Sports will hit differently in the next five years. AI, which has long been an abstract talking point, is becoming increasingly tangible. While foundational models and advanced analytics aren’t new (companies like Zelus started in 2019 and forward-looking teams had been developing them for years prior), the recent surge in powerful, user-friendly AI tools, coupled with high-quality user experiences enabled by LLMs, will continue to bring these capabilities forward.
This AI transformation isn’t just a minor tweak; it’s a fundamental reshaping of how things are done across a wide spectrum of sports operations. From highly personalized player development and streamlined back-office workflows, to more sophisticated front-office decision-making and even direct player interactions with AI-powered ‘Coaches.’ We’ll explore each of these scenarios in future posts. To truly grasp the current shift’s significance, it’s helpful to momentarily look at the foundational component: the journey of data and analytics in sports.
We’ve seen the discipline evolve through distinct phases: from basic box scores (Wave 1), to richer play-by-play data and advanced stats (Wave 2), and into the era of player tracking and sophisticated predictive modeling (Wave 3). These waves brought valuable, often incremental, improvements based largely on analyzing structured data.
The Fourth Wave
We’re now entering what I call the Fourth Wave. This represents a more fundamental shift, driven by AI that can:
- Understand unstructured and multimodal data with remarkable depth (think video, audio, text scouting reports, sensor data, all integrated).
- Generate novel content and solutions (like personalized drills immediately after a game).
- Simulate highly complex scenarios to explore “what-ifs” and anticipate future outcomes.
- Interact naturally through conversational interfaces, making complex analysis more accessible (“Should I trade Luka for AD or Durant?”).
This Fourth Wave isn’t just about doing things faster; it’s about changing how things are done and unlocking entirely new capabilities for teams and athletes at all levels.
What’s Truly Changing?
Let’s unpack what these capabilities will enable.
Unstructured & Multimodal Data
To set the stage on the data side, while we’ve had increasingly interesting Computer Vision (CV) capabilities over the past several years, the coming years will demonstrate a use of advanced computer vision to identify not just player positions, but complex tactical patterns, subtle biomechanical details (even from standard video sources, increasing accessibility), and evaluation of the quality of player decisions within the game’s context. We’ll also employ Natural Language Processing (NLP) to extract nuanced insights from text, including scouting reports, medical notes, and news articles.
The real power comes from multimodal AI (which can understand and combine text, images, audio, and video) integrating these streams: correlating a visual cue (like a slight change in running form) with sensor data (increased heart rate) and a player’s reported fatigue level to provide a much richer understanding of performance or injury risk. This integrated approach allows for insights that were previously impossible or required immense effort.
When we sold Zelus to Teamworks, this stream integration was one of the reasons I was most excited by the deal. It will enable teams to connect the off-field preparation data they had in Teamworks (nutrition, sleep, etc.) with the on-field performance data that we had at Zelus. This will be the first time that the preparation-performance feedback loop will be closed at scale, and it’ll have a powerful impact on future performance.
Generating Novel Content & Solutions
Picture a high school baseball coach working with a promising pitcher. The coach notices the pitcher’s velocity is slightly down, and something seems off with their mechanics, but it’s hard to pinpoint precisely during live practice. Using just phone footage from a pitching session, an AI tool analyzes the pitcher’s delivery frame-by-frame. It identifies a subtle inefficiency – maybe a slight timing issue in their hip rotation or a minor change in arm angle compared to their best sessions. Instead of just flagging it, the AI generates a side-by-side video comparing the suboptimal motions to an optimal model (perhaps derived from the pitcher’s own past data).
Then, it suggests specific corrective drills tailored to that exact issue and might even create short animations demonstrating the proper form for the player to review on their phone later. This kind of personalized, actionable biomechanical feedback becomes scalable and accessible in a way that simply wasn’t feasible before outside of elite, specialized training centers.
Simulating Complex Scenarios
This capability allows teams to explore complex possibilities without real-world consequences.
Counterfactuals (Useful “Monday Morning QBing”)
We’ve all done it—watching a replay and saying, “What if the quarterback had made that pass to Amon-Ra St. Brown instead of checking it down to Sam LaPorta?” In the coming years, we’ll take this beyond casual debate. Using precise player tracking data, AI can reconstruct that exact game moment. Where every player was, their speed, the ball’s location. Then, models will be able to simulate the alternative decision: the QB throws to St. Brown.
