A Decade of Innovation in NFL Next Gen Stats

In 2015, the NFL introduced Next Gen Stats (NGS) to begin measuring the true nature of the game that could no longer be seen from box scores alone. Using RFID embedded in player pads and the ball, along with ultrawideband receivers placed throughout the stadium, the system collected the positions of 22 players 10 times per second and the ball's position 25 times per second, tracking every play down to a few centimeters.

Initially, each team could only view its own team's data, but starting in 2018, league-wide data access opened up, unifying the standard for analysis. That same year, collaboration with AWS ramped up in earnest, and NGS evolved beyond a simple tracking experiment into core infrastructure supporting broadcasts and team operations. Early on, it started with basic metrics like distance between players, speed, and time to throw, but as AWS came on board, more complex modeling became possible.

The first representative advanced metric to emerge was completion probability. An XGBoost model on Amazon SageMaker combined variables such as quarterback pressure, pass distance, receiver separation, and distance from the sideline to express pass success likelihood and difficulty as a single probability. Afterward, AWS tools including Amazon QuickSight expanded real-time visualization and analytics for fans and broadcasts, and metrics like pressure probability led to more than 12 detailed metrics derived from a single rush. NGS has now grown to the point where it generates 500 to 1,000 statistics per play.

Deeper analysis expanded through the Big Data Bowl. Starting in 2018, this competition opened player tracking data to outside researchers, tasking them with problems like how efficiently defenders close space or how to predict player movement after a pass. One outcome from this was rush yards over expectation (RYOE), a metric showing how much better actual rushing yards were compared to the average expected value. This metric quickly appeared in the broadcast of the 2021 NFC Championship Game, demonstrating the speed of real-world application.

Currently, NGS has expanded to more than 75 ML models covering offense, defense, special teams, and overall game strategy. Among these, tackle probability calculates the likelihood of a successful tackle at the moment of contact, and defensive alerts predicts who will rush based on pre-snap movement and game situation, displaying it on broadcast screens. Notably, the latter sometimes fails to catch defensive misdirection, which ironically creates talking points for commentary.

The same tracking infrastructure has also been directly used for player safety and rule changes. The dynamic kickoff, introduced in the 2024 season, is the result of redesigning kickoffs that had too many high-speed collisions, with the league using NGS visualizations and space/speed analysis to find a format that maintains both safety and competitiveness. In 2025, the return rate rose to 75% (compared to 32% in 2024), and despite there being 1,157 more plays, lower-body injuries decreased by 35%, and the concussion rate remained lower than under the previous format.

The next stage is pose tracking. NFL stadiums are already equipped with 4K camera-based optical tracking systems, and now they capture x,y,z coordinates for 29 body parts—including shoulders, elbows, knees, hips, and hands—at 60 times per second. Footage from 16 angles is processed locally in about 700ms, and ML analysis runs in the cloud in under 100ms, keeping the entire pipeline under 1 second.

NGS can now go beyond the limits of 2D tracking, such as how the ball passed between defenders or rolled between bodies. Ultimately, the goal is a hybrid system that identifies a player's center of mass using RFID and combines it with skeletal data, with algorithms filling in obscured moments. Furthermore, with the opening of VR simulations for quarterback training, NGS is moving beyond being a tool that explains the game to becoming an operating system that changes it.