How Math and Data Science Made Liverpool the Best Team on the Planet

It’s been a remarkable season, and here is the untold story at Anfield

Photo by Jack Hunter on Unsplash

If you’ve been following football over the last three years, then it’s difficult to ignore the quiet evolution that has taken place at Liverpool Football Club under Jurgen Klopp’s and Fenway Sports Group’s respective management and ownership. On every level: player recruitment, coaching, in-game management, and commercial contracting, the club has innovated and set new benchmarks for how an elite football club ought to be run. But, here is the thing, if you read between the press lines and try to reverse engineer some of the reasons why Liverpool is punching above their financial weight there is a good tech story.

They have revolutionised how data and models are being used in football.

Here’s is my take as a mathematician (and a data scientist) on what they are doing inside Anfield which is a beautiful lesson in human resources and business management.

Accurate Player Profiling and Recruitment

It’s no secret that Klopp’s teams are intense to play against. They deploy a relentless, aggressive, high tempo, physical style that aims to overwhelm opponents from start to finish. The current (2019/2020) Liverpool team have earned the moniker “Mentality Monsters” for their particular brand of intensity; they’ve also scored more late game, decisive, goals than any other team in Europe by some margin. But how did the identity of this team get built?

There is currently dearth of player statistics available to clubs that are looking to sign new players. And just about everything that a player does on the field can be measured these days. Data is a key part of recruiting players but data has to be applied correctly to build a team identity, where the whole is more than the sum of the parts. To use data for team building, the following ingredients are required:

1. An explicit, quantifiable, sense of how the manager wants the team to play; and a method for realizing this within a specific budget.
2. Models and methods for making player assessments; per position.
3. An understanding of team dynamics.
4. Models for the team’s evolution with each additional player recruited.

The first point (1) is fairly straightforward, it would not make sense to recruit players without having an idea of how the manager wants to set up the team. The difficulty is that this is a tacit, implicit, idea that is never put into a quantitative form. The statement, “I want a team that is intense and unpredictable and that can evolve over a few years,” is hard to quantify, but this is what Ian Graham and Michael Edwards (Liverpool’s recruitment and analytics team) must have done. Here is how to roughly translate it into a model.

• When comparing players of equal skill levels, look for players that have adequate stamina and robustness to fit the playing style: Focus on players that have fast recovery times, good injury track records and can bounce back from the rigors of traveling, training and playing. Look at the rate of games played as well as the volume.
• When recruiting a player and paying a transfer fee plus salary on that player, calculate the discounted value of the player by working backward from the potential transfer fee received, commercial value, and on pitch value of the player.

Of course, there are nuances, but the first basic theme is to deal with downtime and the second is to be maximal yet within budget.

There are also specific criteria for translating intensity into positions on the field: Liverpool play with fast center backs that can catch-up over longer distances with opposition attackers, so the team can play a high line defensively. The goalkeeper is an accomplished reflex shot-stopper, and exceptional at dealing with one-versus-one situations, thus putting pressure on opposition attackers to make exceptional finishes and increasing errors. This allows for the team to squeeze the opposition in their half with defenders pushing up, it also allows the fullbacks to take more risks. This is (2) and (3) above: each position has a profile and that profile translated into the team ethic.

In terms of (2), player dynamics can be quantified with various metrics (apart from the usual suspects goals and assists) including expected goal probability — the measure of how each players contribution increases or decreases the probability of the team scoring a goal, territorial metrics based on how the player controls space on the field, and other metrics like ball recoveries and retention. Creatively constructing softer, unusual metrics, that capture the subtleties of the game have allowed for the recruitment of players that don’t fit the superstar mold (think Gini Wijnaldum) but add key elements to the teams way of playing.

To consider (3), a good characterization of Liverpool’s current way of playing is unpredictability. But it’s not an individual unpredictability in the sense of the giftedness of Messi that produces superlative rare events on the football field with surprising regularity. It’s an unpredictability that has patterns: The front three are positional specialists that work together to overload the central areas of the opposition goal, recycle the ball to supporting wide players (overlapping central midfield players or fullbacks); and they can all finish well. Roberto Firmino is known for his ability to multitask in terms of his role; adding unpredictability. The midfield three players are set up so that they can support extreme risk-taking by the front three and the fullbacks, by playing positionally conservatively. And the fullbacks can both cross and switch the play from flank to flank.

