From squad selection to modelling how footballs behave at altitude, statistician Matt Penn explains how data is helping shape the modern game, and why coaches will always matter more than the numbers.

When the England men's team takes to the pitch, they are supported by more than the coaching staff on the touchline. Behind the scenes, a team of analysts works to turn data into insights that help shape preparation, squad selection and matchday decisions.

One member of that team is Matt Penn, who completed a DPhil in Statistics at the University of Oxford before joining The FA’s insights team in 2025.

His route into football began almost by accident. 'It was a bit of a crazy ride, really,' he says. 'It was Covid. I was doing my master's year, sitting at home playing Football Manager, and thought, "Well, maybe I could do some of this for real."' Matt sent a series of cold emails to football clubs. Oxford City FC replied, giving him the opportunity to work alongside the club while he was still studying. 'I ended up doing essentially just on the side some data analysis for them for a couple of years,' he says. 'I was so green when I started. They were very generous.'

From Oxford City he spent six months with Oxford United before joining Italian Serie A club Como after completing his DPhil. Although he had originally imagined an academic career, those early experiences changed the way he thought about statistics. 'I think it's easy to go in with a very academic mindset and think, "Oh, I'm going to show you something that's statistically sound, but maybe not very useful."' What mattered more, he found, was usefulness over elegance.

For Matt and his colleagues, much of that work happens long before a match begins. Before international camps they analyse potential squad selections, profile players and study upcoming opponents, producing reports that help coaches prepare. 'We might look at a team and say, based on comparing them to other teams of similar quality and looking at the quality of games they've played, they are a team that crosses the ball a lot, for example,' he says. 'It's taking all of their numbers and putting them into the context of what it will be like for them to play against us.'

That often means looking beyond the headline statistics. International football presents particular challenges because teams can qualify having played opponents of very different strengths. 'You might look at a team and they've played San Marino twice and Andorra twice and they've scored fifty goals,' Matt says. 'They're not actually going to score ten goals against you.'

Turning a complex model into something a coach will actually use is its own discipline, Matt says, and one he underrated at first. 'Visualisation is way more important than I thought it was going into football,' he says. 'I always thought the models were the most important bit, and then you make a quick graph and you're done with that section.' The better approach, he's found, is tying every finding back to the pitch itself, rather than leaving it as an abstract number on a graph.

Some of the questions the team investigates have more to do with physics than with tactics. During the World Cup, for example, England's last-16 tie against Mexico was played at the Azteca in Mexico City, more than 7,000 feet above sea level. Using tracking data collected from the ball, Matt and his colleagues fitted a model of ball flight, accounting for factors including air resistance and spin, and compared its behaviour at altitude with its behaviour at sea level. Rather than relying solely on theoretical calculations, the team could test the model against real match data. One finding was that more long passes went out of play at the Azteca, suggesting that the altitude subtly altered the rhythm of the game. 

Sophisticated models only go so far, though. 'If your model really disagrees with what the coach is saying, the model is probably wrong,' he says. 'Coaches are very good at their jobs.'

Coaches see players every day. They pick up on things a model never will: confidence, how someone's been in training that week. But the numbers cut both ways: they can back up a coach's instinct or challenge it.  He recalls an example from before he joined England. 'There was a really tall player who our model said was bad at winning headers,' Matt says. 'That got a lot of pushback at the time. People were saying, no, this guy's great in the air. But when you dig into it, he'd actually only won about 10% of his headers. He wasn't good at heading; he was just tall enough that it looked like he should be.' Even so, Matt is careful not to overstate the model's findings. 'You have to treat it as an opinion and not an authority,' he says. 'There are always going to be other factors that coaches will take into account correctly that you can't capture.'

Football stays stubbornly unpredictable, too. 'The difference between a really good substitution and a really bad substitution might be 1% in terms of your chance of winning the game,' Matt says. Those small improvements, he says, only add up 'over lots of games.' Tournaments give you even less room for that.

'You can be by far the best team in the world, and you would still have less than a 50% chance of winning the World Cup.'

That unpredictability, Matt says, is part of the appeal. 'It's one of the things that's great about football,' he says. 'Because it's low scoring, you can get these amazing upsets.'

Data has moved from a backroom tool to something fans expect. 'You even see it in popular football broadcasting now, they'll show xG (expected goals) as one of the main stats,' Matt says. 'Four years ago that wouldn't have been the case.' Artificial intelligence is changing some of the groundwork behind that shift too, speeding up how quickly analysts can build visualisations and test ideas, but the judgment still sits with people. 'In the past you'd think for a while about an idea and then you'd kind of have time to try one, maybe two ideas,' he says. 'Whereas now you can go, "I've got these ten ideas. Let me spin up ten, compare them and then take the best."'

Looking back, he believes one of the biggest lessons he has learned since leaving Oxford has been the benefits of simplicity. 'I've definitely grown to value simple models much more,' he says. 'I value a model that I'm 99% confident will work and is stable above a model that I maybe think is better, but where there are more pitfalls.'

For students interested in applying statistics beyond academia, his advice is to work with real-world data. 'Getting to grips with some real messy data is really helpful,' he says. 'I think getting that experience of working with some weird data is much more valuable than the classic ‘clean’ datasets.'