- Sport Business
- Members Log In
The evolution of strategy and tactics in the Formula 1, particularly in the last decade, is often discussed and, as Paul Devlin says, the shift was partially a result of insights provided by data.
“No other sport has been as dynamic in its evolution and embrace of data analytics technology,” he tells the Leaders Performance Institute via Amazon Chime.
Devlin is the Enterprise Account Manager for Sport at Amazon Web Services (AWS), a role he has held since November 2020 when he left his position as Head of High Performance at the Brisbane Broncos. It is a transition he has previously discussed with the Leaders Performance Institute.
The topic of conversation on this occasion is the use of data and machine learning. He adds: “By sourcing historical data, and using it to teach complex machine learning algorithms using AWS, Formula 1 can predict race strategy outcomes with increasing accuracy for teams, cars, and drivers.
“While some of the tech goes to helping drivers, who are hitting speeds as high as 230 MPH, taking pit stops in under two seconds, and flying around corners with a force of 5G, much of it goes to enhance the experiences of its growing base of over a half a billion fans.”
We continue by delving into tracking data before progressing onto computer vision and the implications for the tactics and strategies of what he terms ‘field-based invasion sports’ – football, soccer, Australian rules football, both rugby codes, and Gaelic games are all prime examples – and the implications for coaches and practitioners.
What the data isn’t telling us
Devlin speaks approvingly of the advances made in tracking technology and the implications for physical preparation, but spells out its limitations in technical and tactical terms.
“Sports scientists have realised that tracking technology is a poor predictor of performance. It’s not what it does,” he says. “It only tells you what players did – it does not tell you the story of technical or tactical performance. If you’re looking to use that data to ascertain how well a player performed, you can’t do that without information from the opposition.”
Devlin explains that computer vision technology is beginning to provide meaningful insights in this regard across elite sport. Computer vision technology, as described by AWS on their website, is a machine learning (ML) technology that ‘allows machines to identify people, places, and things in images with accuracy, at or above human levels, with much greater speed and efficiency.’
“Technology is advancing quickly in that space,” adds Devlin. “Knowing where the ball is and the Xs and Ys on the field allows you to create technical and tactical insights. You don’t need a trackable or a wearable on the player, which has a number of positives. It’s less challenging from a data management perspective. It also means that you get the data from both teams. That’s the position that artificial intelligence plays: the ability to detect the ball, the basket, the players, the goal.”
He cites a standard example of how ML may be used. “If we get all the GPS tracking data and we say to a machine learning model ‘here’s a kick-off’. We put ten instances of the formation the players get into prior to a kick-off and we then say to a machine ‘here’s a load of game data, now predict when kick-offs happen in games.’ The machine will use the ten inferences you gave it. Or we could say: ‘here’s where we move the ball against this team when the goals were scored on ten occasions. Here’s every goal they’ve ever conceded, here’s how the ball moved; if the ball moves in this way what’s the prediction of a goal happening?’ It’s the creation of predictions from data.”
Field-based invasion sports
Devlin explains that there are clear implications for field-based invasion sports. “When you look at most field games, essentially the goal of the game is to invade the oppositions’ end to score points. You can’t score points if you don’t invade their end. It’s limited by different rules. If you take rugby league, you get six efforts with the ball, six tackles, to invade the opposition end and then you’ve got to hand the ball back; you kick it before but you’ve still got to hand the ball over.
“In rugby union, you’ve got rucks for long as you can keep the ball without infringing on the rules, then you’ve got to hand it over. Soccer is a bit different as you can kick the ball to gain territory but, fundamentally, you ordinarily retain possession and try to work the ball up the field to create an opportunity to have an attempt on goal. If you don’t hold onto the ball and work it up the field, you can’t do that.”
He found that the determinants of performance were not adequately explained by tracking data. “I used to look deeply into GPS data to predict winning and losing and failed, then I started looking at match event data – passes, kicks, shots – to look at performance and predict winning; and that failed too. Then I thought we should try and look at things completely differently.
“In rugby league specifically, I started looking at using the ball in such a way that you can invade the opposition in your first set with it. Then defend and try to protect your invasion, the territory you’ve invaded, then hopefully when you get it back you’ve made a territorial gain. Then you’ll try to work it up field again so you build enough pressure on your opposition from a field position perspective that they’re fatigued, make errors, and you gain an opportunity to score points. Or they give a penalty away and you kick the points.
