How AI is changing rugby and the big question that must be answered
How AI is changing rugby and why players will not need telling
Japan’s Gary Gold is using AI to interrogate a season of League One data, and analysis rooms across the game are changing at speed, but the real test is whether analysis helps a player recognise his options in training and adapt on the pitch without waiting for another instruction.
Gold had five million data points to play with once last season’s 20 rounds of League One were done. Japan’s assistant coach, sitting at Eddie Jones’s right hand, ran them through Claude Code, Grok and Gemini, and what caught our eye was the question he asked of it rather than the sheer volume; he wanted measures the standard feed doesn’t offer, metres conceded off opposition carries among them. He’d still tell you, as he told the Daily Maverick in June, that it’s “never going to replace the human element completely”.
He’s right, and the human element he means is the one that matters most on a Saturday. A player reads and reacts in fractions of a second, a shift in footwork here or a glance at a defender’s hips there, and no instruction from the coaching box can arrive in time for the moment in front of him. So here’s the question for every analyst now armed with AI: does any of this help a player see his options on a Tuesday and back himself to use them at the weekend?
The tools have arrived
Make no mistake, the change in working practice is happening now. Hudl put an AI assistant into Sportscode in August that writes and fixes the code scripts behind an analyst’s reports, and more than 800 analysts across all sports have used it since; the Welsh Rugby Union’s advert for a data scientist this month asks repeatedly for AI-assisted coding.
BBC Sport’s Mike Henson reported last week that all 10 Premiership clubs, more than half the URC and a good chunk of the Top 14 take Oval Insights’ modelled data, and that Simon Orange has bought an AI company which Sale will use for injury prevention and player availability first and recruitment second. New Zealand Rugby’s staff ask questions of their performance data in plain English through Amazon Web Services, whose case study calls the platform agentic whilst leaving rather vague what the agent actually does, much to the delight of marketing departments all over Seattle.
What AI adds to the analyst’s job is mostly speed and reach. It can make the relevant passages quicker to find, turn up situations a busy staff might have missed, and make individual preparation practical across a whole squad rather than for a handful of senior players. Whether any of that turns into better decisions on the pitch is the test rugby should apply, and I haven’t found published evidence that any system has yet demonstrated it.
A script can’t keep up with the game
No coach can script decisions at the speed rugby is played. Staff can adjust tactics during a match, but nobody in the box can direct a player’s response to the defender in front of him, so the plan has to be something he carries in his head, a set of principles and triggers telling him what to look for and what each picture invites.
England: Inside Twickenham’s 2027 Six Nations VIP experience
This, rather than a sequence of instructions he’s waiting to be reminded of; and the analyst’s job in the week becomes building perception, so that by Saturday a player recognises a picture a fraction earlier than his opposite number and already knows what it offers.
What analysis can teach a player to see
It starts with finding the cues that matter. Good analysis can show which early signals tend to come before which pictures: the width of the opposition 13 as the ball leaves the ruck, the position of their 15, a defender’s line speed off the edge or the set of his hips. Help a player recognise earlier where space may develop and you’ve given him something he can act on the moment it appears, adapting as the opposition try to disguise it, which is exactly what a good plan should be made of.
From there, a coach can train the pictures rather than the plays. Analysis can identify the situations a side faces most often in matches and how often each one occurs, so practices can reproduce those pictures with enough variation that players have to read and solve them rather than rehearse an answer, and by Thursday the coaches find out whether they can solve them under pressure. A personal clip library of a player’s own decisions, and of the pictures he’ll face against Saturday’s opposition, lets him see the same problem dozens of times without taking a bang, which matters more than ever given what smart mouthguards are now telling medics about the cost of contact.
The read has to be shared, too. Rugby decisions are coordinated, and a 10’s read is useless if his outside centre has read something else entirely, so analysis of who saw what, and whether the line made the same call, helps build the common picture that lets 15 players adapt together without anybody bellowing instructions across them.
