The NBA and AWS have built a new AI-powered metric called Leverage Score, designed to measure which players and possessions most influence winning. Instead of relying only on traditional box score stats, the system looks at the context around each possession and uses counterfactual modeling to estimate how much a specific action changed a team’s win probability. 

In the NBA’s framing, the idea is to capture impact that standard stats often miss. A rebound late in a close game does not carry the same weight as a routine rebound in garbage time, and a pass that draws defenders and creates an open shot may matter even if it produces no assist in the box score. Leverage Score is built to account for those moments by measuring not just what happened, but why it mattered. 

“Traditional stats capture outcomes. What makes Leverage Score unique is that it captures why those outcomes happened and who made them possible.” – Noah Thro, Software Engineer for NBA Stats Technology 

The system runs in near real time. AWS said the architecture processes roughly 2,500 events per second during peak load across more than 10 simultaneous games, using Amazon EKS, Apache Flink, Amazon S3, Amazon DocumentDB, AWS Direct Connect, and AWS Cloud WAN. Tracking data flows in at 25 frames per second, alongside event data and contextual metrics such as shot difficulty and defensive pressure. 

At the core is a LightGBM win probability model trained on three NBA seasons, about 3,700 games and 500,000 possessions. For each meaningful event, the system runs 3 to 5 alternative scenarios per possession to compare what actually happened with realistic outcomes such as a missed shot, a failed offensive rebound, or an avoided turnover. That gap is what becomes leverage. 

Leverage Score also distributes credit across all 10 players on the floor. AWS said the model uses Expected Field Goal Percentage (xFG%) and other contextual inputs to separate shot creation from shot finishing, while also accounting for on-ball defense, rebounding, and turnover impact. That gives the NBA a way to assign value to actions that often sit outside traditional counting stats. 

“The thing that unlocked Leverage Score for us was the other Inside the Game metrics. Expected field goal percentage sits at the core because it allows us to quantify the value of the shot and distribute that to the creator versus the shooter.” – Noah Thro, Software Engineer for NBA Stats Technology

For fans and media, the NBA said the metric is designed to surface the moments and players driving momentum in a game, not just the headline scorers. Because each value is tied to a specific play-by-play event, the league can connect Leverage Score to highlights and in-game storytelling. 

For the broader industry, the more interesting piece may be the workflow behind it. This is less about one new stat and more about a real-time analytics stack that can run scenario modeling, attribution, and decision logic fast enough to be part of the live product. AWS described the architecture as a reusable pattern for other kinds of impact analysis beyond sports, but in basketball, it gives the NBA a new way to explain which actions truly changed a game.