptual to become reality.
One robot in particular, B.E.B.E, promises to tailor recovery programs and help players optimize their bodies for peak performance. Whether the droid shaped like a cooler ends up making good on the promise remains to be seen, but Bencs and other team decision-makers are chasing solutions to the same problems.
Farrell’s research paper didn’t cover player health, but he teased it as a future use case at the end of his talk. He explained that it could be possible to connect language cues to athlete injuries, building off existing research from sports psychologists that suggests certain psychological factors and stressors may signal an athlete is more likely to get hurt. (He told me he’d need access to a lot more data to test this.) In addition to sports teams, Farrell says he was approached about his paper by one of the largest insurance companies in the world.
There’s already a rapidly growing field of academic research getting better at forecasting injuries through machine learning and other AI techniques. This offers teams more visibility into, say, how height influences back and ligament problems or abnormal running styles contribute to knee injuries. “AI can analyze and compare how players run and jump across a period of game minutes looking for changes that may not be visible to the human eye and not detectable by the player,” Cuban said.
AI will uncover things we haven’t yet considered.Mark Cuban
Some of these changes in movement aren’t even detectable by team doctors. That’s opened the door for a company like Springbok Analytics, which creates 3D renderings of medical scans that give teams a deeper understanding of a player’s muscle quality. The AI-powered tech can spot physical imbalances, like if one hamstring has to work harder than the other, so trainers can address them before they turn into season-ending injuries. Springbok’s client list includes the Utah Jazz, Chicago Bulls, Memphis Grizzlies, and Detroit Pistons.
As helpful as this technology can be in the medical space, it also raises ethical concerns. Can players opt out of this tracking, as they can with any wearable device? What if a team uses these models as a justification to bench a player before they can earn contract incentives? What if a general manager trades or cuts a player based on the likelihood of an injury occurring before it actually occurs? Or what if a front office makes a bad decision because of an AI hallucination? The NBA will need new guidelines to ensure there’s enough transparency to protect everyone involved. The league will also have to reconcile its commitment to player safety with its desire to showcase the best basketball every night.
“Teams have always been making personnel decisions based on new types of technology, obviously AI being one of the latest ones,” said a spokesperson for the National Basketball Players Association. “For us, it’s more about the protection of that data and player safety. That’s our biggest priority when teams are using these tools.”
Based on what exists today, we may not be far off from teams trying to predict the injuries incoming draft prospects could experience at the NBA level. Or exploring other applications of AI and machine learning that have yet to be revealed. The future is now, and the NBA is jumping headfirst into the new era.
“It’s crazy what is happening in the space,” Cuban told me. “AI will uncover things we haven’t yet considered.”
