Skip to content

NHL · 2026-27

StatGrades AI Model

One model, a moneyline call on every NHL game: each team's attack and defence from expected goals, weighted to the last few weeks, plus home ice, against the books' prices. The AI explains each pick in plain English. Graded at the price, in units, after the final horn. No over/under picks: in testing and this season, ours didn't beat the market.

These picks are for information and entertainment, not financial advice, and no pick is a sure thing. StatGrades doesn't take bets or link you to a sportsbook account. Odds are shown for information and can change at any moment; check with the book before you bet. 21+ only where legal. Gambling problem? Call 1-800-GAMBLER.

The record is open. Every pick is free with All-Access this season.

All-Access shows every pick in full: the side, the price, the predicted score and the AI's plain-English why for every game. It's $12.99 a month, and free with an account for the rest of the season.

Boston Bruins @ Detroit Red Wings

Wed, Nov 18 • 7:30 PM ETFinal · BOS 4, DET 1
Matchup Stats

🔒 There's a pick for this game. All-Access shows it in full, free this season.

Montréal Canadiens @ New Jersey Devils

Wed, Nov 18 • 7:30 PM ETFinal · MTL 3, NJD 2
Matchup Stats

🔒 There's a pick for this game. All-Access shows it in full, free this season.

Colorado Avalanche @ Utah Mammoth

Wed, Nov 18 • 10:00 PM ETFinal · COL 1, UTA 4
Matchup Stats

🔒 There's a pick for this game. All-Access shows it in full, free this season.

StatGrades AI Model record

Every game with a line: each team's attack and defence from expected goals, weighted to the last few weeks, plus home ice, compared with DraftKings' moneyline, with the AI explaining each pick. Refreshed until puck drop and graded at its last price. It makes no total pick (see below). Picks made by the earlier StatGrades AI, retired on Oct 26, 2026, are not in this record.

Top calls

14-23

−4.71u at the price · 38% won

Moneyline

135-176

−37.06u at the price · 43% won

The StatGrades AI Model rates every team's attack and defence from our own expected goals (where and how each shot was taken), with a little weight on actual goals, counting the last few weeks most, plus home ice. From the two teams' expected scores it works out each side's chance to win (overtime and shootout included) and compares it with DraftKings' moneyline with the book's cut taken out. It takes the side its chance favours more than the price does; a gap of 6 points of win chance or more is a top call. We tested it before it went live, against DraftKings' closing lines for February to April 2026, after choosing its settings on December and January: on the moneyline it was about as accurate as the market (0.670 log loss against the market's 0.672; lower is better, and the gap is within chance), and its picks with a gap of 4 points or more went 128-132 for +29.26u. On totals it was worse than the market, so it makes no total pick. It counts every goalie as average: in testing, goalies' past numbers didn't help it. Early in the season its ratings rest on a few games, a stretch the test couldn't cover, so until both teams have played 10 games it makes no top calls and never more than a lean. It also makes no top calls and never more than a lean at +200 or longer: when the market gave a team under 30%, the model said about 36% and those teams won about 22% (64 games). It doesn't account for rest, so it can lean toward a tired team the market has already marked down.

The AI's part: it reads the model's numbers and writes the plain-English why under each pick. It never changes a pick. Until Oct 26, 2026 the AI also made picks of its own on up to three marquee games a night; they stopped, so one model makes every pick.

The boxes on each game: Expected Goals is how many goals the model expects each team to score, overtime included (most games land near 3 a side, so we show a decimal instead of rounding everything to 4-3). The arrow marks the team it thinks is more likely to win, which is not always our pick: the pick is the side whose price pays more than the model's chance is worth, so it is sometimes the team the model expects to lose. Before puck drop each box is the best line on offer, with the book that has it, and the percentage is our chance at that line. ML is the model's own chance of each team winning; the puck line and the total come from the same projection, counting goals the way hockey scores fall. Our pick is outlined green, and turns red if it loses.