“An innovative platform that strikes the ideal balance between human coaching and AI-driven automation” is how Vekta’s founder and CEO Paul-Antoine Girard describes the new coaching and training system that launches today with endorsement from Jayco-Alula. The press release comes with a quote from the team's Eddie Dunbar, who won two stages at the Vuelta in 2024. He describes Vekta as a "game changer... we're not guessing, we're evolving".
Artificial intelligence is fast becoming part of everyday life, with ‘narrow’ AI used to speed up or augment a wide range of tasks. Recently Garmin announced Connect+, a premium paid-for plan that provides personalised AI insights. Strava has ‘Athlete Intelligence’ and dedicated cycling AI self-coaching apps have existed for almost a decade now. But Vekta appears to be doing it differently with an approach that combines traditional, human observation and experience-based coaching with digital, high-speed data analysis from multiple devices and sources. “We don’t believe in AI adaptive solutions where AI is going to replace the coaches,” explains Girard. “For us, AI is a tool for the coaches to save time, scale their coaching business and have more insights. Using AI for athletes to have more insights, direct feedback after the ride, but also as a tool for both of them, rather than saying ‘AI is the new coach’. We don’t think AI is the new coach – we know what it is to have a coach and what it brings in terms of psychology. So we don’t want to replace them, we just want to find the balance between the AI, coach and athlete.”
Girard has a degree in mathematics and economics, and a Master's in data science. His two co-founders Samuel Souci and Doron Israel have similar backgrounds in digital transformation and machine learning research. They’re three highly qualified individuals, but it’s unusual to launch your business with the announcement that you’re already in the WorldTour rather than working your way up and gradually gaining credibility via amateur cycling. How did they do it? “We were French, a lot of coaches heard at the time how AI could help them and they wanted to be on top of what might be coming. We did good studies in France at the best schools and French people were convinced by our abilities and expertise to do something. They gave us data, we signed NDAs… I cannot explain why they trusted us in the beginning!”
Girard continues: I’ve always been passionate about sport – cross country skiing in the winter and running and cycling in the summer. And I’m also passionate about mathematics, data science and AI so I’ve always wanted to build my company – sport with data. During my studies I met the other guys, Doron and Samuel, and I shared my will and journey with them. They wanted to jump on with me, so we started out as a school project during our Master's. It was a Master's thesis we were doing – how can we use AI to improve athletic performance?”
From there, their attention turned to the top. “We always thought, if we wanted to do something big in the sports industry we needed to convince the pros first. We needed to work in cycling with pro teams, integrate data science, AI products and performance and once we have that, let’s see if we can build a company on top of it. So we initially reached out to a few WorldTour teams asking to share their data about the riders… give us some data and we’ll do something with data science, predictions, not really sure what we could do but let’s start like this. And so the first team gave us some data, Arkea-B&B Hotels, we built something then we went to Jayco, UAE, FDJ Suez, quite a few teams and we developed a few tools for them.”
Girard explains how he and his colleagues moved not only beyond traditional coaching with the WorldTour teams at the beginning but also beyond the traditional metrics such as FTP and TSS. “We used their historical data to predict kilojoule expenditures for Grand Tours, predict for example Simon Yates for the Tour de France 2023 [above] – his performances on all the climbing stages climbs, his pacing based on his current fitness, performance in the stages before, and a lot of studies on the environmental effects of performance.” The emphasis, Girard explains, is on producing power towards the end of a stage or race. “In professional cycling everything is about durability at the moment. What is important for the pros is peak power output after a certain amount of kilojoules spent. Being able to reach the same power values after five hours. They love to compare power data with fatigue. So as an athlete you can have a few metrics like critical power, W’ [W prime] peak power – those metrics that make up your threshold, your anaerobic capacity and maximal power. Because we think this is the most accurate model available at the moment.”
Additionally, Vekta deployed AI to help understand the effect of temperature and altitude on individual riders and worked with Chris Froome for this. “From that we gained some credibility, gained some connections and finally went out to raise some funds and built a training platform on top of all this knowledge.” An incidental ambition Girard almost casually drops into the conversation is: “And we want to compete with TrainingPeaks because they’ve been around for many years and we think we can do something better.” OK – how easy or difficult could that be? “It’s more that TrainingPeaks has been there for years and all the coaches and athletes are used to using it, so it’s more of a challenge to show the value in switching the whole coaching programme from TrainingPeaks to Vekta and also for the athletes to understand that AI is there, but it’s not the way people usually expect it. In sports when we hear about AI, we think AI is going to coach you. This is not the approach we’re taking.”
