Draymond Green of the Golden State Warriors defends a three-point attempt in Game 1 of the 2018 NBA Finals. (Photo by Lachlan Cunningham/Getty Images)
Soon after Second Spectrum began providing a few NBA teams with positional data gleaned from optical tracking cameras, the Golden State Warriors realized they needed help making sense of the wealth of new information. During a two-week period of the 2012-13 season, the team noticed a disturbing uptick in how many three-pointers opponents were shooting and sinking. Golden State didn’t know why that was happening, or how to address that.
The Warriors were in their first season working with Palo Alto-based Mocap Analytics, an AI data company that has since been acquired by Sportradar. The team turned to Mocap’s algorithms, which scoured through the tracking metrics and discovered the slight openings the players were leaving in defense.
“We were able to pinpoint the actions—and [Mocap] helped us do it—and understand the actions that led to that,” said Golden State assistant general manager Kirk Lacob. “And then go back with players and coaches, look at film of those specific actions and understand that we were actually slightly off on this rotation.”
This was an early, public example of how Big Data was, and is, making inroads into basketball. Last month, panelists at SportTechie NEXT discussed the application of AI to the analysis of performance data, revealing the detailed ways that pro and college hoops are collecting this new information and starting to unlock insights.
Mocap Analytics cofounder Eldar Akhmetgaliyev, now a senior data scientist at Sportradar, said the proliferation of sensor-based and optical tracking sources mean the generation of data is no longer a challenge. However, the task of extracting meaning from that data remains significant. Whether assisting a team like the Warriors—or building newer fan-facing products—Akhmetgaliyev’s team is trying to distill the novel data sets it has access to into digestible insights.
“Our vision was always not to try and do the most complex math or use as much neural networks or machine learning as possible—the focus was always on the stories and trying to tell interesting stories,” he said. “Obviously this data has an enormous amount of information. Our job was to try to use machine learning to uncover that data and tell a story that fans never heard.”
The very collection of this data used be inefficient, at best, and at times inaccurate. ShotTracker co-founder and COO Davyeon Ross described the traditional stat-keeping process. Team managers used to sit underneath each basket and manually note down stats, or someone would record each practice on video, and the coaches would then tally the numbers later. Sensor-based systems like ShotTracker do all of that automatically.
Ross relayed a story in which Kansas coach Bill Self speculated that ball reversals and paint touches had an impact on scoring. ShotTracker reviewed the data and identified the patterns Self hypothesized would be there.
“What we did was start looking at all this data,” Ross said, “and we were able to tell him ‘not just one but two ball reversals [matter]. And if you have a paint touch, these are the things that impact not only your field goal percentage but your points per possession.’”
While Second Spectrum is now the NBA’s official tracking provider and is installed in every league arena, the democratization of its technology to lower leagues is enabling the G League and college hoops to benefit from the new information, too.
“I think the major leaps forward are going to happen on the lower levels thanks to companies like ShotTracker, thanks to just the advancement of this data in this space where there’s now going to be a unification process,” Akhmetgaliyev said. “You can compute the same stat for a division league player or a college player.”
That longitudinal tracking of data will be invaluable to executives like Lacob, whose league doesn’t have the long developmental trajectory of, say, baseball’s minor league system. Warriors guard Klay Thompson endorses ShotTracker, and Lacob said the franchise as a whole is getting more uniform data with the devices.
Also, because basketball is a flow sport with more moving parts and simultaneous activity than baseball, which features wholly discrete actions, its questions and answers are more complex and require more sophisticated analytics tools.
“In basketball, we’re slightly past the part where we can tell you what’s happening and we’re just starting to get to the point where we can tell them why it’s happening,” Lacob said. “I also think, culturally, we’re not quite there yet with teaching players how to understand the data more as a probability than a direct outcome, but we’re slowly getting there.”
Lacob said most of the first two years of the Warriors’ work with Mocap was “really just understanding what we had.” Over time, they began to contextualize the data better and make it more actionable. That’s leading to more helpful inquiries and more refined searches within the trove of information.
“That’s, I think, where we are today in basketball—we’re able to start asking the right questions,” Lacob said. “The good news is we now have the tools to build something that will give us the answers that we need.”
Soon after Second Spectrum began providing a few NBA teams with positional data gleaned from optical tracking cameras, the Golden State Warriors realized they needed help making sense of the wealth of new information. During a two-week period of the 2012-13 season, the team noticed a disturbing uptick in how many three-pointers opponents were shooting and sinking. Golden State didn’t know why that was happening, or how to address that.
