‘We are the only country doing this’: Why England’s AI-driven selection policy is breaking new ground

When Jon Lewis, head coach of England’s women, revealed that he and his backroom staff were using artificial intelligence to help with selection, and that it had contributed to an important decision during the 2023 Ashes, it generated a wave of headlines. Was this a portent, another step on a road marking humanity’s defeat by machines that began with world chess champion Garry Kasparov’s loss to IBM computer Deep Blue in 1997? And didn’t the magic of team sport involve human beings enjoying a shared enterprise, then arguing about it in the bar?

The selection in question was leg-spinner Sarah Glenn for a one-day international at Bristol, even though the broad historical data suggested off-spinner Charlie Dean – armed with more experience, and a better average and strike-rate – was the logical choice. But Glenn conceded fewer boundaries during the middle phase of an innings, when Australia liked to find the ropes. Sure enough, she went for one four in seven overs, and England won by two wickets.

“The AI gave us an idea of a trend Australia used,” says Lewis. “It felt like a really well-informed, thought-through decision. On those pitches, against that opposition, we felt it was a good selection – and it worked. As a head coach, selecting the right XI is your most basic and important role. Having a data-driven science behind predicting what the opposition will do can help you. AI is a useful tool, especially for borderline decisions.”

In fact, it was already being used in cricket and other sports. Lewis first came across it through his coaching role at Uttar Pradesh Warriorz in the Women’s Premier League in India, where it helped inform strategy in the player auction and team selection. Indeed, most WPL and IPL teams were using AI in this way. Lewis was helped by a Wimbledon-based company called Prospect Sporting Insights, who have also worked with the England men’s and women’s rugby union teams, as well as Wigan’s rugby league and football clubs. The Pakistan Cricket Board, meanwhile, used AI to help select 150 players for their Champions Cup domestic competition, and Thomas Bach, the head of the Olympic movement, has espoused its benefits in identifying sporting talent, and improving judging. All the big sports in the United States use it.

“PSI have what they call simulated predictive data,” says Lewis. “I would send them different line-ups, say that these were our possible combinations of players, and ask which they thought was best. They’d run about 250,000 different simulations. They basically gave us a win percentage at the start of the game. For example, in the Ashes ODIs, Australia were 70 per cent and we were 30 per cent. Then we tried to get the percentages more in our favour, and narrow the gap.”

The selection process behind the England men’s teams has not escaped AI’s advance either. It is hard to overestimate how much things have changed since the days, as recently as the late 1990s, when the national averages in The Daily Telegraph were required reading for the chairman of selectors. But these averages revealed nothing about whether a batsman was making his big scores against weak teams on flat pitches or without spinners, or whether a seamer was dangerous only on greentops. Despite being one-dimensional, they were hard to ignore – until better metrics were devised.

Things began to change with the arrival of ball-tracking data through the HawkEye system developed by Paul Hawkins. Originally designed to assist with umpiring decisions such as lbws, HawkEye had a golden spin-off: it created pitch maps, highlighting a player’s strength or weakness depending on where the ball landed, and showing how it swung or deviated. No less valuable were the super slo-mo cameras that captured the ball’s revolutions: which bowlers were doing it, which batters could play it.

These resources were extensively mined by Nathan Leamon, appointed England’s first full-time data analyst in 2009 (he also devised an early simulation system called Monte Carlo). As the pool of Test data grew, so did the potential accuracy of the information Leamon was providing to Andy Flower and Andrew Strauss, then England’s coach and captain. Briefly, they enjoyed a technological advantage over their rivals, rising to the top of the rankings in 2011. When Strauss later became managing director of England cricket, and oversaw the appointment of a new national selector in 2018, he sought candidates who could bring a more scientific approach to selection, and exploit this new data-rich world.

The job went to Ed Smith, who says that, even during his three years in the post, AI was coming into view: “We had access to very valuable statistical tools, such as weighted averages – an algorithm which effectively turned a headline average into a measure that better reflected true performance. During that period, although most of what we did in the data space was conventional analytics, AI became much more of a talking point in society and sport. You are just trying to drill down and have a deeper, more X-ray vision of how good someone is. That is the mission statement of data analytics.” Even so, Smith stressed that all decisions required a balance of sources of evidence – data, scouting reports and coaching instinct.

Shannon Vallor, a professor of philosophy, and author of The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking, describes AI as the ability to look at the past and find patterns efficiently and easily – as good an explanation as any. Smith says: “If you want to make inferences about future probabilities based on past events, AI is an extremely efficient tool. It’s a tool that strategists in sport are going to want to have access to, but it is going to require a lot of skill.”

By 2024, Smith – having written a book called Making Decisions: Putting the Human Back in the Machine – had set up a performance consultancy with Mo Bobat, England’s head of performance while Smith was national selector. They then joined forces with Palantir, a big-data analytics company, to run Derby County Football Club’s sporting intelligence unit. “Our desire is to try and apply our learning and experiences from other sports, mainly cricket, into football,” Bobat told the Derby Telegraph. “Palantir have a number of different software platforms, including their Artificial Intelligence Platform, which is pretty sophisticated. It can help us move to solutions far quicker than standard analytics would be able to. We want them to set Derby up as one of the best-in-class from an analytics perspective.”

