Advanced Hockey Analytics Uncovers Hidden Player Value

The subtle art of hockey analytics reveals that true competitive advantage often lies not in the obvious, but in the delayed, the difficult, and the precisely measured. This conversation with Tyrel Stokes, a statistician with a PhD in causal inference and a Senior Manager of Hockey Analytics at Teamworks, peels back the layers of professional hockey to expose how advanced data is moving beyond simple metrics to uncover deeper truths about player performance and team strategy. Those who can embrace these less intuitive insights--understanding that immediate discomfort can pave the way for long-term gains--will find themselves outmaneuvering competitors who remain fixated on surface-level observations. This analysis is essential for team executives, coaches, and analysts looking to build a sustainable edge in a sport increasingly shaped by data.

The Hidden Cost of "Winning" the Moment

The NHL playoffs, a crucible of high-stakes competition, often present a paradox: a team can play exceptionally well and still lose. Tyrel Stokes highlights this inherent randomness, noting that "you can be really good and and you might not win... maybe 20 of the time the best team is going to win or something like that." This observation cuts to the core of why simple win-loss records or even basic shot metrics can be deceptive. The immediate gratification of a win, or the appearance of dominance in a game, can mask underlying inefficiencies or vulnerabilities that will be exploited over a series or a season.

Stokes’ work with Teamworks focuses on developing sophisticated analytical tools, moving beyond traditional statistics. He describes their "on puck value system," which assigns value to key events like takeaways, passes, carries, and shot recoveries. This granular approach allows for a more nuanced understanding of player contributions. Instead of just looking at who scored, analysts can see how value was generated and where it was lost. This is crucial because many conventional strategies, focused on immediate outcomes like scoring chances or possession, fail to account for the downstream consequences. For instance, a team might aggressively pursue a loose puck, winning possession but leaving themselves out of defensive position, creating a more dangerous scoring opportunity for the opponent. This is a classic example of a first-order win leading to a second-order loss.

The challenge, as Stokes points out, is translating these advanced metrics into actionable insights that resonate with traditional hockey minds. He notes the difficulty in "getting a quick snapshot of like which players and which teams are dominating how." This is where the deeper systems thinking comes into play. It’s not just about identifying who is good now, but understanding why they are good and how that translates into sustained success. The Carolina Hurricanes, mentioned as an example of a team performing exceptionally well, embody this. Their strong playoff start, while impressive, is built on a foundation of analytical rigor that likely influences their strategic decisions beyond just the immediate game.

Furthermore, the conversation touches on the evolving data landscape. The advent of player tracking data, providing insights into player orientation, speed, and even stick location, promises to unlock even deeper layers of analysis. Stokes expresses excitement about understanding "who's good at skating backwards" or "where the sticks are," details that are currently inferred but will soon be directly measurable. This shift signifies a move from analyzing what happened to understanding how and why it happened, allowing for more precise tactical adjustments and player evaluations. The teams that can effectively integrate this new data and move beyond the "obvious" metrics will gain a significant advantage.

"The NHL playoffs is really one of the the cruelest of like the the playoff blow offs like series across sports like just being it's you can be really good and and you might not win."

-- Tyrel Stokes

This inherent randomness means that teams cannot rely solely on short-term performance metrics. They must build systems and evaluate players based on a broader understanding of how individual actions contribute to overall team success, even if those contributions are not immediately obvious or directly tied to scoring. The danger lies in optimizing for the present moment, a strategy that often leads to predictable, and ultimately less effective, outcomes over time. The true competitive advantage is built by embracing the complexity and looking for the patterns that emerge from sustained, high-quality play, not just isolated moments of success.

The Data Engineering Bottleneck: Building the Foundation

A critical theme that emerges is the foundational role of data engineering in unlocking the potential of advanced analytics. Stokes emphasizes that Teamworks aims to "take care of as much of the data and engineering as possible" to make data accessible. This highlights a common bottleneck: the sheer volume and complexity of sports data, particularly with player tracking, require robust engineering pipelines before data scientists can even begin their work.

Stokes describes the incoming hockey data as "36 join angles... a hundred times a second," noting its massive size. He states, "if you really want to be able to use that properly -- you need biomechanists -- you need you need data engineers." This underscores that the ability to analyze complex datasets isn't just about having skilled statisticians or data scientists; it requires a solid infrastructure to collect, clean, process, and store this information. Without this foundation, even the most brilliant analytical minds would be unable to extract meaningful insights.

