How Liverpool are using artificial intelligence to become better at corners

As Arne Slot settles into his role as Liverpool’s head coach, there is an exciting array of fresh ideas from his new backroom staff.

The novel training methods of assistant coaches, Sipke Hulshoff and John Heitinga. The vibrant voice of lead physical performance coach, Ruben Peeters. The creative coaching drills of head of goalkeeper coaching, Fabian Otte.

Off the pitch, Liverpool have been developing a coaching assistant who could help their set-piece approach for the coming season… and it’s not even human.

Liverpool’s analytics department is renowned for its pioneering work, led by director of research William Spearman who took over from Ian Graham in 2023. In their most recent venture, they have teamed up with Google DeepMind, using artificial intelligence (AI) to determine strategies for corner kicks.

The collaboration has culminated in a paper published in Nature Communications — TacticAI: an AI assistant for football tactics. The project, led by researchers Zhe Wang and Petar Velickovic, used data from 9,693 corner kicks collected from the 2020–21, 2021–22, and 2022–23 (up to January 2023) Premier League seasons, feeding information on each player’s height, weight, starting location, and movement throughout the corner routine.

The information from each player enabled the researchers to predict the outcomes most likely to occur within a given corner setup. For example, which player is likely to receive the ball? Will the sequence lead to a shot attempt?

Once the sequence has been played, the analysis can then build a picture to determine whether similar routines have been successful in the past. Crucially, TacticAI can draw from this analysis to generate suggestions that improve the outcome of the corner. For example, moving players’ positioning or body orientation to reduce the chances of conceding a shot from a specific corner.

The graphic below shows four suggestions made by TacticAI to tweak players’ positioning when defending a corner.

As evidence of the model’s capability, experts at Liverpool — including assistant coaches, video analysts and data scientists — were unable to distinguish between the output provided by TacticAI’s suggestions and real-life corner routines, with the suggested routines favoured from the original corner sequence 90 per cent of the time — highlighting how effective the model can be in providing improvements within the tactical structure.

As you can see in the graphic above, the advised tweaks might appear subtle but joint-lead researcher Velickovic emphasises these suggestions are in-keeping with modern-day football. Small changes in distances, timing, and decision-making can be the difference between victory and defeat.

“The aim is not to tell you that a player needs to move two or three metres to the left or the right,” Velickovic told The Athletic. “You make minor adjustments to a player’s location, orientation or velocity which are all relatively small, and that was deliberate.”

A core aim of Liverpool’s project with Google DeepMind is to provide coaches and analysts with a tool to help them with workflow. Opposition analysts often watch hundreds of videos in the lead-up to a game, which is labour-intensive. The ability of TacticAI to sift through similar opponent routines and curate defensive strategies is powerful.

It is more time-efficient to draw conclusions in a fraction of the time and this work provides an objective tactical approach without bias, underpinned by thousands of examples.

“This tool is designed to accelerate a coach’s ability to spot patterns,” Velickovic said. “Coaches are looking at complex situations with 22 players, and they have to work out what the key parts are that made or broke a particular strategy — and which players were responsible.”

“With a system like this — where it immediately produces adjustments to all of the defensive players — you can focus your attention and immediately spot that, for example, a specific defender is doing something wrong.”

“If this happens to your defender over many situations, you can then try to fix it in your coaching. If it is an opposing player, you can work on strategies to exploit that weakness.”

Trent Alexander-Arnold takes a corner against Atalanta last season (Jonathan Moscrop/Getty Images)

It is worth noting this research evaluates the potential use of such an approach, and is yet to be rolled out in Liverpool’s matchday analysis. Nevertheless, the rise of dedicated set-piece coaches shows how many clubs are growing wise to this important part of the game — with 28 per cent of all goals scored from dead-ball situations in the Premier League last season.

GO DEEPER

Set-piece coaches, analysts… and chefs: Your club’s secret-weapon staff member

For Velickovic, the marginal impact this work can have could prove crucial.

“We can never exactly forecast what will happen in a corner,” Velickovic said. “But  if you increase your chances of scoring or decrease the chances of conceding by even one per cent, that can make a huge difference over the course of a season.”

Corners have been responsible for some of the most important moments in Liverpool’s recent history. Jurgen Klopp’s final trophy as manager was thanks to a Virgil van Dijk header from Kostas Tsimikas’ corner to clinch the Carabao Cup against Chelsea in February.

The unlikely figure of Alisson Becker scored an iconic last-minute header to convert Trent Alexander-Arnold’s corner against West Bromwich Albion in 2021 to keep Liverpool’s top-four hopes alive.

And Liverpool fans do not need a second invitation to relive Alexander-Arnold’s quickly-taken corner to Divock Origi in the 2019 Champions League semi-final against Barcelona — one of the most memorable goals in the club’s history.

Ultimately, the delivery Alexander-Arnold provides from set pieces is the crucial component to ensure a choreographed routine is executed. As set-piece specialist Gianni Vio has previously told The Athletic, “The taker is the most important player in set pieces.”

The statistical models can be instructive for coaches, but without strong delivery, the sequence collapses.

GO DEEPER

Gianni Vio on the art of set pieces: ‘They are a game within the game’

Looking at last season, the numbers suggest Liverpool could improve their conversion from corners when assessed against the rest of the Premier League. Their 4.2 goals per 100 corners was their poorest rate since 2018-19, with their overall goals scored and conceded per 100 corners almost perfectly aligned with the league average.

This project was the final milestone in Liverpool’s multi-year journey with Google DeepMind, but the collaboration shows an ongoing desire to push boundaries and find an edge when using analytics in football. It is a partnership that traces back to 2021, where their work has seen them publish two other papers relating to AI in football tactics and its use in analysing penalty kicks.

The constant evolution of football tactics means that the analysis of corner kicks is a dynamic process that should be continually updated. An optimal strategy in 2015 might look different today. The best for the German Bundesliga could be different from the Premier League.

The potential for this AI model to grow across multiple seasons and wider leagues is where analysts can create an extra dimension in their tactical approach and strategic decision-making.

At its core, AI is the ability of a computer to perform tasks that we typically associate with humans. The growth of AI — specifically, Generative AI via platforms such as ChatGPT, Claude, and Google Gemini — has been visible across multiple industries worldwide.

GO DEEPER

‘Will Harry Kane be a good signing for Bayern?’: The rise of Generative AI in football scouting

Advertisement

While these platforms are fantastic resources for coaches and analysts to use as tools, they should never replace human expertise. Data’s place in football is secured, but the combination of objective and subjective analysis is where staff can optimise their in-game decision-making. We are not entering a world where football is played by machines.

For now, at least.

(Top photo: Playmaker/MB Media/Getty Images)