ChatGPT recently dropped the ball on choosing the winner of this year’s March Madness tournament: “I can’t predict the exact winner. March Madness is highly unpredictable,” the AI chatbot replied unhelpfully. Anthropic’s Claude dribbled out a little more information, noting that the favorites to win the NCAA men’s basketball showdown include the Duke University Blue Devils, the University of Michigan Wolverines, and the Arizona State University Wildcats— but still declined to pick a winner.
The AI-based March Madness predictor developed by recent Bryant graduate Jack Sweeney ’25, however, didn’t hesitate to take its shot, picking the University of Houston Cougars to hoist the NCAA National Championship Trophy this year.
If the prognostication proves out, it will be the third consecutive year that the model has identified teams that ended up finishing among the Final Four — the last teams standing in the annual matchup of the best college basketball programs in the United States.
In 2024, Sweeney’s model correctly predicted that the University of Connecticut Huskies would reign tournament champion, while listing other contenders such as University of Houston Cougars, the Purdue University Boilermakers, and Creighton University Bluejays. In 2025, it tabbed the Auburn University Tigers as the tournament champion while highlighting the Florida Gators, Houston Cougars, Alabama Crimson Tide, and Texas Tech University’s Red Raiders as top contenders; Auburn lost in the semifinals to eventual national champion Florida, 79-73.
For the 2026 tournament, the model identified UConn, Purdue, Florida, and Michigan as the other top title contenders.
Sweeney, who graduated with a degree in Data Science and works as a Financial Customer Associate at Fidelity Investments, developed the program while still a student at Bryant, where he also served as a manager for the men’s basketball team. Under the mentorship of Suhong Li, Ph.D., chair of the university’s Information Systems and Analytics department, Sweeney began developing the model during his junior year with an eye on making it his Honors Thesis.
“I didn’t want to be stuck with a project that I didn't enjoy,” explains Sweeney. “I knew that I wanted to do something with sports, and there was a lot of stuff on social media at the time about college basketball statistics, and that's how I came up with the question of whether statistical modeling could predict March Madness.”
Not surprisingly, the AI model wasn’t Sweeney’s first exposure to basketball “bracketology.” He was just 10 years old when his father invited Jack to make some picks in his office’s March Madness pool.
“I started doing it, and I continued to do it every single year,” he says.
Under Li’s guidance, Sweeney used analytical tools like Jupyter Notebook and Python, along with Microsoft Excel, to build his prediction engine through data mining and machine learning.
“Instead of doing the standard points per game, rebounds per game — stats everybody else can see — I wanted to dive deeper into advanced stats to see if I could find a true pattern and predict the winner of the tournament,” he recalls.
“Instead of doing the standard points per game, rebounds per game — stats everybody else can see — I wanted to dive deeper into advanced stats to see if I could find a true pattern and predict the winner of the tournament."
Eventually, he was able to gather more than a decade’s worth of data on 300 NCAA Division I teams, including advanced statistics on offensive and defensive efficiency as well as more obscure measures like continuity and experience that determine “how old your players are and how long they've been in the coach’s and team’s system,” Sweeney explains.
“Things like points per game, rebounds per game, assists per game are only the tip of the iceberg,” he continues. “It doesn't really show how well a team can perform. Really, it's all about whether a team is efficient or not, so I felt like it was important for me to get down to the efficiency metrics. It was important for me to capture the overall team.”
RELATED STORY: In Power Five conferences, women head coaches are associated with higher likelihood of NCAA Tournament appearances
Each season, Sweeney gathered new data right up until the March Madness tournament began. “I got a full overview of how the team performed during the regular season during non-conference and conference games, and into the conference championships. From that, I gained an overall view of the team's actual performance,” he says.
Key variables in the model included physical attributes of players as well as performance measures, including the height of teams’ small forwards, shooting guards, and centers. Defensive rebounding was another, along with a stat called 'wins above bubble' — “which basically measures your team's ability compared to bubble teams, which are teams that just make or miss the tournament,” Sweeney details.
One study limitation was the inability to capture the impact of regional seeding, which could be a disadvantage for teams placed in regions farther from home. Another was the inability to capture in-game adjustments by coaches.
“In-game adjustments measure a coach’s ability to adapt to an opponent’s play style and game plan,” according to the research study co-authored by Sweeney and Li and published in Issues in Information Systems in 2025. “In-game adjustments are important in March Madness, as it’s a single-elimination tournament where every game matters.”
Despite the limitations, however, the model’s predictions have so far have been either spot-on or a near miss. Ironically, Sweeney was so busy working on the tool during the past two March Madness tournaments that he failed to use it to guide any brackets or bets of his own.
“I wish I could have put it to more use, but I'm definitely using it this year,” he says.