Numbers Game
Sports analytics icons Ari Kaplan, BS (BS ’92) and Dean Oliver, PhD (BS ’90) on the next era of game-changing data
by Dan Morrell
Illustration by Gary Neill
This year marks the 20th anniversary of Michael Lewis’s Moneyball—a book that captured a rising tide of data analysis in baseball and ushered in a metrics-driven era in the sports world and beyond.
Dean Oliver, PhD (BS ’90) an engineering consultant and aspiring basketball analyst at the time, landed a job at the Seattle Supersonics soon after the best-selling book was published and tracked down fellow Bay Area resident Lewis to buy him lunch. “When I was trying to get a job in basketball, I would say, ‘Did you read Moneyball? If you want to know how to do that in basketball, here’s how,’” recalls Oliver, now an assistant coach with the Washington Wizards. Ari Kaplan, BS (BS ’92) a longtime baseball front office analytics consultant and one of the earliest analytics acolytes, has watched the industry explode over the past two decades. “When I started working for a team, I was one of only a handful of people employed by an organization to do anything even remotely close to analytics,” says Kaplan. “And now teams across sports have 10, 20, 30 or more data and analytics staffers.”
In this joint interview, we talk to these two forefathers of the movement about the past, present, and future of quantified competition.
When we talk about the data-driven revolution in sports, we tend to focus on front offices’ use of metrics. How has it impacted how athletes approach performance?
Ari Kaplan: From day one, when I was an undergrad at Caltech and [former Dodgers GM] Fred Claire let me into the dugout, I went from being a fan to sitting next to stars like Eddie Murray, Kirk Gibson, and Orel Hershiser. And players back then loved the idea of statistical analysis, because they conceptually loved getting proper credit for their work. That was the first phase. The next leap was strategy: getting information players can use to either understand the strengths, weaknesses, and habits of their opponent, or to develop as an athlete themselves.
Dean, on the basketball side, was there a reticence to embrace statistics? Or did the players immediately see the value?
Dean Oliver: It was more gradual, and it is still working its way into the player development realm. But over the last few years, I’ve had the opportunity to work more closely with players. [Former Wizards forward] Kristaps Porzingis had his best year this year, and he and I worked together on three-point shooting, shot choice, and different elements of his rebounding. And it’s detailed stuff, not “you need to rebound better.” Often it is positioning: he’s going too far out as opposed to sticking around the basket. So there are mechanical elements that can be tweaked with the data that we have right now. And as data evolves, you can get more into some of the detailed mechanics.
How has artificial intelligence impacted your work?
Oliver: Machine learning changed basketball roughly 10 years ago. The video tracking data that started coming in from the NBA—where AI had converted images into dots—wasn’t available league-wide originally; individual teams were actually paying for it. The company that provided it started giving out some of the data to academics. And those academics said, “Oh, I know what to do with this. Let me use this to evaluate rebounding and how much the positioning matters.” Then they expanded it to ask, “Can I use this data to identify a pick-and-roll or a post up?” The automatic tracking of all that dramatically changed analysis in the NBA. Now every team has access to that same information, and they have consistent measures of it. Machine learning gave a kind of a language to the moving dots, and that language was then translated to the coaches, to management, and to analysts like me.
Kaplan: Human talent evaluators have their roles in this work, too. Jerry Krause, who was the Bulls GM and my personal mentor in Chicago for many years, was non-analytical and non-mathematical, but a genius at seeing which players he thought would succeed and which he could give up on. Artificial intelligence and machine learning can now look at the words that talent evaluators like him write in a text report and pick up clues—perhaps, for instance, certain phrases that tend to be more predictive of how a player might do in the long term.
Oliver: I want to note that we’re not just doing data science here. Machine learning is data science. I’m actually working with professors and the basketball coach at Caltech, for instance, on developing some theory around how people interact in a team setting and the value of players in a team setting. It has required some of the deepest math that I’ve done since Caltech, and it’s fascinating stuff. I’m looking at roles on a team, with some players who take more shots and then role players who have to be more efficient. In theory, no player can maintain their efficiency completely as they shoot more and more because they’re going to take worse and worse shots. And so we put those things together to get an idea for how to balance the players on a team.
What are the metrics that front offices are thinking about these days as they’re weighing player decisions?
Kaplan: One thing they might be considering is how to evaluate which players to sign long-term. The old way was to just look at a player in isolation. But each ballpark has different dimensions and weather conditions, and if you are on a team where you’re the sole star, opponents are going to pitch around you to get to an easier player. So teams are now looking at what would happen in context—if they took a certain player and put them in their team with its existing characteristics.
I’m also working on something called multimodal evaluation, where you take many different types of information—both subjective things like scouting notes and injury history along with objective data— and put it into an AI-powered engine to get individualized development recommendations. That’s the next big, solvable challenge.
Oliver: The consideration of analytics by the front office is important, but I’ve always felt like the success of an organization is also dependent on how it deals with a situation when the analytics don’t agree with the eyes, and you have to have that tough conversation about how to reconcile it. That’s not necessarily a quick exercise, but in a front office, you need to go through that and hopefully get it right before the draft or the trade deadline.