Let’s take a look at how Vekta actually works. Some concepts in what can almost be called ‘traditional’ AI are used, such as digital twin technology, in which AI takes all the historical data from an athlete’s devices and apps and builds a digital replica in order to predict future responses. Girard explains: “What we focused on at the beginning was making the training data into something that machines can understand. Everything about training at the moment is doing some intervals, repeat those intervals at specific intensities and if you go to Garmin or TrainingPeaks the only thing they have is average power, normalised power and they can only take those global metrics about training data. So it was interesting for us to interrogate training in a mathematical way: What is training? What are those intervals? So that we can give it to an algorithm and then say to you, in the last five years of data these are all the sessions you did and training plans that you followed, and this is how those plans impacted your performance. We have a few performance metrics that we use so that we can say that, if we have a training plan with these exact sessions, this is exactly – hopefully – how you will react and perform. We can collate digital twins, play them and see where you are if we try different strategies.”
He says this differs from the way AI adaptive plans generally work: “At the moment there’s more of a human approach that is input into a computer and which tells the computer, ‘I believe in this training methodology… this is the amount of time that should be spent at VO2max, at threshold and this is the way we’re going to build the sessions. At Vekta we think the next part of it is having the data so that AI and algorithms can learn from it rather than using a human approach already proved but maybe that’s not going to work for you personally. Also we’re going to integrate some proprietary metrics to measure and quantify training load, some metrics that we developed with the WorldTour teams.”
Girard outlines how Vekta communicates the prescribed workouts from the interface to the pedals: “The big point is having training data, recovery data and some wellness psychological data as well. So you can link training, recovery, wellness and maybe soon nutrition. It all starts with the calendar. If you need to do a VO2max session you describe it and AI will generate the structure of it so that it’s sent to Zwift, Garmin, Wahoo, whatever you use. Making it easy to always have structured workouts. Why? Because we learn from those data.” And for the training itself: “We use the Vekta zones that are adapted to the current performance metric that we have. So if for example you train a bit more for the next three weeks, your critical power improves by 10 watts, your zones will to adapt so that it’s always accurate all the time. Since we are able to detect exactly what training you did in the past, for example how many sprints, what was the average power of those, threshold intervals etc… this is something that is not available anywhere else. We did a lot of studies with pro cyclists who were at the end of their career to understand how different their training was and how the sessions were built. It’s very precise the way we understand and represent training data.” Data from wearables such as sleep and heart rate variability are collected from Whoop bands and Oura rings, with Garmin health also to be integrated.
Obviously what’s not integrated is Strava, which recently restricted access to its user data by third-party apps: “That’s why we did not integrate with Strava,” says Girard. “We never did, because we didn’t want to be in a position where we rely on them. If they wanted to shut us down with their API we would be out of business. So we prefer to go to the data source, collecting data from Garmin, Wahoo, Oura, Whoop, whoever. They’re going to continue to continue to share the data so it's better for us to go to the data source. And also because Strava modifies the data that they share through the API, so we wanted real data.”
So what of the stated aim to become the biggest training platform on the planet? Does that mean making Vekta available in perhaps a different form for enthusiast, self-coached athlete in the future? “The way we see it is we started with the pros and we earned some credibility, built some awareness with them. Now we are convincing the coaches to get athletes on the platform and the next step will be to get more enthusiast athletes using Vekta. Building a good market strategy from the pros to the serious athletes to the more everyday athletes who may not have a coach but who can use a bit of AI to help them. But for this part it will take some time and we still think we have some development to do to bring value to them.”
Can an amateur sign up to Vekta even if they don’t have a coach? “We’re not going to build a training plan for them, at least not yet," says Girard. "Vekta is more like a tool where they can use the platform we have to build a training plan. So it could be more for a self-coached athlete who already has a knowledge of building their own plans. We’re not going to tell you, for example, tomorrow do this, this and this session.”
At launch, the pricing structure is for athletes only, starting at £17.99/€19.99/$21.99 per month; it's free for coaches. Or, a coach can decide whether to pay their athletes’ Vekta subscriptions as part of the coaching fee. Go to Vekta’s website for full details of pricing and for more information.
