Data-driven transfers are football’s new normal

Soon after the Fenway Sports Group (FSG) bought the Boston Red Sox Major League Baseball (MLB) franchise in 2002, John W Henry, FSG’s co-founder, asked Billy Beane to take over as Red Sox general manager.

Beane, general manager of the Oakland Athletics, had revolutionised player recruitment in the MLB using performance statistics to find quality players who’d been undervalued in the MLB draft. This enabled the ‘As’ to compete with wealthier franchises.

Beane refused the Red Sox job, stayed at Oakland, and has since been promoted to executive vice president. The best-selling book, ‘Moneyball’, which became a Hollywood film, told Beane’s story (Beane was played by Brad Pitt).

In 2010, FSG bought Liverpool Football Club. Two years later they appointed Cambridge University Physics graduate Ian Graham to set up Liverpool’s own data research unit.

Part of Graham’s remit was to use live player tracking data to locate and analyse players who would match the club’s playing needs and budget. Liverpool, like the Oakland As, were up against wealthier rivals – Chelsea and the two Manchester clubs at home. And in Europe the likes of Barcelona, Real Madrid, and Paris St Germain.

Back then, some clubs were using data to find new players. But on the whole, player recruitment was still the domain of scouts – experienced pros with a keen eye for a player, who would watch transfer targets live and on video.

Things didn’t work out immediately at Liverpool. Andy Carroll, Mario Balotelli and Christian Benteke were expensive flops. Borini, Markovic, Sakho and others were brought in without much impact. But in the background, things were starting to change.

In 2015, manager Brendan Rogers, unimpressed with the data-led committee approach to transfers, was replaced by Jürgen Klopp. Klopp had managed Borussia Dortmund, a club with a similar business model. To compete with Bayern Munich, Dortmund had to find cheap talent with the potential to be world class, rather than ready-made stars. Klopp also had a specific, established playing style that he believed in totally and would not waver from.

Klopp’s high-tempo attacking approach had already been successful at Dortmund where he won two German league titles and reached a Champions League Final. All he needed at Liverpool was the right players. That was a challenge that Ian Graham and his team of analysts could really get to work on.

In came Sadio Mané, Mo Salah, attackers with explosive pace and lots of energy. The club recruited Georginio Wijnaldum and Fabinho, solid hard-working midfielders to shield the defence and protect against counter attacks. Virgil van Dijk and goalkeeper Alisson were bought to improve the defence.

Over the last five years, Liverpool has spent £108.2m net (difference between players bought and sold), and has won the Premier League, the Champions League, and reached another Champions League final. During the same time, Manchester United spent £484.88m net without winning the title, while Manchester City have invested £601.98m and won the title four times. In 2020’s summer Covid-affected transfer window, Chelsea spent £222.48m on new players.

At all three of Liverpool’s closest domestic rivals, big name players have come and gone without having much impact. Other signings are still there but have underperformed. Think Kepa, Morata and Bakayoko at Chelsea; Cancelo, Rodri, Stones and Danilo at City; Lukaku, Sanchez, Fred, Maguire and Pogba at Manchester United. Arguably some of these players – Stones, Maguire and Pogba, for example – have recently gone some way towards justifying their transfer fees. However, the list of expensive transfer mistakes is actually a lot longer, particularly at United.

Last year, the Old Trafford outfit bought an attacking midfielder, Donny van de Beek, for £35m, who has only made 19 league appearances. This is mainly because United had already spent £200m on attacking midfielders, including Bruno Fernandes bought only last January for £49m.

“Mistakes are made when it’s assumed that a new player will replicate stats from their previous club,” says Dr Bill Gerrard from the University of Leeds. Gerrard is known for applying the Moneyball model to football. He’s worked with Billy Beane on a system to evaluate football players, and currently acts a consultant analyst for Dutch club AZ Alkmaar, where Beane has a 20 per cent stake. “It’s how a club uses the information that makes the difference,” Gerrard adds. “The savvier clubs will interpret the data themselves, to work out how a potential signing might go for them.”

Until recently, data-driven recruitment has been an approach used by forward-thinking clubs to gain an edge over their rivals. Now, because of the coronavirus pandemic, and with revenues down even at the richest clubs, finding the right player for the lowest price is fast becoming a necessity. According to reports from last October, even Manchester United are now looking for a data scientist.

There are only so many Billy Beanes and Ian Grahams around - but there is data available, lots of it. Cameras that track players during live games have been in every Premier League ground since 2013, first operated by ChryonHego and since 2019-20 by Second Spectrum. These cameras capture data at 25 frames per second and there are similar cameras in other European leagues. But the information gathered is available only to clubs in the same league. Clubs have to look elsewhere for data on players in foreign leagues.

Liverpool sources additional data from Skill Corner, a French analytics company that collects data from HD video feeds of every match in 21 leagues. Data is collected by optical tracking algorithms, the sort used in missile technology, which locate and record the position of every player and the ball, every tenth of a second. The system also calculates the movements of individual players, using machine learning to train the algorithms to recognise players by their hair colour, height, and gait. Clubs get a mass of positional coordinates about every player in every moment of a match, from which they can extract insights.

