You are a member of the San Antonio Spurs, facing off against the Miami Heat in the NBA Finals. At some moment over what could very well be an intense, seven game series, you will look up at the clock and see only seconds remaining; one more possession to determine who wins and who loses. You hustle back on defense as you see the best player in the world, LeBron James, barreling toward you with the ball.
What direction do you want to force him? Where should you try and make him shoot? If he passes, where is the ball most likely to go? If Dwyane Wade drives from the wing, how fast can he reach the basket? How far away can your teammates be to provide the proper level of help defense behind you? How much, how often, how successful …
The first rule of SportVU is: You do not talk about SportVU.
How can you and your teammates tilt the odds in your favor?
These split-second decisions used to be made with a combination of coaching staff instruction and raw basketball instinct. Today, thanks to a technology called SportVU, the decision-making can be more akin to an algorithm, churning through data to determine probabilities.
As LeBron makes his move, so do you …
Like any good basketball story, this one begins with the Israeli military’s ability to track and shoot down missiles.
Founded in 2005 by Dr. Miky Tamir, SportVU applied the same principles of advanced optical recognition behind missile tracking to the less threatening world of athletics, specifically soccer. By positioning three cameras, each covering a third of the field, SportVU plotted the X/Y coordinates of every player on the field, as well as the ball, producing scads of information from the number of times a specific player had the ball to the player marking him at the time to how fast each moved throughout the game to the velocity of shots on net.
It wasn’t long before SportVU caught the attention of STATS, one of the largest sports data and analytics providers in the world – but not because of an overwhelming interest in tracking MLS games or friendlies between the US National Team and Honduras.
“We looked at it as potentially a game-changer as it relates to sports data collection,” says Brian Kopp, VP of Strategy and Development at STATS, and self-described SportVU evangelist. “We’re always looking to do more in a global sport like soccer, but the reason we decided to acquire it rather than partner or anything else was we knew that it could be expanded to other sports, and therefore we wanted to be the ones building it.”
STATS acquired SportVU in December of 2008, and quickly began adapting the system for basketball.
The nature of the game demanded some tweaks. Three cameras could cover an entire soccer pitch because the space is large and players are, for the most part, spread out. Basketball is almost exactly the opposite. The floor is only 94 feet long (international soccer fields are more than three times that), and for much of the game all 10 players plus the ball are crammed into the first 28 feet above each baseline. Triangulating and calibrating the Prosilica computer vision cameras so each could capture those X/Y coordinates, accurately differentiating between players – plus a Z (vertical) coordinate for the basketball that didn’t exist in the 3-camera soccer setup – eventually required six cameras, with three covering each half of the court.
Because no two arenas are exactly the same, the cameras can’t be hung in identical spots for each team. Some angles are better than others (overhead view on each half of the court, for example, is ideal) and adapting SportVU to each venue requires a little trial and error. Still, Kopp says, hardware was a minor issue, and the system is simple enough for a high schooler to operate – literally. Apparently one NBA team this year had a 16-year-old running it.
Like any good basketball story, this one begins with the Israeli military’s ability to track and shoot down missiles.
SportVU’s algorithms are where the statistical sausage gets made, and switching from soccer to hoops meant changing the ingredients.
“It’s identifying basically five data points. We have a time stamp, a player or a ball ID, and then X and Y coordinates and for the ball, a Z coordinate,” Kopp says. “The algorithm is telling it what the X/Y coordinate is for a specific object. It goes through the process of identifying an object, and then identifying the coordinates for that object, and it does that 25 times per second.” To do its work of tracking player and ball, the algorithms had to distinguish between different jersey colors and lettering, different shades in team’s court flooring, environmental subtleties in each building (lighting, ribbon boards, and so on). As developers and clients happened upon more things they wanted to know, the algorithms, generated by SportVU’s development team in Israel, continued evolving.
SportVU tracked its first NBA game in 2009, and in the fall of 2010 was in the field, working with clients to deliver data (the Spurs were among the early adopters). Initially, delivery of data from any single game took about 24 hours. Thanks to Moore’s Law, that process has (to put it mildly) improved. Using what Kopp calls their “in-game” system – he hesitates saying “real-time” because different people take that to mean different things – data is gathered by the tracking cameras, sent back to STATS offices in Chicago, tailored to the information teams ask to see, and within about a minute can be delivered to a coach on the sidelines via printout or an iPad.
