Every week of the National Hockey League (NHL) season, fans see TV rankings of the best plays of the week, and every week, fans debate those rankings. Most people agree that a great shot is one that had a low probability of success, and a great save is one that stopped a shot with a high probability of success. But what were those probabilities, really?
A new NHL EDGE IQ metric powered by Amazon Web Services (AWS) lends more fodder to these and other debates and promises new insights across the sport. That metric, Opportunity Analysis, determines how difficult a shot is based on a number of different factors, using a combination of historical and real-time data.
During live games, Opportunity Analysis uses data from the NHL EDGE Puck and Player Tracking system, up to the moment of release on every shot, to measure the factors most critical to the play.
"Opportunity Analysis is the first comprehensive and rigorous analysis that can be used in near real time to understand the shot setup, opportunity, and circumstances around the development of a shot," says Leon Li, AWS principal cloud architect.
The metric could be the genesis of new, more data-driven fan debates — a development the NHL welcomes as it seeks ways to make the game more accessible to fans.
“We're going to be able to use this metric as a tool for fans and broadcasters to help foster understanding and enable them to formulate their own theories,” explained Brant Berglund, NHL senior director of coaching and general manager applications. “It's not about giving people the answer. It's about relying on the accuracy of the data, removing as much of the subjective as possible, and empowering people to assess the data and make their own decisions. We're excited to hear people debate the data — the discussion is the best part.”
Opportunity Analysis assesses the factors that make up a shot, providing an output ranking of high, medium, or low, with "high" being the greatest chance of the shot resulting in a goal. The factors include elements such as the angle of the shooter, proximity to the goal, and how much distance the goalie had to cover to block the attempt.
Opportunity Analysis distills an unprecedented amount of data — dozens of factors, many tracked with sub-second latency — into one comprehensive metric.
“We were able to look at so many factors through the volume of real-time NHL EDGE Puck and Player Tracking data available over the course of the season. That's the comprehensive aspect of it," Li says. "The rigorous aspect of it is us, as data scientists, working with NHL's technology and hockey experts and data engineers to vet the accuracy of the data and generate features that make sense in the context of the game."
Opportunity Analysis is the latest metric to emerge from the in NHL's ongoing effort to develop unique data sources and analytic techniques to help break down the intricacies of the sport. Over the past 15 years, the NHL has implemented the Hockey Information and Tracking System (HITS) as the official scoring and events data platform, and most recently launched NHL EDGE Puck and Player Tracking technology. That system, which is installed in all 32 NHL venues, includes infrared emitters and cameras that track sensors embedded within the puck and the sweaters of every player.
In 2021, NHL and AWS began collaborating to make the most of these data sources. In 2022, Face-Off Probability — the first AI/ML-driven NHL analytic — launched within the NHL EDGE IQ platform, helping determine who is most likely to win a specific face-off based on multiple historic and in-game data points. This built upon the foundation of Shot and Save Analytics, two advanced stats that offer an in-depth look at a team or player's scoring performance and a goalie's save performance, respectively.
The layers of data associated with Opportunity Analysis are a gold mine for fans, broadcasters, and the League alike, according to Berglund. This innovative metric reveals not only the difficulty level of a given shot, but insights such as how fast the puck was traveling, the goalie's height, the shooter's change in angle, and others.
"With this product, we’re going to be able to output massive amounts of data on the play leading up to every shot, curated in very close to real time," Berglund says. "That's even more valuable than the rating in many ways — that we're going to actually output that much, that our talented broadcasters have that at their fingertips to talk about during the game, and that fans will have access via those channels, too."
Opportunity Analysis attempts to answer the common lament — “How good was that scoring chance?!” — with a data-driven approach. Just how tough was the situation generating the shot, from a historical perspective? What, exactly, made the shot a near impossibility, a sure thing, or something in between?
NHL and AWS trained a machine learning model to rate the likelihood that certain combinations of circumstances around a shot would result in a goal.
"We wanted to be open-minded and preserve the possibility that the data could challenge conventional logic about scoring opportunities," Berglund says. "Sometimes it did, sometimes it didn't."
For example, Opportunity Analysis verifies the intuition that, on average, shots closer to the net have a better chance of going in than shots from farther away. But other factors are more subtle. While it's still too early to say why or how much, the data have revealed an association between scoring rates/projected goal rates and where the puck passes the blue line before a shot.
"The beauty of this project is that it's forcing all stakeholders to use data to think about the game in different ways," Berglund says. "And hopefully, consumers will, too."
The beauty of this project is that it's forcing all stakeholders to use data to think about the game in different ways.
AWS's processing power and cloud infrastructure made it possible for the NHL team to approach its data in ways it couldn't before. The security and scalability of AWS SageMaker "allowed the NHL to trust AWS with very valuable, comprehensive data and allowed us to quickly iterate and develop the model," Li explains.
AWS Kinesis made it possible to capture and process live game action, including snapshots of time that occur around a given shot. Kinesis sends the information to the model in SageMaker, which then returns a high, medium, or low rating and the top contributing factors that can be routed to analysts for integration in broadcast analysis.
"That real-time aspect is very important for us," Li says. "So is the scalability, given that the NHL is generating thousands of records per second, and multiple games can be happening in parallel."
Berglund expects that, as the NHL dives further into the key factors of shots’ likelihood of success, other features that could illuminate the sport will emerge. After all, with so many ways to engage beyond the game itself, including second-screen experiences, no one is a casual fan anymore. More access and features will mean more ways for fans — and everyone involved in the sport — to unpack the action and formulate their own theories about what makes a successful player or team.
