NBA Teams Use Data Science to Engineer the Perfect Team [VIDEO]

When a nerd gets pushed into their locker by the captain of the high school basketball team, neither one probably thinks that they’ll be coworkers one day. But nerds have been vital in helping those jocks win NBA championships, and it’s thanks to new technology and data science. “Moneyball” is very real, and it’s transforming both the NBA and manufacturing.

How do NBA teams use new technology and data science?

In 2020, the Phoenix Suns opened a $45 million, 53,000-square-foot practice center that includes 150 cameras, nodes, and sensors to track every piece of data, including ball movement during specific plays and changes in player movement that signal fatigue. NBA teams constantly look for new ways to improve player performance, and this competitive drive has led them to new technology that gathers information like never before. 

Around 2013, several NBA teams started using cameras to track player and ball movement and to gather more intricate data. As they began teaching the computer a few fundamental plays and terms, they discovered something intriguing: they could see which shots were statistically the most valuable and learn which shots gave the best ROI. After this, they dug deeper.

As they taught the computers more and more about basketball, the machines went from knowing the basics, to having professional-level knowledge, to understanding the game better than any single person. The cameras watch every second of every game and use machine learning to understand the relationships and patterns in time, space, and velocity, AKA spatiotemporal pattern recognition. The camera and computers can track every ball and player movement to help teams create better strategies. 

In the beginning, only half of the league used video tracking to gather data. Now video tracking is standard for all 30 NBA teams, and the NBA offers pages and pages of these stats online for free.

Teams have even experimented with devices in player jerseys to further track movement and prevent injury.

Much of today’s data differs from the old data because it measures what used to be qualitative: what makes a play a pick-and-roll or not a pick-and-roll? What makes a good shot or shooter? Teams can answer these questions with certainty rather than just on intuition because they have the data to prove it, and this knowledge has revolutionized offensive and defensive strategies.

How has data science changed the NBA?

Data shows that three-pointers are among the most valuable shots; they carry the highest risk because of the low probability of scoring, but they have the highest reward because they’re worth the most points. In the past, players attempted a lot of mid-range shots just inside the three-point line. Data revealed that these shots give the worst ROI. They’re almost as far from the basket as three-pointers (about the same risk), but they’re worth two-thirds as many points (lower reward).

Teams quickly picked up on this and used it to their advantage.

Over the past 20 years, the number of attempted three-pointers in the NBA has increased steadily, while the number of mid-range shots has decreased.

The red spots in the picture below indicate the most attempted shots during the ’01-02 and ’19-20 seasons (darker shades indicate more attempts): 

(Source: ESPN)

A few decades ago, players were often grouped closely around the basket:

Now they’re spread out around the three-point line:

This new view of the game has changed not only strategies but also recruiting and salaries. Tobias Harris was once a somewhat unforeseen threat in the league, but things changed when teams looked at the data. As a result, he took home a $64 million contract. 

Along with offenses and recruiting, team defenses have adjusted. Teams know that mid-range shots are inefficient, so modern defenses often force shooters to take the mid-range shot (AKA drop coverage).

All of this seems simple enough, but teams don’t just press “Enter” on a computer to make these discoveries. They have offices full of data scientists and statisticians to conduct research and to code, and although the NBA releases public stats, teams like to keep their specific strategies secret. Like anything in pro sports, it’s extremely competitive.

Can NBA Teams Use Data to Engineer Perfect Teams?

The short answer: Not quite.

Even with the competitive advantage of data science, the recipe to winning is still not black and white.

Teams can use cameras and computers to track data and movement all day long, but the computer can’t always see the real-world factors influencing a game. The imperfections of life can still make an impact.

Teams must weigh both shot and shooter quality. They can have a weak shooter who takes a lot of good shots (shots with a high ROI), or a strong shooter who takes a lot of bad shots (low ROI). Strategies change based on each team’s players and which opponent they’re playing.

For example, according to the data, mid-range shots are inefficient, but what if someone is a good mid-range shooter, like Kawhi Leonard?

If the defense uses all of their resources to guard the most efficient spots (the three-point line and just under the basket), then a strong mid-range shooter like Leonard can take advantage. Some of the NBA’s top players — Kawhi Leonard, Kyrie Irving, and, yes, even Stephen Curry — are deadly mid-range shooters:

Mid-range shots are valuable in the metagame, AKA the part of the game that goes beyond the textbook and computers.

Players and coaches don’t compete in a vacuum. They have to make their own decisions based on the real-time actions of their opponents and not just on what they see on paper. Luckily, data now comes in quicker than ever, and in the future, coaches and teams could possibly combine data and real-time conditions to make decisions mid-game.

Can Data and Perfection Travel Outside the Court?

Achievements in data science aren’t limited to the NBA or even sports. Data science and the study of movement have rapidly expanded to all parts of life, including manufacturing.

Just as NBA teams can track player and ball movement, manufacturers and engineers can use high-quality data to track machine movement through digital twins:

Using this high-quality data, engineers can quickly design components to virtually test machine movement and functionality before designs enter the real world. They can also use product data from components to program complex automated tasks (e.g., to program automated arms in factories), a process that falls under Industry 4.0.

Computers in factories can use machine learning to gather movement data (e.g., max speed and range of motion) and to predict when a machine will malfunction. Architects can even use data to predict how the pieces of a structure will function together.

Conclusion: Perfection isn’t everything…In Sports 

Data science can’t predict everything, but NBA teams shouldn’t want it to. The game’s unpredictability sells tickets and advertising spots. If we could accurately predict every piece of a game, what would be the point of watching? Perhaps we should keep this road to perfection out of sports and instead aim for other places like our cars and buildings.