Meet the technology director behind the first ‘digital’ Tour de France
- 31 August, 2017 07:00
There is no part of the sports experience that will not be radically impacted by advances in technology, says Peter Gray, senior director - technology, sports practice at Dimension Data.
Gray has been at a critical juncture to observe - and work - at the most popular cycling sports event in the world, the Tour de France.
“I have the best job in the world,” he says, smiling.
Dimension Data is the technology partner of the Amaury Sport Organisation (ASO) that organises 70 sports events annually, including the Tour de France.
Thus, for the past three years, Gray has been working as the technical director for the Tour de France.
I have the best job in the world
This year, the race began on 1 July in Düsseldorf and finished at the Champs-Élysées in Paris on 23 July.
The first Tour de France was held in 1903, sponsored by L'Auto, owner of the French daily, L'Équipe (ASO is a subsidiary of the Amaury Group, which owns the newspaper).
From newspapers, coverage of the tour evolved to radio, then television and on into the digital era.
"For us, moving into digital starts with having data to allow you to do that," Gray states.
In 2015, he says the focus of the technology team was on capturing information.
“How can we capture vital information, process it in real time and present through multiple channels - television, digital and social media?
“The idea of producing consistent information across multiple channels and having a common platform to do that, was a fundamental building block,” he states.
In 2016, Gray says, it was about how to tell better stories with that data. while engaging the audience.
His team was able to do this because of four things: new ways of capturing race data, new insights into race tactics, new ways of analysing race data and new ways of publishing race data.
“We used data to track the impact of weather conditions on the race,” he states. “We understand wind conditions and the impact on the riders.
“That has changed how broadcasters talk about the sport,” he says. “They talk about the race in a different way, because there is a whole raft of data available to them.”
“We went from publishing data to telling stories.”
How can we capture vital information, process it in real time and present through multiple channels - television, digital and social media?
This year, says Gray, machine learning technologies were used to give cycling fans across the globe an unprecedented experience of the event.
The technology team worked with ASO to develop a data analytics platform. This incorporated machine learning and complex algorithms, that combined live and historical race data to provide even deeper levels of insight as the Tour de France unfolds.
IoT enters the race
The Internet of Things brought viewers to the race, not the race to viewers, he says.
The first step was to apply IoT technologies, putting GPS transponders under the saddles of each bike. The data collected from these transponders is combined with external data about the course gradient and prevailing weather conditions.
This creates insights such as the live speed of riders, the distance between riders and composition of teams within the race.
This year, the IT team analysed more than three billion data points during the 21 stages of the Tour, a significant increase from last year’s 128 million data points.
"We have two years of historical information and used machine learning to predict what are the sorts of things that can happen," says Gray.
"We get the raw information from the sensors, we figure out the gaps and enrich that data. Then we add the prediction capability, being able to profile riders and understand their capabilities. Complex algorithms also analyse historical and live data.
“We ended up with 70 per cent accuracy in predicting the riders in the top five,” he says. These predictions were equivalent to the expert commentators around the race.
This year, they were also able to show live speed data on television for the first time in cycling history.
He looks forward to the year ahead as they continue to create more engaging experiences for fans, media, teams and the riders.
He says the technology team was able to track the growth of audience through the digital channels.
For instance, in the years spanning 2014 to 2016, the unique visits to the website grew from 30 million to 36 million. The downloads of the official app grew from 1.1 million to 1.4 million. Video views on their five digital platforms grew from six million to 55 million. On Facebook, they now have 2.4 million fans (an increase of 20 percent from 2015) and 2.6 million followers on Twitter (a 60 per cent rise from 2015).
Cloud and collaboration in the competition
The enhanced Tour de France solution uses a cloud-based, virtualised data centre that provides scale and means fewer people are required on the ground to enable the solution. The cloud also provides geographic flexibility, as it can be managed from anywhere in the world.
Gray says the technical team that worked on the 2017 tour numbered around 20.
Five were on the ground, but most of them team supported the race remotely, from Johannesburg, London, Melbourne and in the US.