Sophisticated AI behavior models predict how the defenders and other receivers would likely react to this change. By running this simulation many times, the AI doesn’t just guess; it models the probabilistic outcomes. Maybe there was a 70% chance of completing the pass to St. Brown for 25 yards, a 10% chance of an interception, and a 20% chance of an incompletion. It turns speculative “what-ifs” into data-informed explorations of alternative realities, providing genuine insights long after the whistle blows.
Novel Tactical Solutions
How does Steve Kerr come up with a set play against a specific defense at a particular moment in a game? He’ll draw on his experience, watch film, adapt to previous plays in the current game, and sketch something out. With AI, highly advanced techniques like Multi-Agent Reinforcement Learning (MARL) can create a team of “players” in a highly realistic simulation of the game.
Instead of honing in on a specific play, you provide a goal (“advance the ball past half court under full-court pressure with 8 seconds left in the game”). The AI players will run the scenarios millions of times, learning entirely through trial and error. It isn’t constrained by existing playbooks or what’s ‘normally’ done; it’s purely driven by what proves effective in the simulation. It tests countless sequences of movements and decisions, learning from the raw feedback of what leads to achieving the goal and what doesn’t. The AI can uncover genuinely novel and sometimes non-intuitive strategies this way.
Natural Interaction
Conversational interfaces, often powered by LLMs, make these advanced capabilities much more accessible. For instance, a high school coach could ask, “Show me drills to improve our team’s defensive rebounding based on last week’s stats and video.”
The LLM interprets this, then triggers specialized models. Perhaps a computer vision model to analyze rebounding positioning in the video, a statistical model to analyze rebounding numbers, and potentially a generative model to synthesize these findings and suggest appropriate drills.
Similarly, an athlete asking, “What nutritional adjustments should I consider given my training load this week?” would have their query understood by the LLM, which then likely queries a system integrating their specific physiological data (from wearables, perhaps analyzed by another model) and nutritional databases to generate a tailored recommendation. The LLM makes the interaction seamless, but the deep analysis can often rely on these other, specialized AI components working behind the scenes.
AI offers the potential to create more intelligent, adaptive, and efficient sports organizations at all levels. By integrating diverse data streams, generating personalized insights, simulating complex futures, and enabling natural interaction, AI can unlock new levels of performance, development, and engagement.
This won’t be without challenges. It could be league resistance, player adoption, or technical hurdles. But the promise of where this is all headed is what I’m excited to explore here going forward.
Sports will hit differently in the next five years. AI, which has long been an abstract talking point, is becoming increasingly tangible. While foundational models and advanced analytics aren’t new (companies like Zelus started in 2019 and forward-looking teams had been developing them for years prior), the recent surge in powerful, user-friendly AI tools, coupled with high-quality user experiences enabled by LLMs, will continue to bring these capabilities forward.
This AI transformation isn’t just a minor tweak; it’s a fundamental reshaping of how things are done across a wide spectrum of sports operations. From highly personalized player development and streamlined back-office workflows, to more sophisticated front-office decision-making and even direct player interactions with AI-powered ‘Coaches.’ We’ll explore each of these scenarios in future posts. To truly grasp the current shift’s significance, it’s helpful to momentarily look at the foundational component: the journey of data and analytics in sports.
We’ve seen the discipline evolve through distinct phases: from basic box scores (Wave 1), to richer play-by-play data and advanced stats (Wave 2), and into the era of player tracking and sophisticated predictive modeling (Wave 3). These waves brought valuable, often incremental, improvements based largely on analyzing structured data.
The Fourth Wave
We’re now entering what I call the Fourth Wave. This represents a more fundamental shift, driven by AI that can:
- Understand unstructured and multimodal data with remarkable depth (think video, audio, text scouting reports, sensor data, all integrated).
- Generate novel content and solutions (like personalized drills immediately after a game).
- Simulate highly complex scenarios to explore “what-ifs” and anticipate future outcomes.
- Interact naturally through conversational interfaces, making complex analysis more accessible (“Should I trade Luka for AD or Durant?”).
This Fourth Wave isn’t just about doing things faster; it’s about changing how things are done and unlocking entirely new capabilities for teams and athletes at all levels.
What’s Truly Changing?
Let’s unpack what these capabilities will enable.
Unstructured & Multimodal Data
To set the stage on the data side, while we’ve had increasingly interesting Computer Vision (CV) capabilities over the past several years, the coming years will demonstrate a use of advanced computer vision to identify not just player positions, but complex tactical patterns, subtle biomechanical details (even from standard video sources, increasing accessibility), and evaluation of the quality of player decisions within the game’s context. We’ll also employ Natural Language Processing (NLP) to extract nuanced insights from text, including scouting reports, medical notes, and news articles.
