The XV of France puts Artificial Intelligence in its intelligence game

The XV of France puts Artificial Intelligence in its intelligence game

A meeting. This is how Fabien Galthié, coach of the French Rugby XV, and Jean-François Sébastian, general manager of SAS France, explain the genesis of a project launched in 2021 with the French Rugby Federation (FFR). (Photo above: from left to right, Fabien Galthié, Jean-François Sébastian, Julien Piscione. ©Julien Poupart/FFR)

The two men met the previous year and worked to launch a multifaceted new analytical project.

The prototyping phase will last one year. The partnership starts in 2020 with the organization of a Design Thinking workshop between Fabien Galthié, the performance unit of the XV of France, and SAS.

“Data has been used for twenty or thirty years in the business world: their analysis is a decision-making tool on all matters,” observes Fabien Galthié. “So we asked ourselves: can we apply this to our work? In other words, prepare well and beat the opponent we’ll meet over the weekend? And we began to reflect on everything, without forbidding ourselves anything”.

Increased AI key

At first, Charlotte Douette, consultant in artificial intelligence and analysis at SAS, accompanies the data scientist and the “sport scientists” of the XV of France to analyze the results. “We started working with Karim Ghezal, [l’entraîneur spécialiste des touches N.D.L.R]simply because he was already aware of the data,” he says.

However, the XV of France and SAS did not start from a blank sheet. “Within the staff and in many teams there are performance analysts who analyze match results,” confirms Julien Piscione, head of the French Rugby Federation’s performance support department. Even the data analysts of the XV of France were already dealing with Python scripts. But with the new project, analytics will take over.

With the support of the federation, the national team therefore creates a database of 1,438 matches (Top 14, U20 and international) from 2019. “For the XV of France, this represents 36 matches, or 450 successes,” calculates Charlotte Douette.

The French team jerseys are equipped with GPS trackers. These devices produce the vast majority of data recorded during a game. Sports scientists supplement them with statistics on the number and type of actions. In total, there are around 4,500 events, plus a hundred variables per event and per game. In short, millions of lines.

“The goal was to be able to recover all the characteristics of the touch, be it the composition, the speed of release, the position on the surface of the ground, to have the strengths and weaknesses of each within this phase of the game “explains Charlotte Douette.

We remind you that in Rugby the touch is not a simple throw-in, it is a conquest phase, where the two teams challenge each other, most of the time in the air, to regain possession of the ball. Hence the interest of “design” to win the greatest number of conquests.

The predictive analytics resulting from the processed data is now used to prepare the training.

Twelve game prediction modules

Gradually, Charlotte Duette, her team and the staff of the XV de France developed twelve prediction modules. “We worked on scrums, rucks, kicks, penalties, etc. “, she says.

These phases of the game are also contextualised by field area. With the historical data, the analysis cell of the XV of France can thus know the average number of types of actions carried out by its team and by the opposing team.

From historical averages, analysts derive a forecast of the actions that will take place in the next match: number of touches, scrums, rucks, etc. Again to help prepare for training.

The head of analysis of the XV of France, Nicolas Buffa and his analysts rely on a web interface provided and adapted by SAS. This “easy to use” user interface allows you to “cross information between areas of interest, durations, etc. “says Nicolas Buffa. “It is the principle of being able to ask at will”.

Analysts therefore do not necessarily use SQL queries. However, they don’t always get the desired results. “You have to learn to use it, you have to learn to dig,” says the head of analytics. “Sometimes it doesn’t give much, but every week we find an interesting indicator.”

The data is fed into the SAS Viya platform hosted in Azure. As for the modules put in place, they automatically generate reports which are sent to the staff members of the XV of France.

In addition to a data query tool, dashboards are associated with forecasting modules.

Data-driven training

During the ‘Game Observatory’ masterclass which took place on 11 June at the National Rugby Center, Nicolas Buffa explained how he and the performance unit analyzed Scotland’s match history to try to predict the progress of the match on 23 February 2022 against France in order to plan training.

“The performance cell had estimated that training [influencé par ses analyses] accounted for 37.5% of the course of a match against Scotland,” he recalls. This percentage seems low. In fact, there is nothing abnormal. The predictor indeed focuses on the start of the play and the launch of the game.