The data might reveal how many times a shooting guard has received the ball in desired spots, how many drives to the basket a defense has allowed, how the opposition might be defending a team’s star player, and endless other data points that can inform a change in strategy. Over time, information can be compiled on everything from a team’s best five-man lineup combinations against a variety of opponent styles to the essential spacing between players on offense and defense to the habits of individual players (where they shoot, how much they dribble, how often and how well they pass, etc.). How often does a player grab a rebound when he has the opportunity? Who really are the most impactful interior defenders in basketball? Teams can measure player fatigue and fitness based on how quickly and how much they move around the court from game to game, allowing trainers to tailor workout and recovery plans for each player and coaches to better monitor playing time. (Or, in a more cynical reading, know which of their players are dogging it.)
There are myriad applications off the floor, as well. SportVU data is one more tool available to organizations deciding which players to pursue or avoid in an offseason trade or free agent signing, or whom to take in the yearly draft. Would a player’s strengths translate surrounded by different teammates? Would weaknesses be better masked? The answers can help avoid critical mistakes in the allocation of increasingly precious payroll dollars, or conversely find the types of bargains fueling success. It’s not just the athletes who get tracked, either. Teams compile data on coaches, and with good cause. Job turnover is rampant in the NBA – 10 of the league’s 30 teams fired their head coach following the end of the 2012-13 regular season – and it’s reasonable to believe a coach’s tendencies in his old job will carry over to his next. How will his style, successes, and failures (he was fired, after all) mesh with a different roster?
It’s an expansive range of possibilities for a technology still not fully integrated into the NBA landscape. Despite millions of X/Y/Z coordinates logged during well over 1,000 games, there are still holes in SportVU’s data, mostly thanks to sample size issues. 15 teams use the system, which means 15 others don’t. While players on non-SportVU teams still appear in the data when visiting a camera-equipped arena, they do so less frequently, and always as a road team. (Players tend to perform less effectively on the road vs. at home.)
“Until we have 30 teams, we still think we’re in the 5-10 percent range of what we could be doing with this,” Matt Bolero, a basketball operations assistant for the Miinnesota Timberwolves, told ESPN last spring. (That Miami isn’t in the SportVU group means less data available for San Antonio to exploit.)
Nonetheless, there are substantial rewards available for those teams best harnessing the potential of powerful analytical tools like SportVU. Dean Oliver is a pioneer in basketball analytics. He has written books, worked in an NBA front office for the Denver Nuggets, and now serves as ESPN’s Director of Production Analytics. “Exactly what that margin is is hard to say. But it’s several games a year,” Oliver says. “You’re talking about five wins, which could be on the low end, five wins a year, you’re talking about millions of dollars in extra revenue every year. So not small.”
Actually, kind of huge; five wins could easily mean the difference between making and missing the playoffs, or between starting the postseason on the road and earning home court advantage. Those extra games are worth millions of dollars a year in revenue, which explains why team executives tend to give SportVU the full Fight Club treatment. The first rule of SportVU is: You do not talk about SportVU. Organizations rarely discuss the specific ways in which they apply SportVU data for fear of undercutting the competitive advantage it might bring.
Oliver has seen the full evolution of analytical tools in basketball and believes the great value of value of SportVU is its granularity. The options available for data mining are practically limitless. There are less sophisticated game tracking options, but they only “show shots, and rebounds, it has locations and it has information on whether there were assists, and there were steals and turnovers and what locations and times in the game and who was in the game. But it has nothing about what the players who didn’t have the ball were really doing in most cases. It didn’t tell you anything about what the defense was doing. It didn’t tell you anything about the spacing, concepts that are very important when you watch the game that clearly feed into the analysis that you can do with other things,” he says.
But to paraphrase Peter Parker’s Uncle Ben, with great granularity comes great responsibility. Information only has value when properly applied, and while more of it means more avenues to innovation and improvement, it also means more ways to make poor choices or overload the circuitry of those intended to benefit.
“Sometimes when someone is drowning,” says one executive whose team is a SportVU client, “you can throw them 20 different floatation devices and they’ll drown trying to reach all of them.”
Even when a team does something demonstrably cool and forward thinking, it doesn’t automatically yield great results. The Toronto Raptors have been among the most aggressive teams utilizing SportVU data, not only installing the cameras but employing people in the front office writing proprietary code to help turn the numbers into something actionable on the floor. The result, as profiled by Grantland.com, are slick video representations not only of what happened during every Raptors possession, but what their analytics team believes each Raptors player should have been doing during those plays based on the individual skill sets of that night’s opponent.