All in all, this makes it hard for the opposition to predict a line of attack, to recover from an attack, and offensively each cog in the wheel is a contributor toward raising the teams overall expected probability of scoring a goal. This covers the offensive side of the game, defensively, you want the opposite, where each player reduces the probability of the opposition scoring a goal.

In a way, Klopp might have asked the recruitment team to give him generalists, they make the game unpredictable since it’s hard to figure out what tool they will use next. At times you need a pocketknife, and sometimes a sledgehammer is also good to have, having both and knowing when to use them, makes you that much more difficult to work out.

Replacing a Star Player

It was hard to know at the start of 2018 that the sale of Phillipe Coutinho to Barcelona would set the stage for the evolution of the Liverpool team into its current vintage. Investment into Virgil van Dijk and Alisson Becker was an evolution of the team, but one on the defensive side, and between them, they have significantly reduced the probability of opponents scoring goals against Liverpool.

The key to this defensive solidity has been the reduction of unforced errors (sorry Momo Sakho), consistency of fitness from Van Dijk, and another metric which is akin to error recovery: If a mistake is made by an individual player or defender, then what is the probability that the defense will recover. There are a handful of defensive players in football that have a sixth sense of where a fire might escalate, and they are good at snuffing it out. Van Dijk is outstanding at recovery on the rare occasion that is required.

The second factor of the team’s evolution is that by selling Coutinho the focal point of the Liverpool attack evolved too. As a left or central midfielder with an eye for a pass and an excellent shot, the ball was drawn to Coutinho as the last point in the attack before something would happen for the team. But, opponents can deal with this, it’s an easy script and they’ve seen the lines before. With Coutinho sold, a metric like goal variety has gone up for the Liverpool squad, and it is easy to argue that Klopp has encouraged this amongst the players.

A mathematical model for goal variety would be to treat each goal as a story and to measure each of the different “goal” stories that a team is capable of creating. When Coutinho played for Liverpool he was the main protagonist in the story, nowadays there are a few instigators: Trent Alexander Arnold and Andy Robertson play their part, but Mo Salah, Roberto Firmino, and Sadio Mane can take on multiple roles at different times in the story — initiator, propagator, protagonist, twister, teaser, and finisher. And there is also the contribution of Van Dijk and the defense from corners, set plays and raking, accurate, long passes over the midfield to the forwards.

Knowing the Other and Adapting the Self

Apart from recruitment and knowing the self, data and models can play a key role in three parts of a football club: understanding the opposition before the game, understanding the opposition during a game, and understanding the opposition and team in the context of the story of the season. This understanding translates into in-game tactics that are decided beforehand, in-game tactics that are pulled out at specific moments during the game, and into squad management over the season. In each of these areas, Liverpool seems to be doing some interesting, excellent, things.

Opposition Analysis

In the same way that the Liverpool team have their metrics, the opposition also has their metrics, and opposition teams have their own unique identities. If you’ve built all this machinery to analyze and recruit players, as well as study your team, then it would make perfect sense to use it to analyze the opposition. Looking at, for example, how your players respond to mistakes and trying to find that out from the data, means that you have the perfect tool for asking similar, sensible, questions about the opposition.

Here are a few examples.

• Which opposition players tend to drop in performance levels after a mistake? Crowd them out, and make the game difficult for them specifically, force mistakes.
• Which opposition players tend to lose their temperament when fouled? Agitate them subtly and push their abilities to focus on the task at hand.
• Which opposition players are good at producing rare-events of individual brilliance? Cut off their supply lines and reduce the probabilities of them hurting you. Use historical data to identify their key supply lines.

In the recent FA Cup game against Everton, it is noticeable that the Liverpool team were given instructions to shoot from distance. If you’re going to analyze the opposition, then do it thoroughly: At 185cm Jordan Pickford is around 6–7cms shorter than Alisson Becker and David De Gea. He’s tall but he doesn’t have the long arms of the best, elite level, goalkeepers, so there is a chance of beating him from distance. And that is what Curtis Jones did, for the only goal of the game. Understanding the physical, playing, and psychological characteristics of the opposition, and helping to design tactics accordingly is a key role for a data scientist in sports and business.

In-Game Tactical Data Science

Weaknesses in a team’s play style and their defence will translate as extra space in certain areas of the field. This is where in-game data and analytics can hurt opponents.

