Dr Patrick Lucey, Chief Scientist at Stats Perform: “The value of generative AI is that it essentially unlocks the full value of the sports data we collect”

You may remember the scene in When Harry Met Sally where a fellow diner, on seeing Meg Ryan’s ‘enthusiasm’, said “I’ll have what she’s having”. Having heard from Dr Patrick Lucey, Chief Scientist at Stats Perform, we can’t help feel the same. This is someone who loves their job so much that we suspect Stats Perform could halve his salary and he’d keep on working.

It helps that Patrick’s work, unlocking value from sports data, sits at that sweet spot where timing and effect meet. While Patrick has worked in the AI field for over 20 years, the rise of AI combined with the staggering amount of high-quality sports data now available means that Stats Perform can deliver insights to everyone from sports broadcasters to team managers.

Bearing in mind that Stats Perform claims that it’s “world leader in sports AI”, you may wonder why you haven’t heard of them. If you haven’t, you almost certainly have heard of Opta sports data, the power behind everything from betting platforms to video games.

Here, Patrick shares what excites him in the sports tech field (a lot!), how Opta is using AI and machine learning to deliver even more to Stats Perform’s customers, and the impact that generative AI will continue to have on the sports tech sector. And some of his enthusiasm might just rub off on you too!

Related reading: Let the games begin: Paris Olympics put AI to the test

Tell us your elevator pitch

Stats Perform provides trusted, high-quality real-time sports data, content and AI technology to professional sports clubs, federations, media, broadcasters, apps, brands and bookmakers that help them grow their audiences and customers, and win trophies. Our multi-award-winning Opta sports data is known as the Ground Truth of sports by fans all over the world, who rely on it when they check scores, read match reports or previews, place a bet, watch broadcasts, listen to analysis, pick their fantasy team – even when they play video games like EA Sports’ latest blockbuster title FC24.

Coaches, analysts and scouting departments at the planet’s biggest professional sports clubs rely on it to give them tactical insights into their own players and opposition teams, and to make recruitment decisions. We cover over 20 sports, thousands of leagues and competitions, collect our own data via a highly trained and experienced team of humans, and enrich it with AI.

We are also pioneers and leaders in low-latency live sports video streaming. We partner with many of the world’s leading sports properties. We’ve been doing all of this for a long time, meaning our sports media engine is extensively deep, consistent and broad. That also means we have a unique ability to create proprietary generative AI models, which has been my area of focus for most of the past nine years.

What is it about sports tech that excites you? What made you get into this sector?

First of all, I love sport. I love playing it, watching it, analysing it and discussing it. Sport is the ultimate theatre and it is really the only one of the shared, live experiences that unites billions of people across the globe. Through another lens, it really is the universal language of the globe.

Also, sports tech relies on data which in my opinion is the most interesting data that exists in the whole field of AI – as it is multimodal (visual and metric-based), fine-grained, multi-agent, adversarial but also tied to the real-world. It also enables us to answer questions of team and player performance in an objective way.

So the intersection of these two passions of mine has enabled me to live out my dream job. But even though I’m excited about the work I do every day, it comes with great responsibility as the statistics we collect and generate, and the metrics that we create on top, and that we distribute around the planet, are essentially that “universal language (of sport)”.

This language is consumed by billions of people around the globe every year – it powers their fantasy games, match reports, what they hear in commentary, and talk to their friends about. So we take enormous measures to be accurate and reliable.

Our data also enables pro teams and coaches to analyse previous and upcoming matches, and make recruitment decisions – it really is very widely trusted. Producing to such high standards is a responsibility that everyone who works here at Stats Perform embraces, that is why our Opta data brand is the best-known, most trusted and reliable sports data brand in the world, and also why we have an unassailable lead in Sports AI, since only we have access to our deep proprietary data and we’ve collected it to be highly usable to train models. In fact our company has been at the cutting edge of doing just this for over 40+ years.

For me personally, because sports fandom evolves, and technology evolves, so every day is still more exciting, with more possibilities and opportunities than the last. I’m an ex-athlete, and a huge sports fan, so grasped the chance to combine my lifelong passion with my data science profession when it arose back in 2015. Since then I’ve been in the enviable position of leading an AI team with unique access to the world’s biggest and best sports data, information and video engine… for me, it’s a union made in heaven.

Recommended: Paris 2024: The greenest games ever

What sports does your tech apply to? And have you been surprised by its use?

At Stats Perform, we like to think of ourselves as the keeper of the public record of “ALL” sport. Even though we collect data and generate AI metrics and insights across all sports – we go particularly deep in soccer, basketball and tennis with our tracking data solutions.

I wouldn’t say we have been surprised by the importance of our AI-infused data and metrics as this is essentially capturing the story of performance in an objective way – but I would say we have been surprised by how long it took people to realise that this is important across all consumption verticals (not just teams and betting, but for fan engagement) – but I think people are realising the importance now across the sporting ecosystem.

