LAS VEGAS – National Football League CIO Michelle McKenna-Doyle may have never lined up under center and taken a real snap in a live game before, but she too must make intelligent reads just like any quarterback.
In her latest move, McKenna-Doyle has decided to make a bold decision to leverage machine learning technology to help evolve the game of football for coaches and scouts, while also improving fan engagement and protecting approximately 2,000 players throughout the 32-team league.
McKenna-Doyle is responsible for the NFL’s entire technology strategy and her plan has to extend outward to support the more than 180 million fans worldwide. The league also has another issue besides wins and losses: it’s also a large content creator with NFL.com and the NFL Network.
Machine learning is the focal point for a new NFL initiative called NextGen Stats, which is replacing the NFL’s existing player-tracking system. Powered by cloud provider Amazon Web Services (AWS), NextGen Stats will track all basic stats like touchdowns, interceptions, yards rushing, passing yards, tackles and fumbles, but it will also manufacture a new collection of stats such as real-time location, speed, and acceleration data. The data is analyzed on AWS and used to contextualize movement on the field and displayed as a new experience for fans to see on NFL Media properties, game broadcasts, third-party digital platforms, and in-venue displays. These new stats will also be available to players and coaches.
The issue for McKenna-Doyle boils down to the fact that the main stats take hours to collect manually. So for her it’s an efficiency problem. The other issue is delivery of these stats are simply not quick enough for such a fast-paced game.
The strategy being employed by McKenna-Doyle is to run these new stats in an automated fashion from all 31 stadiums, along with games being played in Mexico City and London, U.K.
The biggest change for players is that sensors with radio-frequency identification (RFID) tags will be embedded into all shoulder pads. “Now you can see how far Emmanuel Sanders of the Denver Broncos ran during the game and at what speed,” McKenna-Doyle said. “By the way, Sanders ran at 19KM per hour.”
AWS is also providing the NFL with a new cloud framework for NextGen Stats that will include cloud, artificial intelligence, and machine learning technologies to track live games for the media and post-analysis for football experts and pundits. “This will build a digital storytelling experience for fans,” she said.
McKenna-Doyle demonstrated what NextGen Stats is capable of at the recent Re:Invent Conference in Las Vegas. She showed a pass play run by the Buffalo Bills against its division rival New York Jets during a Thursday Night Football game this season. Bills quarterback Tyrod Taylor was in a formation that consisted of 35 basic routes he could call. In this formation there are at least 1,000 combinations based on many factors such as the players on the field or whether it’s a day or night game.
In those 35 routes, Taylor would most likely focus on nine of them, she said, but of those nine there are 50 advanced routes that are possible. This is why machine learning is needed because otherwise it would simply be too hard to draw out any insights from this particular play.
Machine learning can recognize and record all of those routes, and a quarterback like Taylor can learn from its deep insights to become more efficient. One of the deep insights being tracked is how long Taylor held onto the ball before throwing it and how that time impacted the receivers out on the field.
Machine learning can also look at field conditions at the time of play. Weather has always been a factor in the NFL, as the sport does not accommodate rain outs or rain delays like baseball and will play in snowy conditions. McKenna-Doyle said that weather will be fed into AWS Sagemaker and from there it will create models for all stakeholders. “The goal is to automate all formations and routes,” she added.
The play Taylor ran against the Jets resulted in a first down pass to Andre Holmes for 12 yards. McKenna-Doyle said that for most fans the play looks simple to execute because Taylor was untouched by the Jets defense.
NextGen Stats will be able to determine why that play was successful and quantify it. “It will look at key events such as offensive formation,” she said. “The play is called trips left [when three wide-receivers line up on the left side of the ball]. In pre-snap reads, the defensive formation matched all the receivers in a six-defensive back formation. The key match-up was Holmes and Jets defensive back Darryl Roberts. Holmes ran a post route [post routes usually has a wide receiver run in a straight line and then sharply cut towards the goal post] and NextGen Stats will look at when the ball was snapped and determine if the pass would be successful.”
New insights like pre-snap reads and probabilities of success are the main reason for the development of NextGen Stats, according to McKenna-Doyle. In the Taylor to Holmes pass play, NextGen Stats determined that it had a 60 per cent chance of completion, if Taylor decided to throw it to Holmes on that play.
But what can make NextGen Stats special is that during mid-play when the ball is in the air, it can read the play – through machine learning technology – and find out at that moment in time if Taylor made the right decision. When run through NextGen Stats that play now had a 75 per cent chance of being successful. So, McKenna-Doyle said, Taylor made a good decision to throw the ball at that time to Holmes.
This new arrangement with AWS will mean that the Seattle-based cloud provider will be called an official technology partner of the NFL.
For coaches and scouts, they will get records of each play and those plays will be in diagram form detailing things never thought of before such as fitness summaries for each player, heat maps to show exactly where each player went on the field.
Even NFL referees will be including in the NFL’s machine learning program as they will get analysis from heat maps as well.
“The best place to see a game is in the stadium and we will be sharing these NextGen Stats with fans in the stadium,” McKenna-Doyle added.