Michigan-Alabama Rewind: How Our AI Predictions Performed

Getty Michigan's J.J. McCarthy and Blake Corum celebrate after beating Alabama in overtime to win the Rose Bowl.

When we looked last week at Heavy Sports’ five AI-powered projections for the Michigan-Alabama game, this was our synopsis of the forecast:

Lots of offense, relatively speaking. A heavy dose of Blake Corum. And at the end of the day, a Michigan football victory over Alabama on New Year’s Day in the Rose Bowl.

So how did the projections do? Overall, pretty darn well, though we benefited significantly from the game going to overtime.

Before we completely turn our attention to next week’s Michigan-Washington national championship game, here’s a look at how the projections performed, at a broad level and on specific items:

Michigan-Alabama Projections Scorecard: The Big Picture Was Spot-On, With a Big Assist from Overtime

Our forecast called for a higher points total and a larger Michigan margin of victory than the consensus of sportsbooks in our database. When Michigan trailed Alabama 20-13 with less than two minutes left in the fourth quarter, it looked highly plausible that the forecast would have whiffed on *both* of those elements. (With 4:35 left in the fourth quarter, ESPN’s win probability metric had Alabama at just shy of 89% to win.)

Things changed massively when J.J. McCarthy hit Roman Wilson for a 4-yard touchdown pass that brought the Wolverines within an extra point of tying the game at 20 with 1:34 left in regulation. (James Turner’s extra point did exactly that. More on Turner later in this post.)

If you evaluated the big-picture forecast on regulation alone, you’d conclude that our model, powered by Quarter 4, was slightly *less* accurate on the point differential and points total than the sportsbook consensus. Regulation featured 40 total points and a 20-20 tie.

We’ll do our best here to acknowledge the good, the bad, and the ugly. And the regulation-only snapshot wasn’t exactly a showpiece for eh model.

That said: If you’d read our projections post last week and placed a parlay bet on Michigan to cover a 3.5-point spread and the points total to go over 46.5, you’d have earned a handsome return.

Blake Corum’s 16-yard touchdown run and the Wolverines’ subsequent defensive stop sent Michigan to the championship game and left the model with a 2-for-2 performance on the correct winner and the points total.

Aside from the points total and Michigan margin of victory, our post last week highlighted three other elements: Performance by Michigan running back Blake Corum; a low-turnover output by both teams’ starting quarterbacks; and the importance of Michigan kicker James Turner.)

Here’s how the model fared on those elements:

The Model Was Right About Blake Corum’s Workload, but a Bit Too Optimistic About His Rushing Performance

The first header in our projections post posited that, according to the model, Corum would “Run All Over the Alabama Defense.”

Whether that actually happened depends on how you define a few things. Corum wasn’t unstoppable by any stretch of the imagination, and the model was overly optimistic about Corum’s rushing yards total. The model, though, came close to nailing Corum’s number of carries. And it’s not like Corum was a non-factor. Far from it, he scored two of Michigan’s four touchdowns — one receiving, one rushing — including the game-winner in overtime.

A snapshot of Corum’s performance compared to the model’s projections and Corum’s season averages:

The Model Was Mostly Right That Quarterbacks J.J. McCarthy & Jalen Milroe Would Take Care of the Ball

The model’s expectation that both McCarthy and Jalen Milroe would play relatively clean games survived a *major* scare in the game’s first play from scrimmage, when McCarthy threw what was ruled on the field to be an interception.

Fortunately for Michigan — and the model’s accuracy — officials reviewed and then overturned the call, ruling that the Alabama defender on the play had stepped out of bounds and failed to re-establish himself in-bounds before intercepting the pass.

McCarthy shook off the initial hiccup and played what was, given the stage, one of the best games of his career. McCarthy finished 17-for-27 passing, with 3 touchdowns and no interceptions, and was at his best late, engineering an 8-play, 75-yard touchdown drive with the season on the line to tie the game late in the fourth.

Milroe, for his part, also turned in an interception-free game, though he did lose a fumble for one of the game’s two turnovers.

The Model Was Right That James Turner Would Be a Key Player, but Missed High on His Field Goal Attempts

Our headline prediction that “Michigan Kicker James Turner Will Be a Key Player” proved accurate in one clear sense: The extra point that Turner attempted — and made — at the end of regulation was about as high-leverage an extra point as you can have, given that a miss would have more or less ended the game in favor of Alabama.

If you look under the hood, though, the results are a bit less flattering for the model, which “thought” Turner would attempt 2.1 field goals and make 1.6. In fact, Turner attempted only one field goal — a miss from 49 yards early in the fourth quarter, with Michigan trailing 17-13.