At the heart of it, we believe there’s magic in the detail of elite sport and that if you can unlock it, and make it useable by those who serve fans and athletes, you can make sport mean more to current fans, future fans, and everyone who’s passionate about helping the sports industry thrive.

You can power more colourful, contextual, real-time stories that inform, illuminate and enrich the on-field action, bonding fans closer to more teams and players, growing viewers, spectators, readers, subscribers, users and sponsors. You can uncover new insights and patterns that power sharper predictions and smarter decisions by coaches, scouts and analysts, improving their on-field and off-field performance and helping them win more trophies. You can make sport more relevant, more accessible and more entertaining, to more people, in more countries.

Technology is at the very core of capturing and delivering the magic in detail.

Can you give an example of a complex problem in sports that you – or your company – have been involved in tackling with technology?

As mentioned above, in my view, sports data (particularly tracking data), is the most interesting dataset that exists in the whole field of AI as it is multimodal, fine-grained, multi-agent and adversarial. It’s entirely different to human language which is why models trained to predict the next word in a sentence can’t predict the next thing a player will do. I described our Opta sports data as the Universal Language of Sport – in a way, the deep data we’ve generated through AI technologies could be called the “hidden language of sport”. It is very complex and requires sophisticated methods to find.

For example, in soccer – the language used to discuss team tactics centres on formation and the role of each player within that formation. However, players are dynamic and tend to interchange positions fluidly across the game. This causes a permutation problem, meaning effectively that like a ball of yarn, it gets jumbled up making it near impossible to analyse (there are more permutations than atoms in the universe!).

But using machine learning techniques, we have effectively been able to discover the hidden structure of teams in an automatic way – enabling us to measure formation and role in soccer, at scale – which also enables us to do other things like “visual search” (think of circle to search which was recently popularised by Google), “ghosting” (think of a simulation in EAFC), but also generate complete positional data for every player on the pitch remotely, even when they can’t all be “seen”, which enables us to go back in time and scale data collection for every soccer game that has been played, pre the invention of multiple in-stadium cameras, or to supply detailed positional and movement data about players and teams to power analysis that previously couldn’t be done (if you are interested in learning more, we recently did a two-part deep dive on the topic here and here).

Related reading: Ryan Beal, CEO & Co-Founder of SentientSports: “Sports generate some of the richest datasets globally”

What is an example of a club or team improving their performance through technology that you find particularly interesting? And why?

I think a really good example is our work in basketball with a product we developed called AutoStats. We pioneered the collection and use of player tracking data in the NBA back in 2010 with 4 teams and then NBA-wide in 2013 via our SportVU in-venue camera system.

Now tracking data and the associated metrics are the language used to describe team and player performance beyond superficial actions because they include the all-important movement and positions, whether a player has the ball or not. The NBA used our in-venue camera systems to collect this level of detail.

Outside the NBA however, in NCAA/College and international basketball leagues that feed future draft players for NBA teams, the only data available has been high-level play-by-play data, which just gives the outcome of each possession like pass, rebound, or foul.

Given the improvement in computer vision, our deep data archives and advances in machine learning technology, 6 years ago we embarked on the bold task of collecting tracking data not from cameras in the venue but synthesising it from game video. This enabled us to provide detailed analysis of many more aspects and capabilities of a player such as their offensive and defensive ability on things such as pick-and-rolls, isolations, drives, off-ball-screens etc. The power of this technology is that we can effectively scale data collection at a granular level, and effectively “go back in time”. Presently, we have collected this hyper-detailed data for over 12,000 games of NCAA and international basketball games, and teams like the Orlando Magic are users of this information to help inform their draft decisions (see Athletic article here). Last season they made a selection described as ‘controversial’ in the media, who then went on to be Rookie of the Year.

As I said, there is magic in the detail! We are now doing the same in soccer, via our Opta Vision product, which is orders of magnitude larger in terms of its users and its applications, which we are very excited about.

What do you believe is the biggest opportunity in sports tech right now?

Like every sector, it is definitely generative AI. The value of generative AI is that it essentially unlocks the full value of the sports data we collect. However, this isn’t a case of just asking a sports question into ChatGPT or Gemini as those models are trained in natural language.

As mentioned above, sport has its own language (one that we have essentially created) that consists of the event (action) and tracking (positional and movement) data we have collected, which makes it multimodal. Effectively Generative AI or the Large Language Models (ie, LLMs) that underpin them enable us to provide more precise predictions and recommendations about players to recruit, tactics and formations to adopt, create player tracking data, or quickly answer questions from our deep, dynamic database that couldn’t be done at such depth and scale before. It means our data will be more useful, and more personalisable, and that our users and their users (fans) will be more entertained than ever before.

That’s the goal we always work towards, as that’s what keeps the sports industry thriving.