One particularly interesting area is space. Good managers have an intuitive idea of where the chinks in the opposition defense happens to be, but this can also be turned into a data science problem that can run in real-time. Football happens on a two-dimensional playing surface and each player can cover in real-time a specific area that depends on how fast he accelerates. A team, collectively, covers space and also forces the opposition to make choices when attacking or defending.

By leveraging imaging data and video feeds, it is possible to understand how a given game is evolving in terms of patterns of space, and then communicating to the players where to focus their energy. Think of Spock communicating to Captian Klopp, “The fullback is tiring and space is opening up between him and the center back, on average he is covering around 0.75 square meters less as the game gets longer. That’s where we attack between 85–90mins.” This is Peter Krawietz’s job during half time at Liverpool. He provides the team with half-time video analysis. Are they also using data?

For the 2019/2020 season, Liverpool have won a few games late, and they’ve won a lot of games by managing out the opposition from the game. Tactically, the team hit the right metrics when it comes to creating a block that closes off the options for opponents and narrows down their choices. This fits with managing areas on the field, and, of course, the advantage of fast players with stamina, is bigger area control throughout a game.

Squad Management

The smallish size of the Liverpool squad (due to a combination of Klopp’s preference for camaraderie and closeness in the team, as well as providing entry points for younger players) along with the demands of playing more games per season due to success in the Champions League has meant that the squad has had to be carefully managed to get the most out of players in terms of mental and physical freshness.

This has translated to an in-game tactical strategy that is less intensive at certain times and has seen the team managing out games where they are ahead, but also over a whole season, a data-centric squad management system. From the outside, one can surmise that the Liverpool analytics team keeps a track record of players training, muscle conditioning, and performance on the field as well as the team’s travel commitments and the like. Using these data points and matching them up to on-field performance levels gives a model for connecting preparation behavior with in-game behavior. Besides, it gives insight into how to manage the squad, over the season, to get consistently optimal performance levels.

Maybe all this modeling and emphasis on data takes the romance out, but it’s good to know stuff if you are going to punch above your weight.

Skin in the Game Contract Incentives

Apart from the on the field contribution of data science to the way that Liverpool have carried out their business, there is a sense that off the field things are healthy for the club too. They are actively pushing the boundaries of the commercial side of football, and it seems that fine details like player contracts have been designed to optimize performance.

Based on little snippets of information in the media, it is clear that contracts are awarded to players according to a set structure and that incentives have been put in place to mitigate risk factors and to maximize player performance. Liverpool paid out more bonuses than any other club last season, and part of that was due to on-field success in the Champions League, but mainly due to the way players are incentivized.

Start-up companies tend to punch above their weight when it comes to results obtained for resources put in because everyone has skin-in-the-game — their individual and collective risks are correlated. Liverpool’s recruitment strategy has been around finding underrated players and making them into stars, but their players have also been given contractual incentives to meet certain key performance indicators. This is a lot different from the Alexis Sanchez and Paul Pogba deals at Manchester United, where expensive established players where brought in and found to not have “skin-in-the-game” in terms of the clubs interests on the field.

Game theory is used extensively in bargaining models in economics. The key idea is that there are equilibriums where the selfish motives of an individual agent (not a football agent!) may produce a less than optimal global outcome. For an analogy, think about the bar scene in A Beautiful Mind, the film adaptation of Nobel Laureate John Nash’s life.

In the dance between club and player, there has to be a lining up of incentives, so that the whole is more than the parts, in turn, this allows for clubs to punch above their financial weight at an elite level. This is how Klopp’s Liverpool caught up ground on Manchester City in England, and the cash-rich Real Madrid and Paris Saint Germain continentally.

To see how ingrained this culture of skin-in-the-game is at Liverpool, consider the commercial deal that they’ve made with apparel giant Nike. They have gone for less guaranteed income, a stake in global sales and a chance of global exposure through brand alignment. It’s clever, but it also has a sense of identity and integrity about it. The club is operating the same way on multiple levels.

Lessons for Business

If there is one clear message that you take from this article, it ought to be, hire more mathematicians and problem solvers, and then give them support. Get them properly integrated into your business.

The key skill of being mathematically inclined is the ability to abstract and generalize, which is a huge advantage for a business or country that wants to compete at the elite level. We only solve global-scale problems by creating a whole that is more than the sum of its parts, and the key to this is abstraction and generalization. Once you understand a problem abstractly, then you can rinse and repeat; and have multiple, purpose-built, arrows in your quiver. YNWA.

I’m a freelance mathematician and data scientist, and this article is my own opinion based on information that is available in the public domain.