URC, Round Five: Predictions, teams, kick-off times, how to watch and referee appointments
Grade the read
None of that works if the review room grades the wrong thing. Cricket’s scorebook has always had this problem; a duck is a duck, whether the batsman nicked a wide half-volley or was done by the ball of the summer. Judge a player purely on whether the move came off and he learns to fear the result, but judge him on whether he chose the best option available, given what was actually in front of him, and he starts to trust what he sees. A good read that ends in a dropped pass gets credit, and a poor one rescued by luck gets questioned.
It also means knowing what he was meant to be doing. Picture Siya Kolisi on a night when South Africa ask him to play as a roving sweeper, speeding up their ruck retention. A stat sheet that counts carries and turnovers can make him look as though he’s had a quiet night, and a model that doesn’t know his assignment could mark down his limited carrying whilst failing to credit the breakdown work that occupied him.
Want more from Planet Rugby? Add us as a preferred source on Google to your favourites list for world-class coverage you can trust.
Then there’s the player who does exactly as he’s told. Take a winger told to rush up on the outside of a blitz defence. He comes off the line bang on cue, his line speed is the best in the squad and every model built around the defence coach’s plan scores him highly; but if the opposition ten has spent the week practising the chip into the space behind him, the winger has done what he was asked and what the opposition wanted, all in the same moment. A winger taught to read the ten’s shoulders and to know where his own full-back is covering would at least have had the chance to make a better-informed choice; holding might have stopped the chip, or it might have opened the outside channel, and that’s the decision the review room should go through with him.
A system built around the coach’s own definitions gets very good at telling you who held width, who rushed on time and who hit the ruck when told, and it can leave the plan itself barely examined, which turns an expensive piece of software into a very sophisticated way for a coach to defend what he already believes. So point some of it at the plan: measure each passage against what the opposition were trying to do, and pull out the times somebody broke the plan and it worked, alongside the times it was followed to the letter and failed.
The player has to be part of that, and his account is evidence too, to be set alongside the instruction, the footage and the circumstances; he can be completely honest and still misremember a call or rebuild his reasoning after the event. The review works best as a joint investigation, with coach, analyst and player testing each other’s version against the footage.
What better measurement adds
Most of this can be done with what clubs already hold, but richer measurement would sharpen it. Shared match data is still a record of selected actions, a tackle here and a ruck entry there, and the contact area remains hard to read automatically. A study published in 2024, using data from eight Tests played in 2020, found Catapult’s microsensors picked up between roughly 53 and 85 per cent of tackles made, depending on position, and I couldn’t find published independent validation for the camera-tagging products being sold to clubs to code that area.
Tracking would change the decision-making conversation most. Give every player an x and a y, many times a second, and you can reconstruct the picture around him: where the space was, which options were open, how early it formed and whether he moved before it closed. Coordinates can’t tell you where he was looking or what was blocking his view, so that reconstruction is where the conversation with the player starts rather than a standard to judge him against. The NFL tracks every player in every game with chips in the shoulder pads and the Premier League has had optical tracking for years; rugby could follow if its leagues fund it centrally and agree how clubs get access.
The smart mouthguard, now worn in elite competitions that use the head injury assessment, records the linear and rotational acceleration and angular velocity of the head; it doesn’t measure the force exchanged between players or record every contact, but laid over video it can show which tackle shapes and body positions produce the biggest head readings. In my view it’s the most important new data stream in the professional game, and its first job is welfare.
Mouthguard information used to assess a player’s health is protected as health data, and the hard questions are who reviews it, what it obliges a club to do and whether it may ever influence selection or contracts; football’s FIFPRO data charter is a useful comparator for whatever rugby agrees.
Where this leaves rugby
The software will keep getting better and cheaper, and because it works on information clubs already hold, it’s the easiest part to adopt. Two clubs renting the same models will often turn up overlapping patterns and candidates, but the side that gains most will be the one whose analysis has taught its players to recognise the picture a fraction earlier and trusted them to act on it.
And every Monday, its coaches will keep asking themselves one question: what evidence would convince us that a plan we strongly believe in is wrong?