At a meeting last year with staffers from IBM and the U.S. Tennis Association, former world No. 1 player Ivan Lendl recounted an insight he had gleaned from scouting Novak Djokovic against the player Lendl coached, Andy Murray. The Serbian star never hit a forehand down the line before the fourth shot of a rally.
Andre Agassi’s longtime strength coach, Gil Reyes, described intently watching how players stopped and restarted their lateral movement during a match. Did they stay low and push off or did they need an extra step to reverse momentum? That extra effort seemed to be a sign of fatigue.
IBM’s technology program manager, John Kent, and the USTA’s general manager of player development, Martin Blackman, shared those recollections because they spurred a new partnership announced this week. Leaving that meeting, Kent realized that IBM Watson could be trained to recognize those patterns and observations. The collaboration to aid the USTA with player performance analytics began about a year ago but really materialized into actionable intelligence after trials this summer.
The hypotheses that coaches have gleaned over years from observation can now be incorporated into a sophisticated computer algorithm. The same AI technology that began generating highlights at the 2017 U.S. Open also can be used to index video for player development purposes.
Reyes’ intuition has already been converted into a new metric called “redirect steps,” or “red steps” for short. IBM culls positional data from Hawk-Eye’s player tracking system to quantify the rate at which a player changes direction. That information can guide both tactics and training decisions and, in the future, can grow to include workout data from wearables such as the instrumented insoles USTA offers players (basically, force plate orthotics in a sneaker).
“It’s almost a new stat—what are your redirect steps?” said IBM’s Elizabeth O’Brien, the program director for sports and entertainment partnerships. “So it’s like, ‘We’ve found the on-base percentage of tennis!’”
O’Brien’s half-serious comment perfectly encapsulates the state of tennis analytics to date. The realization that on-base percentage—the rate at which hitters reach base in baseball—is objectively more important than batting average is a central plot line in the bestselling book, Moneyball. The real story of that 2002 Oakland A’s team was the process of finding undervalued assets in the market (such as players with a high OBP). The A’s triggered an arms race by which every other major league club added or ramped up data-driven analytics operations.
The USTA is now following suit. The IBM partnership is broad and won’t serve only the top pros. At the lowest levels, players aged roughly eight to 11 will now be able to upload video of themselves playing for review by the USTA. This will help the governing body identify the most promising young players. It will also make the sport more accessible, because the traditional methods of getting scouted—competing in youth tournaments—can be expensive.
For older players, up to about age 15, there will be some basic technique instruction based on their video, as well as more involved work with coaches. The hope is that this partnership can enhance the development of American tennis players. The women’s semifinals of the U.S. Open last year consisted exclusively of Americans—Sloane Stephens, Madison Keys, Venus Williams, and Coco Vandeweghe—but that is a rare occurrence.
The pros will receive plenty of tactical support, especially during Grand Slam tournaments. Within an hour of one match concluding, scouting reports for the next round will be distributed. Those briefs will contain both oppositional analysis and self-scouting reports. (They will not be produced, however, if two Americans are set to play each other.)
Blackman said most players want data from three areas: statistics that might be connected with their chance of victory (e.g., winning 71 percent of first-serve points correlates with a 90-percent success rate in matches); benchmarks against the elite (e.g., how their serves compare with those of, say, American No. 1 John Isner); and track record of pressure points (e.g., how successful they are in break points when they stay on the baseline). The players have generally been receptive, Blackman added, although their appetites for data vary considerably.
“Some guys and some women want a lot of information,” Blackman said. “Some want just one thing.”
Younger players tend to want a video playlist of instruction, he added. If a coach emphasizes looking for an inside-out forehand return on a backhand cross-court shot, a series of clips showing that exact sequence is helpful. When O’Brien spoke with a number of players and coaches at the USTA National Campus in Orlando this summer, everyone told her that video was better for memory retention than any series of statistics.
“Any kind of learning or training has to be really visual for tennis players,” O’Brien said. “Video itself is data. Treating data like a video source instead of just like a tape you’re watching. Video is not new, but treating it like a data source is new.”
The idea of using existing knowledge to enhance data analysis is at the heart of what Watson is offering the USTA and, by extension, its players. One coach told O’Brien that it’s important to practice patterns of strokes to create habits and make the sequences intuitive. Learning opponents’ default rhythms can be a useful insight and can be derived from a coach’s observation combined with the processing power of a computer.
“You’d have to have a little bit of a starting point or framework of what to look for, but if you observe a pattern of things and are able to correlate it with something else, then you have your outcome,” Kent said.
Previous work in studying opponents by USTA was “really labor-intensive,” Blackman said, especially the manual tagging of video. Most of the work was outsourced, and only a few players could be supported. Now, all of the top juniors and pros can receive this analytics support. The program also plans to expand to more inputs, such as the wearable tracking and biometric data in the organization’s custom athlete management system.
“Watson is able to look at that data against performance,” Blackman said, “and look for correlations that we can use to modify training.”