How AI is helping the NFL improve player safety

From the initial kickoff at Allegiant Stadium in Las Vegas for Super Bowl LVIII on Sunday, an artificial intelligence platform will be tracking every move on the field to help keep players safer.

Like many other professional sports leagues, the NFL has been at the leading edge of data-driven transformation for years. For example, in 2015 the league dramatically increased its data collection efforts by equipping all players with RFID sensors that pinpoint every player’s field position, speed, distance traveled, and acceleration in real-time. This season, the NFL has worked closely with Amazon Web Services (AWS) to debut a new club portal for their joint effort: Digital Athlete.

Digital Athlete is a platform that leverages AI and machine learning (ML) to predict from plays and body positions which players are at the highest risk of injury. The platform draws data from the players’ RFID tags, 38 5K optical tracking cameras placed around the field capturing 60 frames per second, as well as other data such as weather, equipment, and play type to build a complete view of players’ experiences. One of those data sources is the Next Generation Stats System (NGS), which captures real-time location, speed, and acceleration data for every player.

During each week of games, Digital Athlete captures and processes 6.8 million video frames and documents about 100 million locations and positions of players on the field. During practices, it processes around 15,000 miles of player tracking data per week — translating to more than 500 million data points.

“We’re running millions of simulations on in-game scenarios to tell teams which players are at the highest risk of potential injury, and they use that information to develop individualized injury prevention courses,” says Julie Souza, global head of sports at AWS.

Souza has headed up the Sports practice at AWS for more than three years after stints as head of business development and strategy at both ESPN and Second Spectrum, a data tracking and analytics provider for the NBA and other sports leagues. Now she and her team at AWS are helping sports and entertainment organizations build data-driven solutions that encompass everything from fan engagement and venue management to game strategy, scouting, and rules development.

The NFL piloted Digital Athlete last season and made it available to all 32 teams in the current season.

Changing the game

The first step in building Digital Athlete was using computer vision and ML to teach the AI to glean information from game and practice footage. For example, before the AI platform could track head impacts, it needed to ingest images of helmets from all angles to learn how to identify helmets. Once it was able to identify helmets, it was taught to recognize helmet impacts and cross-reference NGS data to determine which players were involved.

By using all the data at its disposal, Digital Athlete can reconstruct the conditions of how and when an injury occurred and run simulations of any play using different sets of players. Risk Mitigation Modeling can then be used to analyze training data and determine a player’s ideal training volume while minimizing injury risk. The team is currently working on a feature called pose estimation, which assesses players’ movements through space and time to better understand how body positioning can lead to injury.

Souza notes that this data is not only helpful for creating personalized training programs for players, but it’s also driving decision-making at the league level. The data used by Digital Athlete was a key factor in the NFL’s new fair catch rule for kickoffs, which debuted in 2023. The old rule required teams to attempt to catch and return a kickoff unless the kicker kicked the ball into or past the end zone. Now kick returners can call for a fair catch even if the ball is kicked short of the end zone, ending the kick return play, and putting the football on the returning team’s 25-yard line.

The goal of the new rule was to reduce kickoff returns by 7%, which the data suggested would lead to a 15% reduction in concussions from those plays.

“There was a reduction in kickoff runbacks, which is when you get more of that head-on-head situation,” Souza says. “I think that indicates how rules and how you play the game is changing.”

Part of Digital Athlete’s aim is to help uncover similar correlations between play scenarios and injury outcomes to shed light on risks that can be mitigated.

“If we can find the particular plays or rules that facilitate a greater likelihood of injury, then those rules can be changed,” she says.

Data over instincts

Ultimately, the goal of Digital Athlete is to use data rather than hunches and instinct to understand what’s happening on the field during games and practices. This has been borne out in other areas of the game, like the increasing tendency by teams, informed by analytics, to attempt fourth-down conversions.

“We couldn’t talk about stuff like this before because we didn’t really know,” Souza says. “There were all these hunches and things like that, stuff they knew in their gut. Tell me what you know instinctively and let’s put math to it. I bet we could try to prove that out or disprove it.”

Souza says this is true of all businesses, not just sports.

“It’s really about having a mindset where you’re curious,” she says. “The first thing is having a data strategy, having a foundation of data, and then asking questions of it.”

After that, she says, successful data-driven transformation requires knowing that building out AI capabilities is an iterative process and patience is required to allow those capabilities to grow over time.

“You’re not building a model, setting it, and going, right? The model gets smarter as you go,” Souza says.