With 147 quintillion possible outcomes, picking the winners of all 67 games in the annual NCAA basketball tournament is nearly impossible. But Microsoft is attempting to model an infallible bracket, feeding 15 seasons of statistics into its Bing Predicts engine and constantly iterating to optimize your chances. Just in case you haven't filled out your own bracket—and hurry up, the tournament starts tomorrow!—we asked Walter Sun, an architect of the Predicts algorithms, for mathematically proven tips to help you score office-pool glory.
The Rule Defensive efficiency is key. Bing tracks the number of points that are scored against a defense per 100 possessions—the lower, the better. Game pace slows in the tournament, which benefits D-minded squads.
The proof: Hawaii and Stephen F. Austin, two teams with strong defense, were ranked low but pulled off first-round upsets last year.
The Rule Strength of schedule weighs heavily in the NCAA’s ranking tool, so teams that ruled smaller conferences are likely to be underrated.
The Proof: A surprise 2014 win by Stephen F. Austin was actually predicted by Sun’s model—they’d gone 32–3 in a comparatively smaller conference and hadn’t lost since Thanksgiving.
The Rule You can base a model entirely on coach rankings—veterans Roy Williams and John Calipari have each led multiple teams to deep tournament runs. If it’s a toss-up, go with the established leader.
The Proof: Louisville’s Rick Pitino is 11–1 in the Sweet 16, and Michigan State’s Tom Izzo is 21–4 in the second game when playing twice in a weekend.
The Rule Tournament games may be played on neutral courts, but proximity to home still makes a difference. When in doubt, bet against the team that has journeyed the farthest and across the most time zones.
The Proof: In the past two tournaments, teams playing in their home states were 13–2.
This article appears in the March issue. Subscribe now.
