The 2025 NCAA Men’s Basketball Tournament concluded with the Florida Gators securing their third national championship, defeating the Houston Cougars in the title game. In the lead-up to March Madness, the NCAA Selection Committee was tasked with seeding 68 teams, a process informed by seven different ranking systems. These included NCAA Evaluation Tool (NET), KenPom, Torvik T-Rank, ESPN’s Basketball Power Index (BPI), Wins Above Bubble (WAB), Strength of Record (SOR), and KPI. These systems incorporate varying methodologies, including efficiency-based models, résumé-based assessments, and hybrid approaches combining performance and opponent strength.
To evaluate the predictive validity of these metrics, I conducted an experiment during the 2025 tournament. Using each ranking system independently, I constructed a complete bracket based solely on the relative rankings as of Selection Sunday. No adjustments were made once the tournament began, and no subjective inputs were included. The objective was to assess which system most accurately predicted the actual outcomes of the NCAA Tournament.
A Chalky March Madness
This year’s tournament produced very few upsets relative to historical norms. All four No. 1 seeds advanced to the Final Four, and three of the four No. 2 seeds reached the Elite Eight.
In all the tournament produced only eleven upsets, defined here as any lower-seeded team defeating a higher-seeded opponent. Even among those results, many were minor deviations from expectation. Two were 9-over-8 matchups, and two others were 10-over-7. According to historical data, a No. 9 seed wins about 49% of the time against No. 8 seeds and No. 10 seeds win about 39% of the time against No. 7 seeds.
The four South region upsets included: No. 9 Creighton over No. 8 Louisville, No. 10 New Mexico over No. 7 Marquette, No. 5 Michigan over No. 4 Texas A&M, and No. 6 Ole Miss over No. 3 Iowa State.
The West region accounted for the largest share: No. 12 Colorado State over No. 5 Memphis, No. 11 Drake over No. 6 Missouri, No. 10 Arkansas over No. 7 Kansas, and No. 10 Arkansas over No. 2 St. John’s.
East region upsets included: No. 9 Baylor over No. 8 Mississippi State and No. 6 BYU over No. 3 Wisconsin.
The lone upset in the Midwest region included: No. 12 McNeese over No. 5 Clemson.
This relatively low rate of disruption suggests a high level of accuracy in the Selection Committee’s seeding decisions. Given that the Committee’s process is informed in large part by advanced ranking systems, the tournament’s chalk-heavy nature may also reflect the strength and consistency of the underlying metrics used to evaluate team quality.
Measuring The Accuracy Of Ranking Metrics
To evaluate the predictive accuracy of the ranking systems used by the NCAA Selection Committee, I submitted bracket entries into the 2025 ESPN Tournament Challenge with picks made purely based on each of the seven rankings used by the NCAA Selection Committee in order to determine which ranking system proved most accurate this season.
The selection process followed a systematic approach. For each ranking system, the team ranked No. 1 was selected as the national champion. The remaining picks were made sequentially: the team ranked No. 2 was advanced as far as possible without facing a higher-ranked opponent, and so on, until a complete bracket was filled. This process ensured a consistent and objective methodology across all seven systems.
To assess performance, each ranking system was evaluated using two scoring frameworks:
- Tournament Challenge Score: This metric follows ESPN’s official scoring format, in which each round increases in point value (10, 20, 40, 80, 160, and 320 points per correct pick, respectively). This weighted model rewards accuracy in later rounds, where correct picks are more difficult and impactful.
- Unweighted Accuracy: This metric measures the raw percentage of correct picks, calculated as the number of games correctly predicted divided by the total number of games in the tournament (63). It provides a more neutral assessment by treating all games equally, regardless of round.
Together, these two metrics offer complementary views of performance and capture both bracket efficiency in a competitive context and underlying predictive accuracy.
Results Based On 2025 March Madness
None of the brackets constructed using the seven ranking systems correctly predicted the Florida Gators as national champions. However, five of the seven systems accurately projected all four Final Four teams. Only Wins Above Bubble (WAB) and KPI omitted Duke in favor of Alabama, a divergence that ultimately reduced their overall predictive performance.
The Tournament Challenge results present an interesting pattern. Only 290 out of a possible 1,920 points separated the top performer, KenPom, from the bottom performer, KPI. However, these brackets landed in very different percentiles. The percentile is an indicator of how the bracket performed relative to other brackets submitted on ESPN. A KenPom bracket was better than 97.5% of all other brackets while a KPI bracket was only better than 68.6% of all other brackets. This gap underscores how even modest differences in bracket accuracy can translate into substantial variance in competitive standing.
Unweighted accuracy showed less variability across systems. KenPom and Torvik T-Rank led with 51 correct picks out of 63 games (81.0%), while KPI trailed slightly with 46 correct picks (73.0%). The narrow five-game spread between the top and bottom performers illustrates the relative consistency in baseline predictive ability. However, the compounded value of correct predictions in later rounds, particularly the Elite Eight, Final Four, and championship, amplified the separation in weighted performance.
Takeaways For March Madness 2026
This analysis demonstrates that while all seven NCAA-endorsed ranking systems offer a reasonably accurate view of team quality, not all are equally effective at predicting tournament outcomes. KenPom and Torvik stood out, combining strong overall accuracy with high-impact picks in later rounds. In contrast, systems like KPI and WAB lagged behind, missing key matchups and failing to match the predictive performance of their peers.
In a year with relatively few upsets, the best models closely mirrored the actual trajectory of March Madness. This performance reinforces their value not just to the Selection Committee, but to analysts, fans, and bettors seeking a competitive edge. When the margin between a good bracket and a great one is only a few games, the choice of ranking system can make a meaningful difference.