Premier League: How to find the edge in data analytics - examining trends and what is to come (2024)

The brains of football data departments from around the world descended on Stamford Bridge last week.

At a similar analytics conference a decade ago, the room would have been full of far more data sceptics, with very few practitioners working at clubs.

But today, more clubs than ever have data analysts, scientists or engineers in the building, spend more money than before on data and tools, and see analytics as a genuine means of finding an edge.

The event was held by Statsbomb, which operates in the professional, media and gambling spaces. Deals with the majority of Premier League clubs, alongside teams from Roma in Italy to Peru’s Alianza Lima, reflect the company’s rise. It also keeps a keen eye on developing the use of data in women’s football, with teams in those leagues able to get access to their premier platform — Statsbomb IQ — free of charge.

“Doing analytics” isn’t a binary thing though, and the impact on decision-making at a club can be limited if the data department is understaffed, under-funded or under-skilled, or its work is undervalued and people are not being listened to. Case in point: a club’s head of analytics was side-tracked during one of the talks having been asked to order some IT supplies for another member of staff.

Football certainly lags behind other sports.

Most NBA basketball franchises, for example, will have four or five analytics staff. In the Premier League, it is thought only Liverpool and Manchester City have an analytics headcount exceeding that.

The hires are coming, though.

Aston Villa are expected to appoint a data scientist to underpin their football research team in the coming weeks and will look to expand the department before the end of this season. Everton are currently on the lookout for a data engineer to help beef up their internal analytics offering. During the conference, news broke that Manchester United had hired their first director of data science — as first trailed by The Athletic almost a year ago.

Considering three of the sessions on the day gives us a good overview of where football analytics is at, and offers some hints as to what the future might hold, too.

Liverpool’s director of research, Dr Ian Graham, gave a talk focusing on how to maximise the impact of analytics within a club.

As is the case with Graham, the presentation was punchy, filled with eyebrow-raising anecdotes while at the same time being extremely digestible. “It’s like you’re having a chat with a mate down the pub,” noted one esteemed Championship analyst.

Graham’s message was to focus as much as you can on doing work that has an impact. That may seem like common sense, but common sense in football is not common practice. Compared to trying to prevent injuries or improve performance using data, Graham spoke about how “player recruitment and retention is the most important work, by a factor of 10”. Few in the room disagreed.

Graham also showed, with some basic maths, just how difficult player recruitment and retention can be. In a simplified example, he explained how it’s roughly a 50:50 split between the transfers that succeed and those that fail, even when you have 90 per cent confidence across a group of categories, such as in the chart below.

That, perhaps, goes some way to explain Liverpool’s lack of transfer dealings this summer, with the exception of Ibrahima Konate being bought from RB Leipzig and young Harvey Elliott becoming a full-time member of the first team. It’s better to keep players whose abilities you know and can trust, retaining them for their peak (Trent Alexander-Arnold) or post-peak (Jordan Henderson) years, rather than roll the dice trying to replace them.

There was also a peek into Liverpool’s thinking from a transfer strategy point of view, with Roberto Firmino’s move in 2015 used as an example. Firmino had two years left on his deal at Hoffenheim, a mid-level Bundesliga team, at the time. He had no international appearances for Brazil and barely scored, notching 38 in 126 starts, roughly 0.31 goals per 90 minutes — the sort of numbers put up by Che Adams and David McGoldrick last season in the Premier League.

None of that tallies with what you might expect for a starting striker in an elite Premier League team, but that is exactly how Liverpool think. “Work out what’s overvalued by the transfer market and pay less, work out what’s undervalued and that’s cheap anyway”, was Graham’s summary.

Lastly, there was also an insight into the metrics used by Liverpool.

Graham spoke about how a player’s “impact on goal difference compared to the average Premier League player” is of particular interest, although how that is calculated was kept quiet.

Probably the smart choice to keep the secret-sauce secret, with Liverpool’s sporting director Michael Edwards among those in the room.

If Graham’s talk was light on technical details, the talk by Dr Will Morgan, a data scientist at Statsbomb, was anything but.

Morgan was using Statsbomb’s “360” dataset, which advances traditional event data — every shot, pass, tackle, et cetera — by adding the position and team of each player on the broadcast footage at the point of action.

For example, traditional event data looks like this:

Statsbomb’s 360 data looks like this:

That increased level of granularity opens up plenty of angles of analysis — in particular, those situations where something didn’t happen but could have.

