
Come Opening Day, when a manager takes the slow walk to the mound to pull his starting pitcher, will it be because of a gut feeling about his ace or what the analytics show about the batter matchup? Or, will it be because Artificial Intelligence told him to do so?
“I’m telling you AI will be working in real-time and… will come up with insights that will knock our socks off—on injuries, on recruitment, on game tactics,” said Roger Mitchell, a venture capital investor and owner of Albachiara, an Italy-based digital change consultancy specializing sports and media. “It is total game theory.”
You may not see a robot between the Gatorade jug and a bag of sunflower seeds in dugouts, but major sports teams in Europe and the U.S. are starting to embrace AI as the next step in the Moneyball approach to team management. Liverpool FC has a theoretical physicist and an astrophysicist on staff doing research on game management—and their findings are credited with defensive formations that slashed goals allowed last year. At the end of 2020, rival Manchester City started a contest with Google to crowd-source AI strategies and then hired its own lead AI scientist, Laurie Shaw, an astrophysicist who taught at Harvard and designed trading systems for a hedge fund. In his spare time, Shaw writes data-driven blog posts about player formations and what age goal-scorers peak. Stateside, at least one Major League Baseball club is testing AI, as are teams in the NFL, NBA and Major League Soccer.
Much of the shift toward AI is being driven by team owners, who are increasingly coming from finance and bringing their ideas around risk with them, says Mitchell. “It’s the mindset about management of volatility; they want to minimize as much as possible risk,” said Mitchell in a video call from Italy. “If I’m going to do a capital expenditure between full-year contracts and transfer fees of close to a billion dollars, I want the data. The finance guys will not accept the old lads in the room saying, ‘Trust us, this guy will be OK.’”
The biggest risk AI is taking on: injuries. Mitchell invested in Zone7, a California-based outfit that specializes in developing the complex mathematical strategies that are called AI and applying them to reams of data in order to project when a player may be at risk for an injury. Other investors in the business include Dallas Mavericks center Kristaps Porzingis, Manchester United center back Phil Jones, women’s soccer great Bex Smith and Shenzhen FC coach Jordi Cruyff.
“There’s a myth about AI, that it’s a magical black box that will know everything,” said co-founder and CEO Tal Brown from his Palo Alto office. “But we can forecast really, really accurately. We can place you in a red zone, where [that] implies you’re highly likely to break down over the next few days. And that is useable, as long as you can figure out the intervention.”
Brown, who previously developed AI for Salesforce, keeps the names of most of Zone7’s team clients close to the vest, but he says the company is working with “dozens” of top league European and U.S. clubs. Among them: Serie A’s Bologna, Scotland’s Rangers F.C. and Getafe of Spain’s La Liga. No matter what the sport, teams are asking Zone7 AI to help them figure out the short-term trade-off between player performance and risk, Brown explained. “One of the key things people care about is: Can I play this guy a full game tomorrow? And it’s not just a binary question of play or don’t play. How will this player’s risk change from playing 20 minutes or 40 minutes?”
AI answers this by taking all sorts of information and crunching it through complex mathematics to find patterns and correlations a person may never notice. Generally speaking, AI uses computing speed to make millions of calculations between potentially thousands of data points, finding statistically significant connections among two or more points in combination. Whether deployed on medical imagery to see risk indicators a team doctor might miss or to improve officiating, AI has near limitless applicability. But it does rely on data, which plays to Zone7’s advantage since it is among one of “older” sports AI outfits, founded in mid-2017. For teams, using AI means feeding the algorithms with everything from a player’s physical statistics to past injury history. Practice and game playing information from wearable trackers, as well as data from other players, too, can help AI make better insights over time.
One of Zone7’s publicly disclosed Major League Soccer clients, Real Salt Lake (Toronto FC is the other), began using Zone7 for its 2019 season. The tech firm says it identified nine of 13 injuries RSL experienced, tagging them as high risk or medium risk up to seven days before the injury occurred. Because of the training staff’s early insights into their athletes’ conditions, the team suffered 30 days lost per month by its players in May to October 2019, compared to 52.3 days a month the same period in 2018 without using AI, according to a case study provided by Zone7.
Zone7 began working with an undisclosed Major League Baseball franchise in 2018, also intent on managing injuries. In a separate, backward-looking analysis of a MLB franchise’s 125 major league and minor league baseball pitchers in the 2019 season, Zone7 says its AI would have detected 71% of injuries. In that study, the AI was ‘trained’ on third-party data on MLB and minor league players from 2015 onward, and then released on the unseen 2019 data. Zone7 also identified factors that contribute to pitcher injuries and ranked them according to the risk they add to a player. For instance, release position adds the most risk to an arm, followed by release acceleration then release spin rate.
AI like this is, by its nature, inscrutable to outsiders, being a collection of highly complex calculations that needs to be shielded from competitors. That means results can sometimes seem surprising and perhaps even contrary to perceived wisdom. For instance, the pitcher study tagged the knuckleball as the pitch having the most “impact on risk,” while curveballs—long the forbidden pitch for Little League hurlers—have about half the impact. Splitters? They have the least impact among the eight pitches tallied, and less than one-third the impact of the knuckler Phil Niekro threw in the big leagues until he was 48. Pitch count, this century’s contribution to the pitcher management gospels, also seems to have relatively little impact on risk, according to the analysis.
“AI is not the goal. The goal is to be more effective, more efficient as a human operator,” Brown explained. “We are a human performance company. We’ve launched the product with rugby [Wales’ Ospreys] and are doing this with NFL and NBA clients. We also think people in more environments are susceptible to well-being risk.”
The company has partnered with healthcare providers and the military, looking into managing fatigue and burnout in those non-sports areas. But change consultant Mitchell sees plenty of growth for AI in the years ahead within sports alone, from the plays on the field to which players get their names called by the NFL commissioner and when. “Player recruitment in America is all the draft, and the draft is going to be AI-dominated,” he said. “You will see Ph.D. physics guys all around the draft rooms, all around them.”