“There is not a match where the scenario we have planned takes place. »

Fabien GalthieCoach of the XV of France of Rugby

If the French team does “predictive”, it doesn’t do “predictive” (but forecasts, correct translation of English “predictive”).

“Rugby is a very disruptive game that changes every three seconds: whoever was standing is down, whoever was down is standing. We may be in our camp and we find ourselves in the opponent’s camp. We may be in possession, and [d’un coup] we are in defence”, recalls Fabien Galthié.

Therefore, it is impossible to simulate, or predict up to the minute, the course of a rugby match between two nations. “There isn’t a match in which the scenario we have planned takes place,” underlined the coach. This in no way calls into question the usefulness of the analysis. In reverse.

An algorithm to better evaluate opponents

After a game, for example, key statistics are analyzed to debrief the team and anticipate the next game.

Even if they don’t predict anything (since they are statistics and therefore probabilities), historical data on other teams is also very useful for establishing attacking strategies.

For example, during the Observatory of the game, Laurent Labit, coach in charge of the attack, evoked elements of preparation for a hypothetical match between France and New Zealand.

Among the win conditions, analysts noted that teams that won against the All-Blacks had a game that limited the number and length of rucks.

But the XV of France goes further. With SAS, Nicolas Buffa has developed – with the participation of Fabien Galthié himself – an algorithm to record the experience of the collective and its evolution during a match.

The strength of a team is established, among other criteria, by counting the number of international matches played together by all pairs of players (each player has 14 possible “pairs”) making up a team (105 pairs of players in a team of 15 players). “Let’s take about fifteen nations and compare the figures of the last three or four years to estimate the experience of the teams,” anticipates Nicolas Buffa. “We realized that for South Africa we had to go even further back and that the team had four times more collective experience than us.”

During the “friendly” match against South Africa (two red cards), the French collective’s score went from 12 out of 24 at kick-off to 7.4 out of 24 after captain Antoine Dupont was sent off for dangerous play. “It is a team with” 7.4 “of collective experience that has gone looking in this last quarter of an hour to regain the score against a team that had 24 out of 24”, welcomes Fabien Galthié. Proof in passing that an isolated metric isn’t everything.

In fact, the coach used other levers and “secret” criteria, and focused on “other objectives” to win against the Springboks. But he won’t say more.

Collective intelligence comes before artificial intelligence

“It’s not the numbers that guide us, we guide them. »

Fabien GalthieCoach of the XV of France of Rugby

Advanced analysis is therefore increasingly used by the XV of France. But Fabien Galthié insists “it’s not the numbers that guide us, we manage them”.

For the farmer, analyzes remain tools Help which are available to staff. Important decisions remain the responsibility of the human being. For example, the former international scrum-half does not do real-time analysis. “During a game, I don’t connect to anything,” he says.

Likewise, the selection of the individuals that make up the collective remains the exclusive responsibility of the recruiter and his staff, without there being the need to involve algorithmic models.

As for the mental support of the players, it can certainly be based on indicators, but even in this case it is the coaches who have to find the keys to understanding and advising rugby players.

Analytics takes precedence over video

Intuition has therefore not disappeared. It is now verified by the numbers. And analytics is much more popular than traditional video analytics.

“We put collective intelligence at the heart of the project. And in the center is the analysis cell. »

Fabien GalthieCoach of the XV of France of Rugby

“We put collective intelligence at the heart of the project. And in the center is the analysis cell”, assures the technician. “The main evolution is there. Previously, coaches had an intuition and used the video cell as a match library. Today we can spill everything: we work on methodologies, ways of thinking, etc. »

The data is also used, in a more traditional way, in training (Thibault Giroud, performance director of the XV of France, relies on the data at his disposal to plan the physical preparation of each player), or to reduce the tiredness associated with the large number of meetings.

However, the XV of France and the FFR validated the potential of the analysis after the Grand Slam achieved during the 2022 Six Nations tournament. Fabien Galthié and his staff are already anticipating the next tournament as well as the World Cup in 2023. Initially , it will be about improving the modules in place. But FFR has plans to launch other analytical projects. Initiatives “to be followed” therefore. Like a candle from another time.

Leave a Reply

Your email address will not be published. Required fields are marked *