Morgan’s talk focused on predicting whether a possession would be less than five seconds in duration before the ball is lost, more than five seconds long before losing it, or likely to result in an effort on goal. Understanding the reasons why a possession is likely to not last very long in the first instance is interesting — such as there being a lot of pressure on the ball, or very few obvious passing options.

Identifying these “near-misses” lend themselves nicely to a coaching context too. Situations in games found using Morgan’s model could be linked to video, allowing coaches and analysts to question why something happened, using the answers to inform tactics in a future game or to feed back an individual coaching point to a player.

There’s also the obvious recruitment use of a model like this too, such as finding those players who excel in consistently retaining possession in difficult situations.

It was complex work, but the outputs are clear and lend themselves to a more macro trend about the economy and where life is going in 2021: more and better data means that work can be automated, placing further emphasis on the skills needed to interpret and use the data, rather than to collect it in the first place.

Another big pull of the day was the sporting director panel, featuring Leeds United’s Victor Orta, James Cryne from Championship side Barnsley and Will Kuntz of Major League Soccer’s Los Angeles FC.

The panel touched on topics including the impact of Brexit and introduction of governing body endorsem*nt (GBE) points — the system to assess a player’s eligibility to play in the UK — on recruiting both players and managers, to how the trio of influential figures learn from their mistakes and improve processes at their clubs.

Cryne and Orta, given the clubs they work for, were both particularly negative when considering the impact of GBE on the game. Orta’s slant was that, if the end goal was to increase the quality and volume of English talent coming through academies, it was more harmful than helpful to restrict the talent pool from which clubs can recruit. Having better quality on the pitch will raise standards and create a better pipeline of talent but reducing that quality will only make it more difficult.

Cryne, who is also a director and co-owner of Barnsley, noted how the likes of Valerien Ismael and Gerhard Struber — two recent managerial hires for the Yorkshire club from continental Europe who are now at West Bromwich Albion and MLS side New York Red Bulls respectively — would be far tougher to hire now owing to Brexit.

Orta also spoke about some of his earlier forays into analytics. With datasets limited, he took a look at creating metrics based on the player ratings from the big European newspapers — Marca, L’Equipe, AS, et cetera — and used that as a scouting filter. Fantasy points were also an example raised as an early dataset that was explored as a means of finding undervalued players.

Some photos from the afternoon session#StatsBombConference pic.twitter.com/pYFKlj9zab

— StatsBomb (@StatsBomb) October 8, 2021

Kuntz’s LAFC obviously operate outside Europe, and have to deal with the labyrinth of MLS rules. His proximity to the American sports market also gives him a different perspective and he noted how, in the last three or four years, football has been “having the same fights as baseball did 15 years or so ago”. The fact that those fights for influence, input and understanding are taking place are a sure-fire sign of the progress of analytics in football.

Kuntz also spoke of rigorously reviewing every transfer window to question what went well and what could be improved upon.

The 2015 book Superforecasting was name-checked by Orta during the panel, with the extensive reviews advocated by Kuntz being reminiscent of one of the main takeaways found by its author, Philip Tetlock.

So, what did we learn?

The datasets, such as Statsbomb’s 360, and the tools needed to get the most out of them are making it cheaper and easier to conduct complex yet insightful analysis and build tools that anyone across a club can have access to in order to make objective, informed decisions. Teams now are limited by their ambition and investment, not a lack of raw materials.

The real edge when it comes to applying analytics, though, isn’t how good your model is (although a bad model is worse than not having one at all) or how smart you are, but how you can get buy-in and have a real, direct impact on the course that your club is set on.

Graham will happily mention some of the metrics Liverpool use in passing, or how they have such a keen eye on focusing almost all of their energies on player recruitment and retention. Similarly, Orta will speak about the ease of working with Leeds head coach Marcelo Bielsa, because he knows exactly the sort of player he wants. Kuntz will tell you the importance of looking to learn from your wins as well as your losses.

What you’ll never be told, though, is how they resolved conflicts in those earlier days, getting everyone to pull in the same direction. They’ll also not say — publicly at least — how to structure these post-window reviews, or how to translate a manager’s player profile into something that’s searchable with data.

They won’t tell you how these things are done, just that you should do them because they worked for them.

That, ultimately, will always be the challenge clubs face when trying to get smarter and more informed.

Sometimes the edges are what’s left off the slides — not what’s actually on them.