The NFL and Amazon are using AI to invent new football stats

The National Football League, like most professional sporting industries, is embracing artificial intelligence. Through a partnership with Amazon Web Services called Next Gen Stats, the NFL is hoping that intelligent algorithms, with the help of high-tech data collection tools, will be able to extract meaningful data from games and decipher patterns in player performances. AWS says it was inspired by submissions to the 2023 Big Data Bowl, an annual software competition organized by the NFL, when it set out to invent a new category of analytics that pertains to the analysis of “pressure” in the game of football.

AWS helped build out AI-powered algorithms that can analyze player behavior on the field and can pick up on how aggressive a defender played, how fast they were and even how quickly a quarterback responded. This granular data quantifies pressure and in doing so, allows game analysts to dissect the strategies that might influence plays. This innovative suite of analytics rises above traditional statistics that are limited in how much they can reveal. While traditional data can tell you if a rusher passes a quarterback, it may not be able to provide insights on how much of a fight was put up. This is where the pressure probability being tracked by “Next Gen Stats” delves into more detail.

The AWS and NFL partners have focused on developing machine-learning models that can provide data relating to three areas in game play, according to Amazon. The first application is giving the AI the ability to identify blockers and pass rushers in pass plays. Second, teaching the tool how to quantify “pressure” in a game. And lastly, the development of a process to detect individual blocker-rusher matchups. Ultimately, the development of this AI-tracking technology provides professionals in the football league with valuable information on player stats that can help scouts or coaches select new players. For example, knowing which player blocked or passed a rusher may help determine if they are a good fit for an offensive lineup.

In the game of football, quantifying the performance of offensive players and the rushers that tackle them can be a difficult feat, even for game experts who have the eye for these quick movements. Player reactions can happen in split moments and an individual’s performance in these high-speed exchanges can be hard to track and let alone quantify. Things like how close a defender got to the offensive lineup can help a coach understand the strength of their plays.

The NFL collects data for these AI-powered processing softwares using tools it installs in its own fields. In every participating NFL venue, there are at least 20-30 ultra-wide band receivers inside the field and there are 2-3 radio-frequency identification (RFID) tags inside each players’ shoulder pads and on other game gear, like balls and posts. These data transmitters collect information that is fed through a graphic neural network model (GNN), which allows the data to be relayed in real time. Using AI, the stats being extracted can be made into meaningful insights.

These insights can look like a number of interactive graphics found on the Next Gen Stat game landing page. You can get a breakdown of individual player movements in any given game in 2D models and graphs. For example, you can track the movement of both players and the ball during a 40-yard passing play in the San Francisco 49ers’ game vs. the New York Giants on September 21.

While the AI tool is hosted on AWS infrastructure, the final product is a compilation of a multidisciplinary partnership between the NFL, Zebra Technologies, and Wilson Sporting Goods. The Next Gen Stats project, which began in 2017, now makes up a data pipeline that contains historical data available for every pass play since 2018.

Meanwhile, in a parallel project, AWS engineers shared that they are working on automating the identification of blockers and rushers so that eventually, the AI models could autonomously ID players’ roles on the field. Currently, this kind of information is gathered manually through charting is prone to label errors, and often takes hours to generate by humans.