“We use a lot of collaboration technology,” he explains. “Most of our platforms is cloud based, so we can support them from anywhere in the world.”
As to how he got into the role, Gray says throughout his career, he has always worked with data, including business intelligence and data warehousing.
Even before term ‘big data’ came about? “Exactly,” he says. “My roles have evolved all the time.”
“I don't think these days you can just be a technologist and be effective,” he says. “You need to be very focused on a lot of the business outcomes I am trying to drive. How can I use technology effectively to enable those business outcomes?”
He says some of the company's experiences on the tour are applicable to other industries.
Right after the 2016 tour, when they were evaluating the event, he says, “We talked to commentators, we talked to fans, we looked at the data sets we have available and we agreed on the most important questions that we were going to focus on.”
The team took an iterative approach to the development of the project.
We tested it early and continue to refine the models, he says.
“Even throughout the tour, we continue to do this,” he says. “Not only did the models improve, but also our knowledge of how they behave and the kinds of scenarios that we may need to account for.
“The way we use data evolves as well,” he says. “We continue to refine it through the course of the race.”
We all need to be data savvy
He also highlights the importance of diversity and in their case, “bringing together teams from different disciplines”.
"We built a cross-functional team for the project,” he says. These involved data engineers responsible for making sure the data was available, together with the machine. Along with these were the data scientists who were developing the models."
They were bringing all of the knowledge, the algorithms, the mathematics and computer science to create the models, Gray explains.
“We worked closely with ASO. They brought in commentators to help us think about what are the messages that will be particularly interesting for their people to talk about, and the information that is relevant to them.”
They also worked with domain experts. One of these included a cyclist who has raced at international levels, “who knows cycling inside and out”. The cyclist worked with the data scientists and engineers to develop those models.
The lesson from all these is, Gray says:
“Don’t get caught up in the technology, think about the questions that the data can help you answer. What are the exciting things you can learn by using and looking at the data? Then figure out how the tech can help you answer those questions.
“The reality is the world is becoming more data oriented, so we all need to be data savvy.”
As to what is next for the 2018 tour, he says, “the machine will continue to get better, the more experience that we give it.
“We will be looking at giving it more and more data to work with to continue improving its experience.”
This was how it worked out this year.
“We have the two-year historical data and we supplemented that with external third-party data, like weather and social media, and brought in race history data. The latter provided the background and context.
“So, mashing together all of that information allowed us to create models using machine learning to predict the riders who will perform well and how hard the riders are performing.
“Each day we looked at the prediction. Does that make sense? We talked to people who are experts and they said 'these are good predictions'.
“What is interesting for me is the machine learning model was able to make predictions that were equivalent to [those of] cycling experts on the ground.
“That does not mean you are going to replace those cycling experts,” he says.
“But if I take that to industrial applications, those experts, there aren’t many of them," he says.
With AI, “I can provide a similar level of expertise at scale, at low cost. That does not mean I don’t want them in my organisation.
“They [the experts] now have the ability to apply their knowledge and capability to a much larger problem or space.
“That is where I see a lot of opportunities as the AI models continue to develop and evolve,” says Gray.
“This is the first year we have done it. As it gets more intuitive, the quality of the predictions will improve.
“We also know that in its early stages of learning, the machine will make mistakes. Like a human learning, we make mistakes and then we correct those.
“We are making sure we have got the right controls in place around that before releasing the results into the world.”
It is not unlike teaching a child to cross the road, he states.
“You don’t teach your child to cross the road by saying,’ you go for it a hundred times’. You teach the rules and you put structures around to protect them in learning,” he explains.
“Over time, they have learned the rules and you can release them and you can say, it is okay for them to cross the road on their own.
“It is similar to developing machine learning algorithm, you are allowing it to develop its knowledge over time.”
Send news tips and comments to firstname.lastname@example.org
Follow Divina Paredes on Twitter: @divinap
Follow CIO New Zealand on Twitter:@cio_nz