What is NBA win probability?
A win probability is one number for how likely a team is to win a game. Here is what it means, how it is made, and how to tell a good one from a bad one.
BY UĞUR ALTUNBAŞ · OCT 23, 2026
A number for how often this game goes one way.
A win probability is a single number for how likely a team is to win a game. When Clutch gives the Thunder 90% against the Clippers tonight, it means that in games like this one, a team in that position wins about 90 times in 100 and loses about 10. It is not a promise: a 90% favourite should lose about one game in 10.
How to read the number
| NUMBER | CLUTCH SAYS | WHAT IT MEANS |
|---|---|---|
| 50–54% | Narrowly | Close to a coin flip. The favourite wins a little more often than not. |
| 55–61% | By a possession | A real edge, but the other side wins about two games in five. |
| 62–71% | Clearly | A solid favourite. Upsets still come about one game in three or four. |
| 72% and up | Comfortably | A strong favourite. Even at 80%, one game in five goes the other way. |
The words are the ones over each number on Clutch's game pages. The number always belongs to the side Clutch favours; the other side's is 100 minus it.
Where the number comes from
A pregame win probability starts from how good each team is this season and lately, then adjusts for the night itself: rest and back-to-backs, travel, altitude, a neutral floor, and who is listed on the league's official injury report, weighted by how much each player has been worth. Home court matters too: home teams won 54.6% and 55.5% of the games in 2024-25 and 2025-26.
Clutch's model learns how much each of these is worth from past seasons, then shows the top reasons behind every number in plain words. How Clutch's model works →
The market's number
The prices on a game can be read as a win probability too: take the margin out and both sides add up to 100%. Just before tip that number has seen almost all of the day's news, which makes it the hardest one to improve on. Clutch shows it beside its own number on every game and grades both after the final.
Pregame and live
TV broadcasts and score apps often show a live win probability that moves with every possession. Clutch's number is the pregame one: made on the day of the game, sealed in public before tip, and graded after the final, so it can be checked against what happened.
How to tell a good one
Two checks. First, does it come true as often as it says? Of all the games a source called at 70%, the favourite should win about 70 in every 100. That is calibration. Second, is it confident when it's right and cautious when it isn't? The Brier score grades that: lower is better, and a coin flip on every game scores 0.250.
Clutch prints both for its own numbers, band by band, beside ESPN BPI, the market and picking the home team. How accurate Clutch is → · How accurate is ESPN BPI? →
Why favourites lose
On Oct 24, Clutch's favourites won 6 of 8. Its numbers for those games added up to about 2.3 expected losses, so the night went about as the numbers said. A source whose favourites never lose isn't giving probabilities; over a season, a 70% favourite should lose about three games in ten.
The example and the night above are rebuilt from Clutch's data every day. More on reading the numbers →