AWS and NHL unveil new NHL EDGE IQ Ice Tilt metric to help measure momentum during the game

Ice Tilt helps crack the code on how action on the ice translates to momentum shifts within the game

While traditional stats like goals scored and save percentage are straightforward to track, more nuanced aspects of a hockey game have been deemed immeasurable – until now. For the first time ever, the National Hockey League (NHL) is working with Amazon Web Services (AWS) to  go under the hood with Ice Tilt, a NHL EDGE IQ stat that provides accurate, real-time visibility into which team has the edge – and by what margin – at any given point during game play. The innovative analytic, which averages each player’s positional data on the ice mapped to game data over the course of a period, is helping transform the fan experience and how teams track and analyze their performance.

“Puck-based metrics like zone time and possession don’t always tell the full story of the play itself. Analyzing an accrued average location of all the players within the NHL EDGE data gives us a better feel for control of the play, the territorial control of the surface, and even the current state of momentum at that time in the game,” said Brant Berglund, Sr. Director, Coaching and GM Applications for the NHL. “We can pinpoint key moments in time during the game where drastic changes occurred to facilitate storytelling around those notable events. Future versions of this will look to tell us why the Ice Tilt changed, and even more interesting if it is about to change.”

How it works

During NHL games, the NHL EDGE Puck and Player Tracking (PPT) system transmits positional data from the Players on the ice and puck at 100Hz. With five skaters and one goaltender on each side plus the puck, 3,600 data points are transmitted per second for each game – 12,960,000 data points per period per game. To calculate Ice Tilt, the NHL EDGE data stream is analyzed by AWS, along with the game events data stream, which provides information like clock start/stop times, Player line changes, and power play durations. Using these streams, Ice Tilt calculates a rolling average in two-minute increments, providing a near real-time sense for the center of the action gravity during the live game.

“Ice Tilt is a quantitative metric that can provide data-informed perspective around game control. Puck possession doesn’t always tell the story; instead, we’re working with the NHL to take it a step further to help paint a picture about where the action is occurring on the ice,” said Andrew Reich, Senior Sports Consultant for AWS Professional Services. “It’s one thing to say that a team is dominating based on a qualitative view, now we have a data point to validate and inform that assessment.”

Often, the stronger team will spend more time on the opponent’s side of the ice, attacking the goal. When all the players and the puck are on one side of the rink, then ice is tilted in favor of the offensive team. As each period progresses, the Ice Tilt value fluctuates based on activity, and that metric is mapped to significant game events to provide context. When a team goes on the power play, the expectation is that the ice will ‘tilt’ in their favor. But if their opponent kills the penalty, what happens to the Ice Tilt going forward – is this a turning point in the game?  In the event of clock stoppage or game delay, the Ice Tilt technology enacts a decay process through which the Ice Tilt value slowly drifts to neutral, since the longer the pause, the more likely the game momentum has been impacted.

Figure 1: Ice Tilt values during the Seattle Kraken at Vegas Golden Knights game on October 10, 2023

Under the hood

To calculate the Ice Tilt value, the NHL EDGE Puck and Player tracking and game data streams are sent from the arena via the NHL’s OASIS Platform data hub, for AWS to analyze.  Amazon Elastic Compute Cloud (Amazon EC2) ingests the data, which is then placed onto Amazon Kinesis Data Streams. Amazon Managed Service for Apache Flink then receives the live data for processing and calculates the Ice Tilt value. These calculated Ice Tilt values are sent back into OASIS for distribution to downstream systems including broadcasters for integration with live broadcast signals, and archived in Amazon Simple Storage Service (Amazon S3) to aid in long term analytics. With the potential for multiple games to run simultaneously across arenas, various orchestrations within the system ensure that it is turned on and off based on game schedule, so that AWS resources are only running when needed and the NHL can optimize costs.

NHL stakeholders will be able to use the Ice Tilt value to better understand how certain events impacted game play, which could help with future strategy and preparation. While Ice Tilt is currently a straight logic-based calculation, as it is used across games, the captured data can be leveraged to build machine learning models for future data analytics and add predictive aspects to the feature. Since data is securely archived on Amazon S3, it can easily be fed to thousands of other AWS services, such as Amazon SageMaker, depending on what features and capabilities the NHL may roll out in the future.

Figure 2: High level architecture of the Ice Tilt system running in AWS, integrated with NHL arena data sources and data consumers. Live, in-game data streams are sent from NHL arenas during up to 16 simultaneous games to the cloud. In the AWS Cloud, Amazon EC2, Kinesis Data Streams, and, Amazon Managed Service for Apache Flink compute Ice Tilt with less than 1 second of latency, and send results downstream to broadcasters and other data consumers to power live, in-game graphics, commentary, and analysis.

The need for speed

The significant engineering achievement Ice Tilt represents is its ability to keep up with the speed of a game. NHL Players skate at speeds of 20-25 miles per hour in a space roughly one third the size of a football field, all while passing and shooting a puck that can travel 100 miles per hour. This incredible speed of action presents a challenge in figuring out ways to visualize game control. Ice Tilt indicates where the puck and Players are spending the most time on the ice and maps that information to in-game events to provide additional context and insight as to what factors might be influencing this territorial advantage.

“One of the most difficult factors in developing Ice Tilt the stat was the real-time calculation. With the NHL EDGE data and game streams andwith simultaneous games in up to 16 NHL arenas on the same night, we’re using up to 20,000 data points per second.” said Leon Li XX. “We’re ingesting multiple real-time big data streams and making all of our calculations, then sending them downstream with less than one second of latency.”

Low latency is also crucial for delivering the real-time metric. For Ice Tilt to provide valuable insights, real-time data streams must be ingested, analyzed, and exported in seconds and at scale. Continuing a tradition of innovation with AWS, the NHL has achieved an Ice Tilt data loop that’s completed in less than three seconds, and the metric can be leveraged for games running simultaneously throughout the NHL’s 16 arenas across North America.

Ongoing evolution

Ice Tilt marks the fourth data metric to be derived from the NHL EDGE IQ powered by AWS collaboration – following Opportunity Analysis, Face-Off Probability, and Shot and Save Analytics. Collaborating with the NHL, AWS Professional Services engineers design and implement systems to create scalable performance data analytics using NHL EDGE and  real-time data streams. NHL fans will be able to see the dynamic storytelling tool in action during 2023-2024 season broadcasts, where the data can be used to drive exciting visual representations.

Learn more about how the NHL leverages AWS services to provide advanced analysis to teams, officials, and media partners and build innovative technology solutions.