AI in sports: A look back at the analytics pioneers
Where once baseball's Oakland A's gained a strategic edge by discovering undervalued statistics, teams across all sports are now using emerging technologies to inform decisions.
Editor's note: This is the first article in a three-part series examining the evolution of AI in sports.
Once a futuristic fantasy that could capture reams of previously unattainable data and turn it into actionable insights, AI is no longer the next frontier in sports analytics.
Now, it's the norm.
There are strata within the different sports -- and teams within each sport use AI in their own way -- but its deployment to analyze data in ways previously impossible is now widespread across all major sports, according to Peter Zaimes, a senior lecturer at the University of New Hampshire and founder of the university's Sports Analytics Lab.
"Everybody is using AI now," he said. "It's still to different degrees in terms of accessing data -- what teams choose to access is still going to differ -- but it's widespread. … I can't point to any teams in the major sports not using it."
While now widespread, AI representing the evolution of analytics in sports is still somewhat nascent.
Just as OpenAI's November 2022 launch of ChatGPT marked sudden and significant improvement in generative AI (GenAI) technology, it has similarly enabled the extensive use of AI in sports.
Before GenAI -- and, more recently, the emergence of agentic AI -- analytics in sports was limited by the human capital needed to manage, model and analyze data, as well as the constraints of the systems being used to ingest, integrate and prepare data. Most of the same data now used to inform AI models and applications was there to be analyzed in reports and dashboards, but making use of it all was impossible.
GenAI enabled a breakthrough, reducing and eliminating some of the previous barriers to using AI to evaluate players, scout opponents and determine strategy. And it has opened a future filled with new ways of making the litany of decisions -- personnel, training, strategic planning and in-game choices -- that go into determining the outcome of a given game.
[AI in sports] was inevitable. In any super-competitive industry -- especially sports, where there are only a certain number of teams trying to win a championship -- every team wants to try to get a step ahead.
Addison HunsickerSenior manager of soccer analytics, Philadelphia Union
"[AI in sports] was inevitable," said Addison Hunsicker, senior manager of soccer analytics for the Philadelphia Union of Major League Soccer. "In any super-competitive industry -- especially sports, where there are only a certain number of teams trying to win a championship -- every team wants to try to get a step ahead."
Power hitters were put in the middle of baseball lineups because of their home run prowess, running backs handed footballs with frequency because of the yardage they amassed, plays designed for the tallest basketball players because they had the highest shooting percentages, swift skaters with quick hands put at center because of their scoring prowess.
Analysis became more sophisticated over the decades. For example, as baseball evolved, lineup decisions and pitching moves were made factoring in the statistically demonstrated tenet that it's advantageous for hitters to face pitchers with the opposite dominant hand -- right-handed batters fare better against left-handed pitchers – and, conversely, advantageous for pitchers to face hitters with the same dominant hand.
But it wasn't until the turn of the 21st century that advanced analysis, true analytics, came to sports.
The Oakland A's were pioneers, realizing that certain statistics such as on-base percentage and walks were undervalued, and using those realizations to their advantage. Over the ensuing two decades, analytics spread, becoming pervasive in all major sports.
Because sports is a copycat industry with teams following the leads of others that have success, analytics spread quickly in baseball. Basketball and soccer followed suit, with the Houston Rockets among the analytics pioneers in the late 2000s and international soccer clubs including AC Milan investing in analytics departments the same decade.
Football was perhaps the slowest professional sport to widely adopt analytics.
Kelvin Beachum, an offensive lineman with the Arizona Cardinals who began playing professionally with the Pittsburgh Steelers in 2012, wasn't exposed to analytics until a few years into his career.
"I would say 2013-14 is when analytics started to become a thing," he said, noting that most of the data used by NFL teams was for player personnel making decisions, rather than for improving player performance, scouting opponents and making strategic decisions.
Finally, late in the 2010s, players began getting information on opponent tendencies and their own performance to gain advantages and improve their play, Beachum continued.
"The last 5 to 7 years it's become something they [provide] so I can look at defensive ends that I'm about to face over the next couple of weeks and see where and when they're lining up, what kind of defensive tendencies they might have," he said. "Those statistics, those tendencies, became more prevalent over the last five years. But when [analytics in the NFL] started, it was just to evaluate an individual."
AI, meanwhile, was emerging as a means of advancing sports analytics by the start of the 2020s, with teams and leagues expanding into machine learning and predictive modeling. Baseball and soccer, sports with a dearth of action compared with others but featuring bursts of intense activity, lent themselves to AI.
"Not all teams were doing it," UNH's Zaimes said. "Not all teams were as advanced. But the technology did exist."
For example, with camaras strategically placed throughout baseball stadiums, hundreds of images are captured of each pitch that enable teams, beyond speed, to know the spin rate of the ball, which contributes to how the ball moves in any direction, and the amount of vertical and horizontal movement.
Employing such metrics compiled over the course of time, the Minnesota Twins in 2020, using a data management and model development tools from data and analytics vendor Databricks, built simulation models that helped the team quantify how much of what happens when a pitch is in a given location is luck versus skill.
Similarly, the Bundesliga -- Germany's top soccer league -- partnered with AWS in 2020 to capture millions of data points per match and combine those with historical data in a machine learning model to calculate probabilities such as the likelihood of any shot taken throughout a match resulting in a goal.
Such use of AI to gain strategic advantages, however, was limited until the past few years.
After teams like the Oakland A's of the early 2000s pioneered the use of analytics in sports, AI is now taking statistical analysis to a new level.
Barriers to AI
Lack or technical expertise was a hindrance for many teams.
Not all, especially those below the top level in their sport, can invest in entire departments dedicated to data science. Even in Major League Soccer, the top U.S. soccer league's lack of manpower was a major barrier to analysis beyond basic statistics, let alone AI, according to Hunsicker, who joined the Union in 2020.
"There really wasn't anything in 2020, no data infrastructure, no architecture to have everything in one spot," he said. "There was a lot of my boss and I sending files back and forth [to analyze data]. It was all manually stored on our own laptops in spreadsheets and presentations and OneDrive."
The available technology itself was another hindrance, with AI and machine learning development platforms sometimes unable to process the data volume needed to properly teach models to deliver accurate outcomes.
"A lot of it was data processing speed, the number of metrics that you're actually able to process to make a decision," Zaimes said. "All the advanced analytics metrics were there, but not everybody had processing mechanisms to deal with large-scale data sets to make predictions."
Still another hindrance was isolated data, which is one that still prevents NFL teams and players from getting the full benefit of AI-driven analysis when it comes to player performance, according to Beachum.
"Right now, there are so many different silos," he said, specifically referencing data a company called Catapult collects about running, force plate testing to determine the amount of force players use during movements such as sprinting and jumping, body monitor data and neurological data. "There are so many different platforms, so many different use cases, but there is nothing that pulls it all together."
Recent advances in technology have reduced the need to have a team of experts to build and analyze AI models, helped break down some data silos and improved processing speeds. And now, AI is becoming pervasive.
"There's a race to incorporate GenAI and figure out how to apply it to use cases," Hunsicker said.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.