Recently, the ECB’s data department have grown rapidly under the leadership of Stafford Murray, their “head of analysis and insights”. One of the big advances has been the ability to harvest data from domestic cricket across all formats through the iHawk system, developed by Hawkins, and introduced in 2023. It involves umpires wearing camera kits on their chests which, among other things, can monitor a bowler’s speeds and lengths, as well as movement through the air and off the pitch. From this, it was possible to determine who was bowling “deliveries of international class”, and which batters were dealing with them best – and to establish, in the lead-up to the 2023 summer Tests, that Josh Tongue was the fastest bowler in county cricket. Partly because of that data, he played in two Tests, and did well. The data drawn from iHawk was not in itself AI, but it was vital information to feed into the AI system.

AI also helped Murray’s team determine the most advantageous match-ups – which bowler to use against which batter – and predict how a certain ground might play. But a more ambitious project was to use it to measure the impact of players, so that, in Murray’s words, “when it comes to giving data to selectors, coaches and captains, we’re not just giving them output data, as in averages or scores, it’s also their impact on the match… what effect did a spell, or action of a player, have on the likelihood of winning?” The overall aim, he said, was to ensure that the best talent, or most appropriate players, rose to the top. He added: “We are the only country doing this.”

The term “AI” covers a range of systems and tools. Which of these are truly intelligent is something even computer scientists and AI developers argue about. But, in the cricketing context, the systems used are perhaps best categorised as Good Old-Fashioned Artificial Intelligence (GOFAI), in which a machine is supplied with all the data relevant to a specific task, such as finding the bowlers with the fastest median speeds across a Championship season, or running simulations of matches based on various selection options. In essence, doing things much faster and more accurately than a human. The more information, the more accurate an AI system is likely to be, hence “self-learning machines”. Lewis understood this, since there was less historical data available around HawkEye ball-tracking in the women’s game than there was in the men’s. “It will become more accurate,” he says.

Speaking in May 2024, Murray also suggested that a change in selection might be evident that summer. “While some of it might look a bit weird, it’s with the 2025/26 Ashes in mind,” he said. “It’s a performance-backwards approach, looking at the long-term goal.” It is understood this sort of talk created tension with Brendon McCullum, England’s Test coach, probably because it ran counter to the philosophy he espoused when he took up the job in 2022 – of picking a team to win the next match.

Perhaps one reason the weight of science behind England men’s selection was rarely given its due was because of the hoopla surrounding Bazball and the free-spirited approach it encouraged. But McCullum was not unreceptive: he had been open to data analysis during his time with Kolkata Knight Riders at the IPL. And while he might back seemingly out-of-form players in the Test side, he often draws on the impact argument – that a player is making more meaningful contributions than the conventional numbers suggest.

Rob Key, the managing director of England men’s cricket and de facto chairman of selectors, was fully engaged in the data-science projects run by Murray’s team. Key and full-time selector Luke Wright worked with Murray to ensure that the right types of players for any assignment were put before the coaches and captains of the red- and white-ball teams – what Murray called “amplifying the culture and philosophy already in the dressing-room”.

What was striking was how radical England’s selections have become under Key, a sure indication that data was being followed, since one of the central takeaways from data analysis – in many spheres, not just sport – is that it tends to encourage greater risk-taking, contrary to the human instinct for caution. Several players were chosen after only a few county appearances, and with little in their traditional stats to support promotion. Shoaib Bashir’s call-up for the India Test tour in early 2024 owed something to Stokes watching social-media clips of him bowling at Alastair Cook in a Championship match, though Bashir’s unusually high action and high revs had already been identified. A similarly eye-catching, successful pick was Tom Hartley, another tall spinner with a high release point and brisk pace – his style closely resembled Axar Patel, who had a proven record in India.

While the decision to steer James Anderson into retirement could have been justified by reference to the speed gun, it was bolstered by broader research undertaken by Murray’s team into the speeds necessary to thrive in Australia – speeds Anderson was not producing.

Might the chairman of selectors one day simply be a software package on the head coach’s laptop? Not any time soon, says Smith. “I don’t think AI will ever take away the need for human creativity, innovation, imagination or insight. The data that it is drawing from has to be correct, which is a massive assumption, and the context of the future has to be something like the context in the past. We need to focus more than ever on creativity and imagination in our problem-solving, because the bits that are open to computational intelligence are going to be done very quickly and efficiently by machines. The art is going to be getting the best out of human beings and the best out of AI.”

Murray agreed. “You can’t replicate what the skipper can see, feel, smell, taste out on the pitch. There’s always contextual stuff going on that the numbers won’t measure. And that’s why AI will never replace the decisions of people. That’s the fun. If you could measure everything, it would be boring.”

Simon Wilde has been cricket correspondent of The Sunday Times since 1998.

Follow Wisden for all cricket updates, including live scores, match stats, quizzes and more. Stay up to date with the latest cricket news, player updates, team standings, match highlights, video analysis and live match odds.