The implication for teams is clear: investing in data engineering is not merely a technical necessity but a strategic imperative. It's the prerequisite for any advanced analytical work. Teams that underestimate this or try to bypass it will find their data science efforts hampered, unable to leverage the full power of their data. The ability to "map, clean, re-representing" data is essential for building accurate models, whether they are expected goals models or more complex systems designed to infer player movement.

"if you want to be at the bleeding edge of statistics at this point like we're about to get hockey hockey data is you know it's it's something like 36 like join angles you know like a hundred times a second... and if you really want to be able to use that properly -- you need biomechanists -- you need you need data engineers."

-- Tyrel Stokes

This also explains why Teamworks provides "foundational models" and "lower level outputs" to their clients. They recognize that different teams have varying levels of internal expertise and infrastructure. By handling the heavy lifting of data engineering and providing robust models, they empower teams to focus on the higher-level analysis and tactical applications, rather than getting bogged down in the complexities of data infrastructure. This approach acknowledges that the "middle" tier of teams, those with some analytical sophistication but not massive dedicated engineering teams, can benefit immensely from a platform that bridges this gap.

The Unseen Value: Off-Puck Play and Loose Pucks

Stokes reveals that a significant portion of a player's value, perhaps as much as half, comes from actions "with and near the puck." This challenges the common viewer's focus on more visible actions like shots or goals. He elaborates on two specific areas where unseen value resides: defensive positioning and loose puck recovery.

For defensive positioning, Stokes highlights players like Jacob Slaven and Adam Pelech. He describes how these defenders excel at "reducing pass threat" and are "really incredible" in their positioning, often appearing out of place to the untrained eye but making crucial plays. This type of defensive acumen is difficult to quantify through traditional stats but is vital for preventing scoring chances. It’s a testament to how systems thinking, understanding how defensive structure prevents opportunities, is more impactful than simply looking at blocked shots or hits.

The recovery of loose pucks is another area Stokes identifies as "extremely predictive" and surprisingly high in value. He notes that players who are adept at this, like Yanni Gourde (though he names him as Celli Brunini in the transcript), consistently add value regardless of their team's success or context. This skill, seemingly simple, involves a combination of anticipation, agility, and stick work that allows players to gain and maintain possession in chaotic situations.

"you being the person that is like recovering loose pucks like -- is extremely predictive -- it's one of those things that like one of the things that i always try to look at when i'm thinking about how predictive is the statistic i really care about how does this translate across teams and how does this translate if i was to take those players that get like swapped or they're in completely new context like how predictive is it and something like loose puck recovery like if you're on a good team if you're on a bad team if you're in ncaa like whatever you're in the nhl like the guys that get a lot of value out of that like it's extremely predictive."

-- Tyrel Stokes

These insights demonstrate the power of looking beyond the obvious actions. By understanding that "off puck behavior" and the ability to capitalize on "loose pucks" contribute substantially to a team's success, analysts and coaches can identify and develop players who excel in these often-overlooked areas. This is where delayed payoffs come into play; a player who is excellent at defensive positioning might not light up the score sheet, but their contributions prevent opponents from scoring, a critical factor in winning close games and series.

Key Action Items

• Develop a "second-order thinking" framework: When evaluating any strategic decision, actively map out the potential downstream consequences, both positive and negative, beyond the immediate impact.
• Invest in data engineering infrastructure: Prioritize building robust systems for data collection, cleaning, and processing, as this is the essential foundation for any advanced analytics.
• Focus on player evaluation beyond scoring: Identify and value players who excel in areas like defensive positioning, puck recovery, and effective passing, even if they don't have high goal totals.
• Embrace the "pain now, gain later" principle: Seek out strategies and player development areas that require upfront effort or short-term discomfort but offer significant long-term competitive advantages.
• Integrate player tracking data with existing models: As new data becomes available, proactively explore how it can refine existing metrics and uncover new insights into player mechanics and tactical execution.
• Train coaches and staff on interpreting advanced metrics: Bridge the gap between traditional hockey language and analytical insights by providing context and education on how advanced data translates to on-ice performance.
• Conduct "What If" scenario planning: Regularly simulate how opponents might react to your strategies and how your own team might adapt, fostering a proactive, systems-level approach to competition.