Over the past few years, you may have noticed a curious element in NBA game previews: Several publications have begun to quote the NBA 2K series when previewing real-life matchups between NBA teams in the regular season. As the latest edition of the video game series, NBA 2K25 is often trotted out for game previews for the 2024/25 season. But how useful is it?
The first thing to say is that there are certainly limitations for using the game to simulate game outcomes. Perhaps the best way to put it is that it can be a useful tool to provide some context for a game, but it probably shouldn’t be your primary tool for making predictions about NBA spreads or fantasy picks.
Yet, it is useful. In the preseason, USA Today simulated the Oklahoma City Thunder’s entire season using NBA 2K25. In fact, the media outlet used the game to simulate the outcome ten times and published details of every team’s final standings and route through the Playoffs. The results were interesting, but they weren’t markedly different from what you see with the season overs/unders on a sportsbook or supercomputer projection.
USA Today Models Entire Season
For example, USA Today’s model picked out the Dallas Mavericks (61-21) and Boston Celtics (62-20) as leading the Western Conference and Eastern Conference, respectively. The focus of the article, Oklahoma City Thunder, were projected to finish the season on 59 wins. None of this is particularly earth-shattering news. Moreover, the projection failed to pinpoint certain anomalies, such as the Cleveland Cavaliers’ red-hot start (12-0) at the time of writing.
In addition, the modeling of an entire season is much different from modeling an individual game. Last season, the Boston Celtics were historically good and completely dominant across the regular season and postseason, yet they still lost games against the Charlotte Hornets and Atlanta Hawks (twice in a row), teams that were vastly inferior to the championship-winning Celtics. You can’t model these things. They just happen.
Real-time data is an issue
Clearly, part of the issue is what also occurs with AI models when used for simulating outcomes. The bots, or computer games in this instance, are only as good as the data they have access to. In addition, there is an inability to react to human qualities, such as dips in form. A case in point is the form of Victor Wembanyama this season. The French kid is going to be a superstar – there’s little doubt of that. But the reigning Rookie of the Year is having a few issues in his sophomore season. He’s still thrilling to watch, but he has looked clunky at times.
Wembanyama’s form perfectly illustrates the issue: It’s something that we can put a finger on – perhaps it’s the weight of expectation; maybe it’s lethargy after an Olympic summer – yet it is something that you cannot really model with statistics. “Wemby” will be just fine in the grand scheme of things, but he’s young, and in an individual game, his numbers might not match up to what is projected.
In the end, the lack of adaptation to human variables is the main issue. Unlike real-time betting algorithms or models that can respond to updated injury reports, day-to-day team morale, or even last-minute coaching adjustments, NBA 2K25 is based on static data updated periodically. On the other hand, it can be useful to provide fantasy players with insights into player metrics. For example, if the simulation predicts that a player like Shai Gilgeous-Alexander will be dominant against weaker defenses, this insight might help a fantasy player spot a good opportunity. Still, the limitations remain. By all means, check out NBA 2K25 for your sports betting or fantasy sims, but don’t put too much stock in it: It can only do so much.