The real power comes from multimodal AI (which can understand and combine text, images, audio, and video) integrating these streams: correlating a visual cue (like a slight change in running form) with sensor data (increased heart rate) and a player’s reported fatigue level to provide a much richer understanding of performance or injury risk. This integrated approach allows for insights that were previously impossible or required immense effort.
When we sold Zelus to Teamworks, this stream integration was one of the reasons I was most excited by the deal. It will enable teams to connect the off-field preparation data they had in Teamworks (nutrition, sleep, etc.) with the on-field performance data that we had at Zelus. This will be the first time that the preparation-performance feedback loop will be closed at scale, and it’ll have a powerful impact on future performance.
Generating Novel Content & Solutions
Picture a high school baseball coach working with a promising pitcher. The coach notices the pitcher’s velocity is slightly down, and something seems off with their mechanics, but it’s hard to pinpoint precisely during live practice. Using just phone footage from a pitching session, an AI tool analyzes the pitcher’s delivery frame-by-frame. It identifies a subtle inefficiency – maybe a slight timing issue in their hip rotation or a minor change in arm angle compared to their best sessions. Instead of just flagging it, the AI generates a side-by-side video comparing the suboptimal motions to an optimal model (perhaps derived from the pitcher’s own past data).
Then, it suggests specific corrective drills tailored to that exact issue and might even create short animations demonstrating the proper form for the player to review on their phone later. This kind of personalized, actionable biomechanical feedback becomes scalable and accessible in a way that simply wasn’t feasible before outside of elite, specialized training centers.
Simulating Complex Scenarios
This capability allows teams to explore complex possibilities without real-world consequences.
Counterfactuals (Useful “Monday Morning QBing”)
We’ve all done it—watching a replay and saying, “What if the quarterback had made that pass to Amon-Ra St. Brown instead of checking it down to Sam LaPorta?” In the coming years, we’ll take this beyond casual debate. Using precise player tracking data, AI can reconstruct that exact game moment. Where every player was, their speed, the ball’s location. Then, models will be able to simulate the alternative decision: the QB throws to St. Brown.
Sophisticated AI behavior models predict how the defenders and other receivers would likely react to this change. By running this simulation many times, the AI doesn’t just guess; it models the probabilistic outcomes. Maybe there was a 70% chance of completing the pass to St. Brown for 25 yards, a 10% chance of an interception, and a 20% chance of an incompletion. It turns speculative “what-ifs” into data-informed explorations of alternative realities, providing genuine insights long after the whistle blows.
Novel Tactical Solutions
How does Steve Kerr come up with a set play against a specific defense at a particular moment in a game? He’ll draw on his experience, watch film, adapt to previous plays in the current game, and sketch something out. With AI, highly advanced techniques like Multi-Agent Reinforcement Learning (MARL) can create a team of “players” in a highly realistic simulation of the game.
Instead of honing in on a specific play, you provide a goal (“advance the ball past half court under full-court pressure with 8 seconds left in the game”). The AI players will run the scenarios millions of times, learning entirely through trial and error. It isn’t constrained by existing playbooks or what’s ‘normally’ done; it’s purely driven by what proves effective in the simulation. It tests countless sequences of movements and decisions, learning from the raw feedback of what leads to achieving the goal and what doesn’t. The AI can uncover genuinely novel and sometimes non-intuitive strategies this way.
Natural Interaction
Conversational interfaces, often powered by LLMs, make these advanced capabilities much more accessible. For instance, a high school coach could ask, “Show me drills to improve our team’s defensive rebounding based on last week’s stats and video.”
The LLM interprets this, then triggers specialized models. Perhaps a computer vision model to analyze rebounding positioning in the video, a statistical model to analyze rebounding numbers, and potentially a generative model to synthesize these findings and suggest appropriate drills.
Similarly, an athlete asking, “What nutritional adjustments should I consider given my training load this week?” would have their query understood by the LLM, which then likely queries a system integrating their specific physiological data (from wearables, perhaps analyzed by another model) and nutritional databases to generate a tailored recommendation. The LLM makes the interaction seamless, but the deep analysis can often rely on these other, specialized AI components working behind the scenes.
AI offers the potential to create more intelligent, adaptive, and efficient sports organizations at all levels. By integrating diverse data streams, generating personalized insights, simulating complex futures, and enabling natural interaction, AI can unlock new levels of performance, development, and engagement.
This won’t be without challenges. It could be league resistance, player adoption, or technical hurdles. But the promise of where this is all headed is what I’m excited to explore here going forward.