Transfer deadline: Using AI to recruit football talent

Historically, to be scouted as a professional footballer boils down to a lot of luck.

Football – or soccer, depending where you kick a ball – is the world’s most popular sport, with over 250 million people playing the game across 200 countries. In England alone, 1.5 million boys play the sport at any one time, and out of them, just 180 get signed.

For that to happen, a scout will typically head out each weekend to analyse a team themselves, and usually, it is a club they or a peer have had success at before.

“It’s subjective by nature,” says Richard Felton-Thomas, COO and director of sports science at AI-scouting platform ai.io.

Finding talent is a big money game, with high rewards for clubs across the football pyramid. According to an EY report, between the 2012/13 and 2020/21 seasons, some £1.6 billion was invested through the Premier League’s Elite Player Performance Plan (EPPP) into Academies at more than 90 clubs in the Premier League and EFL to develop homegrown talent.

This can pay off, too, with player sales often a key part of footballing business. Aston Villa made £100 million – all profit – when it sold academy graduate Jack Grealish to Manchester City in 2001. Grealish had been scouted while playing for Highgate United and joined the Villa Academy as a six-year-old.

But scouting capacity is limited by both time and staffing resource, especially at smaller clubs. This means that limiting the scouting pool to clubs and schools they have confidence in is generally the best course of action for scouts.

“You don’t have that many scouts, so the ability to actually see everybody or give everyone an equal chance was something that wasn’t in the industry,” says Felton-Thomas. “You had to be very, very lucky for a scout to come see you.”

From a player’s perspective, even if they’re performing their best each game, being noticed is largely beyond their control, and clubs also may not be aware of who they’re missing out on.

To tackle this, football clubs are becoming much more dependent on data.

At Web Summit last year, Damien Comolli, director of Toulouse FC admitted: “All the decisions we make on the football side of our operation are driven by data. Whether it’s how we recruit, we sign the player, or the way we play.”

ai.io, for instance, supports the likes of Chelsea and Burnley FC in aiding in player recruitment and analysis by using artificial intelligence.

ai.io’s founder, Darren Peries, launched the platform after his son was released by Tottenham Hotspur at the age of 16. While scouts requested data and videos before coming to watch him, Perries realised the gap in which amateur players didn’t have access to this information to offer, prompting the launch of aiScout.

aiScout is a mobile app where players can use the camera to perform football-related drills, and its AI will analyse the footage, and provide a score benchmarked against the clubs the platform now works with.

“We knew from the very beginning that the computer vision of actually tracking through a mobile phone was going to change the game in grassroots and amateur recruiting,” says Felton-Thomas.

ai.io started to build the tracking technology themselves, and has now acquired Intel’s 3D Athlete Tracking (3DAT) platform, as used in the Tokyo Olympics on professional athletes to further enhance the experience for both amateur and professional football players and clubs.

How to get scouted through AI

“So if you’re trialling for Chelsea through the app, you’ll get scored in relation to a Chelsea player,” explains Felton-Thomas.

The scouts “see all the scores, they see leaderboards, and they can [even] see that who is on top of the leaderboards might be better than the players that are currently inside the academy,” suggests Felton-Thomas.

“It’s just a flag to say, that’s a player you should be watching this weekend.”

With this, in this season alone, it has had 30 of its players go for trials at Chelsea and Burnley.

“Players have come into the app, downloaded, got their data, and that’s been enough of a signal for the club to bring them in,” enthuses Felton-Thomas.

In particular, footballer Jez Davies was released from an academy last October, but, within two weeks, he downloaded the app, uploaded his data, and Burnley had him signed.

“The speed of him leaving one club and finding a new club from getting his data into the system was a great success case.”

In the beginning, when Chelsea was initially trialling the app, it spotted player Ben Greenwood’s athletic data and agreed to take him on for a one-day trial to validate the system.

While he didn’t get signed by them, Felton-Thomas explains that once a player who’s never been scouted before has a trial with a club such as Chelsea, the rest of the clubs know, “and now he’s actually on his second pro contract with Bournemouth.”

Plus, Peries’ son, Reef, who was let go from Tottenham, is now with the Sri Lankan National team, as scouted through the app.

“This year we’ve had an under-9 signed at Burnley, an under-14 signed in the summer at Chelsea, so we constantly are having this trickle of successes by using the platform.”

Toulouse or not to lose?

While AI has allowed the scouts to dramatically widen their recruiting pool, this doesn’t mean clubs are becoming more dependent on it.

As Toulouse FC’s president puts it, the French club wants to go from “to lose” to, “to win” by using its own data analytics to find the best players.

For Comolli, once the players in its system are narrowed down to the best options for the team depending on age, cost, and of course, skill, then the pick is down to the ethics and morals of each pending recruit.

“We want good people, and we want good citizens in the team,” and it may come to choosing someone who has better morals than one who may be a winner, said Comolli.

“How do we identify this with the data?”

As recruiters in enterprises across all industries may be concerned that data and AI may take their jobs, ai.io voices that this challenge of finding a personality fit gives reason that scouts will still have a place on the side lines.

“From the very beginning we’ve seen the value in going back to that subjective side,” Felton-Thomas says. “They’ve been watching football for decades and there are things that AI can’t do.”

For instance, AI can’t see how a player might react to adversity, how they deal with teammates when they’re not winning, and how they speak to the opposition, the coaches, and the referees.

“We’re kind of early talent detection,” explains Felton-Thomas.

Therefore, rather than a scout heading out to a club they don’t know and hoping it isn’t a waste of their time, they can watch a game with the knowledge that one or two players is actually at the top of their leaderboard on aiScout.

“There’s so much subjectivity in a sport where you still need your eyes to see it,” he says.

On the other side, an AI model trained on conventional techniques may also be missing out on a talented player who plays unconventionally.

“There’s always cases where if you speak to most people, there’s features and attributes they want in a football player,” says Felton-Thomas. “But then everyone’s got use cases of where a player didn’t have [those] and they’ve turned out to be one of the greatest players of all time.”

To approach these missing elements, ai.io is coming out with cognitive testing this year, so that it can start to analyse the mental side of the sport.

“So, visual competence, the attention processing, the working memory, we start to bring that because that’s kind of a missing element even to watch the game you don’t really know the cognitive side of that player,” he says. “So that will certainly help start to identify some other talents.”

“At the same time, there is definitely going to be talent who don’t fit the conventional box and we might not have the answer for that today,” says Felton-Thomas.

But, what it hopes to do is by collecting lots of data over time, it will start to see where those players are falling through the net.

For example, during maturation, someone might develop into an unexpectedly exceptional footballer.

“Think of someone like Harry Kane, who was passed up from multiple clubs before he landed at Tottenham,” he says. “There are clubs that passed on him because he was too small, and ended up being six foot three and a huge striker.”

So, by looking at maturation and including the heights and weights of players and their parents, the platform can predict growth spurts and who might not be small tomorrow.

“Those are little things we’re learning as we go, and is going to start to make sure those cases where someone might be missed, won’t get missed, and I think that’s the beauty of AI and machine learning.”

“Hopefully over time we’re just increasing the confidence level of finding the best talent.”