“Then I started looking at all the data, metrics and stats. In rugby league, we’d have things like completion rates, percentage possession and missed tackles; all these things that I’ve looked at for 20 years which is not data analytics, it is just counting. People have been convinced that if you miss lots of tackles then you’ve got a poor defence – but this view often doesn’t correlate with points conceded, the epitome of poor defence. Equally, if you have a good completion rate, which means you’ve not made many errors, that often doesn’t correlate with scoring points and winning either.
“I looked at every stat in rugby league to try to see which ones correlated well with winning, and none of them do. I was then looking for a stat that indicated an ability to build pressure on your opposition through invading their territory. The challenge is about sustaining pressure, and there were no metrics I could find that told that story.
“If you take the 10,000-foot view, it’s about the relationship between the Xs on one team, the Ys on the other team, and the ball in relation to the field position. If you look at it generally, that is the predictor of scoring points and therefore winning. ‘How good are you at protecting your territory and invading the other territory to score points?’
“Then, of course, players make mistakes and some players are geniuses. You can make a line break or make an interception and score from your own line, but these are most often anomalies in the data. As a rule, the best teams in all the field-based invasion sports are the teams that are really good at getting from one end of the field to the other and either building pressure there and then scoring points or getting up the field so quickly that the opponent cannot maintain defensive alignment and they score points just because of the speed at which they entered the territory. But it’s all still related with invading the opposition’s territory.
“That’s the path that lots of sports are now looking at, taking a step back and saying: ‘we’ve got the ability through artificial intelligence and machine learning to be much smarter about how we predict what might happen based on tracking players and events’. It’s looking at data and analytics, and the invasion of opposition territory, to decide how you might invade it most effectively to score points. That’s the next world.”
Goal prediction in the Bundesliga with AWS
Some of the world’s best sports organisations are using AWS to build data-driven solutions and reinvent the way sports are watched, played, and managed.
As the football league with the highest stadium attendance globally, Germany’s Bundesliga is among the leaders of the world’s most popular sport.
Bundesliga is bullish on AWS, using ML and analytics to enhance the fan experience and deliver new game and player analytics. Firstly, this has reduced operational cost by 75 percent but just as importantly, with Amazon Personalize has increased fan engagement by 68% percent. Using AWS technology, Bundesliga built new cloud-based services that automate processes, increase operational efficiency, and enhance the viewing experience for the league’s rapidly growing global fan base.
“Through a next-generation statistics platform, using Amazon SageMaker, a fully managed service to build, train, and deploy ML models, Bundesliga offers fans real-time predictions on when a goal is likely to be scored (xGoals), identifies potential goal-scoring opportunities, and highlights how teams are positioning and controlling the field, based on live data streams and historical data from over 10,000 Bundesliga games”
Services such as xGoals and other Bundesliga Match Facts are setting new standards by providing data-driven insights in the world of soccer.
The implication for the coach’s eye
There are concerns in some quarters that if information is fed to coaches and practitioners from ML it might dull their coaching instinct. “There’s also an argument that you can develop a coach’s eye quicker with the use of appropriate data sources,” says Devlin. “It’s using these in the right way and making sure you’re improving your coach’s eye but learning from data and video feedback.”
“Artificial intelligence and machine learning is now allowing us to detect athlete limb positions in real time to predict improved execution of skill and technique changes. This delivers an ability to potentially fast-track coach development but also self-correction of technique through use of technology”
“Coaches from older generations such as me, didn’t have that back in the day. You couldn’t watch an athlete move, make some opinions, and then instantly watch it back on your iPhone to validate your opinion. You just hoped that you got it right without that instant feedback loop, whereas now we’ve got the potential instant digital feedback loop, and that is a remarkable difference. Used appropriately, it should be able to enhance the ability of someone to develop a coach’s eye.”
All teams, coaches and athletes stand to gain from being able to tell the story of success. “I like to say unexplained success doesn’t count,” says Devlin. “You need to know how you won or how you can replicate it. You need to know what impacted your performance from all the different things you can measure and track to be able to replicate it and tell the story.”
Download Performance 23
Our latest Performance journal has landed, with Gareth Southgate, Head Coach of the England men’s football team leading the way with his reflections on defining and developing resilience. Elsewhere we spoke to the Arizona Diamondbacks of Major League Baseball, as well as the world-renowned New Zealand Rugby, and British Wheelchair Basketball, who runs some of the finest programmes in the sport. Edd Vahid of Premier League club Southampton FC also penned a column focusing on talent pathways.