It’s provocative stuff, but in the end shows the limitations of analytical data. The Raptors finished the 2012-13 season at 34-48, with a defense ranked 22nd league wide. Moral of the story: Good ideas don’t automatically translate into success, and more importantly, in the NBA personnel trumps all things. No amount of data papers over an ill fitting or under-talented roster.
The oft-discussed gap between old school and new school isn’t nearly as large as it once was, or as it might still be portrayed. (At this year’s MIT Sloan Sports Analytics Conference, sort of like SXSW for the statistical community, all but one NBA team – the Los Angeles Lakers – sent a representative. Stars like Oklahoma City’s Kevin Durant employ personal analytics consultants. The war is over. Data won.) Still, in a game overloaded with disparate temperaments, skill sets, and locker room culture, implementation of SportVU data is as much a management challenge as one of x’s and o’s. A common coaching cliche is “KYP,” meaning Know Your Personnel. In context, it refers to players understanding the strengths and weaknesses of an opponent.
The same principle applies to those responsible first for choosing which bits of information pack the most potential for successful outcomes, then for making sure the lessons available in the data are delivered and understood.
“It’s knowing how much to give certain people, because of their appetite for it and their ability to assimilate. Some people, they’re feel players. They go out there and they’re going to play the game. Other guys, they want to arm themselves with information. Coaches are the same way, GM’s are the same way. Some people when they pack, they pack everything because they want to be ready for anything. Other people? Give ’em a toothbrush and their underwear and they’ll figure it out. You have to account for everybody in your organization, and what’s important to them,” says Tommy Sheppard, Vice President of Basketball Administration for the Washington Wizards, another SportVU squad.
“You have to get everybody on the same page from the jump and really explain why we are doing these things and what we hope to accomplish with this stuff.”
In 2012-13, the Wizards, among the worst defensive teams of the last half-decade, finished fifth in defensive efficiency (points allowed per 100 possessions), a 16-spot leap. It’s an improvement Sheppard attributes in part to personnel improvements and coaching, but also to adjustments made possible by SportVU data. Mortar helping bind the bricks, so to speak. Just don’t expect him to tell you exactly what those adjustments were. (See the first rule of SportVU.)
Given the tens of millions required to run an NBA team each season, examples like those make the cost of SportVU (around $100,000 per year) feel like a no-brainer. But Oliver understands why so many teams still haven’t adopted the system. Using it well requires some organizational infrastructure backing it up. “It does take an investment in really understanding what it is and how to use it,” he says. Not every franchise is there.
They’d be wise to catch up.
The key going forward is asking the right questions, something Oliver believes NBA teams do far better than a decade or so ago and will continue to do as the role of data in the game evolves. Really, they have no choice, because those franchises lagging behind will cede important territory.
But for those teams successfully integrating SportVU into their basketball operations, finding a consistent competitive edge ultimately becomes becomes an exercise in curiosity and creativity.
“What do you use a hammer for?” Sheppard asks. “Well, if you really stop and start writing it down, you realize that you can use a hammer for about 150 things, not just what it’s ‘used’ for. That’s kind of what I’ve found with this information.”
The 2013 NBA Finals are a testament to high-end talent on the court, along the sidelines, and in the front office. It’s also are a validation of analytical thinking in basketball. The Spurs have lived ahead of the curve for a long time, using available data to help strategy choices on the court and in player personnel. That they were among the first to use the SportVU cameras likely surprised absolutely no one around the NBA.
MIami doesn’t have the cameras, but head coach Erik Spoelstra is known for poring over endless stacks of data looking for any available edge. LeBron is a basketball savant who absorbs information like the near-perfect hoops cyborg he is. Chris Bosh has changed his game based on data, while Shane Battier has extended his career thanks to a profound understanding of analytics.
Both teams bring an intense attention to detail, but only Miami brings the incredibly significant advantage of James, easily the league’s most dominant force. “If you’re going to start the game spotting a team eight to ten points, it makes it very hard,” Oliver says of the four-time MVP. “Which is essentially what he’s giving you.”
But who knows? Maybe San Antonio can squeeze a little knowledge from their SportVU data, one weapon the Heat won’t have. Could it be enough to swing a critical moment in a critical game?
Maybe, maybe not. Either way, the Spurs won’t tell you.
Remember, the first rule of SportVU…