“A club might have six players they want to look at and ask us to provide data,” says Hugo Bordigoni, Skill Corner’s co-founder. “They might want the fastest player in Netherlands’ league. Or they might just access our road-tracking data, which shows distance covered by players, how long they are moving at high intensity, the number of sprints made, peak speeds reached.”

There are other analytics companies working with an increasing number of football clubs on player recruitment. According to Gerrard, these companies either pull information together or provide analytical insights, or both. “

Tim Mitchell, principal in charge of football analytics at DecTech, where Graham worked before joining Liverpool, explains that clubs use data to make their appraisal of how good a player is objective. “When a scout watches a player, the view is subjective,” Mitchell says. “Humans have limited memory, perceptions and cognition, we can’t watch a match and form a well-balanced objective assessment of how good a player is.”

To avoid expensive mistakes, those in charge of transfers at football clubs, need deeper insights into the capabilities of any potential signing. “Data reconstructs the story of the match,” says Patrick Lucey, chief scientist at sports data and analytics company StatsPerform. “The more granular the data, the better the story, and the better predictions that can be made.”

To establish a general context, analysts consider the style of play favoured in the player’s current team, by opponents and overall in the league. Is a physical approach prevalent or do teams play a more skilful game? Do teams defend deep, press their opponents, prefer possession tactics or a direct approach? Is the potential new player coming from a dominant or weaker team?

Gerrard says: “The volume of work [a player does in a poor defensive team] doesn’t necessarily translate into quality or what the player can do for the new team, in a different system, where the player is not going to be switched on as a defender all the time.”

To understand what a player might actually do for their new team, analysts also require specific detail about the player’s individual actions during matches and, importantly, what is happening around that player at the time they perform each action.

From this, the analyst can extract data about a player which shows how they perform in the type of situations they might encounter at their new club. It was this sort of analysis that enabled Liverpool analysts to see that Andrew Robertson had the potential to become a world-class left back, even though he was part of a Hull City defence that had let in 80 goals on the way to being relegated the previous season.

Lucey explains that it’s also possible for analysts to measure a player’s potential quality for a new team by simulating how the player would perform if playing in the new team’s system. “We can see how the game would evolve around a decision, say if we moved the back four up (to defend further from the goal, as many top teams do),” he says.

Mitchell adds: “To be useful, data needs to associate a value to every action and relate this to a team’s likelihood of scoring and conceding – either a positive or negative value.”

All this is a far cry from the days when track-suited scouts stood in the rain with clipboards and a keen eye, making judgments on players that would inform, if not decide, a club’s transfer activity. Wise old owls like Piet de Visser, now 83, who scouted Neymar, Ruud van Nistelrooy, Kevin de Bruyne and Ronaldo (the Brazilian one) before any of them were famous. Or Geoff Twentyman, Liverpool’s chief scout throughout their period of dominance in the 1970s and ’80s, who helped bring in a procession of unknown players from lower leagues who would go on to become stars. Keegan, Clemence, Rush, Hansen and many more.

The most famous of all the old-school talent spotters died 30 years ago. Peter Taylor was Brian Clough’s assistant manager at Derby County and Nottingham Forest. And it was his eye for a player, as much as Clough’s motivational skills, that helped turn two provincial clubs into league champions and, in Forest’s case, double European Champions.

It was Taylor who scouted the apparently injury-prone and over-the-hill Dave Mackay as the man to get Derby get into the First Division, discovered future England defender Roy McFarland playing in the third division, and helped turn overweight midfielder John Robertson into the best winger in the league. There are many other similar stories and Gerrard thinks that modern clubs, despite their increasing reliance on data, still need their Peter Taylors.

“A coach or scout with a deep understanding of the game automatically contextualises what they see,” Gerrard says. “They see the whole picture and process information both analytically and instinctively. Artificial intelligence relies solely on its programming and can only process the data.”

A good scout can analyse things that data cannot quantify. A player’s ability to read the game, that knack that only some have, of being in the right place at the right time to make a telling contribution.

The personality and attitude of a potential transfer target is something else that cannot be gauged from the data. Footballing legend has it that Taylor once followed Birmingham City’s Kenny Burns to the dog track to see if Burns’ drinking and gambling was as bad as rumoured. It wasn’t. Nottingham Forest signed him, Taylor and Clough converted the striker into a centre back and in their first season together, Forest won the league and Burns was named Football Writers’ Player of the Year.

Talent spotters don’t always get it right, though. A certain Manchester United scout once insisted that Man United’s new Brazilian signing, Anderson, was a better player than Wayne Rooney. Anderson proceeded to score an underwhelming five goals in 105 games before he was released, subsequently drifting off into obscurity. The scout’s name? Martin Ferguson, Sir Alex’s brother.

Gerrard suggests combining data analysis and expert perspective as part of an evidence-based approach to player recruitment. “Liverpool have scouts who watch players,” he says. “Even Billy Beane would watch potential draft picks, he didn’t rely solely on data.”

As another Premier League season gets underway, who will prove to make the best transfer decisions, and will they be based on the data or will they be scouted based on diligent observation? At the time of writing this article, the biggest transfer news was Manchester United’s proposed signing of Jadon Sancho from Borussia Dortmund for a reported fee of £73m [which has now been completed - Editor]; a big investment for a player of only 21, but he has already proved his abilities in the German Bundesliga and has huge promise. The trick for United’s competitors is to use the data to find equivalent talent at a fraction of the cost.