Camino El Colegio 46, Santa Brigida, Las Palmas, 35310, Spain
        Contact the ABCC Today Call us on +34 696 313 209

The Future of AI in Endurance Sport

by Armando Mastracci

Artificial Intelligence (AI) applications have skyrocketed over the past few years. With the advent of large-language models (LLMs) like ChatGPT, AI has come into the limelight. Much of the hype and speculation is still beyond the capabilities of what these technologies can do. Artificial General Intelligence (AGI) which aims to replace humans and has the capability to *self-learn* is still just a dream. For the moment, AI needs humans to create them, teach them what to do and guide them towards creating real value. While a lot of the hype is click-bait, some of the hype is real, and ChatGPT and other technologies are delivering new capabilities never before imagined.

AI in Endurance Training

AI applications in fitness and endurance training have been around for a number of years, many making promises to be your “AI Endurance Coach”. These AI systems are largely based on encoding of rules: assign this training in this situation; assign this other training in this other situation. They codify general coaching and exercise physiology principles and make them accessible to athletes via an easy-to-use system where they can choose a goal, set availability and out pops a training plan. Some may offer ways to adapt plans as availability changes or there’s a change in goals.

These bear some similarity to services offered by real coaches but fail to deliver some of the essential elements of coaching. Coaches develop a relationship with the athlete, gathering more than just training data from them, interpreting sentiments and sensations using language and terminology that coaches often hone over years of experience. Every coach has their own unique method of interacting with athletes and that interaction and the alignment of their athlete towards goals and training interventions they can implement is what makes a great coach.


Some AI approaches involve the use of ChatGPT and other technologies to attempt to create an artificial dialogue with the athlete, mimicking a dialogue they would have with a real coach. This approach is new in the AI space for endurance training and it remains to be seen whether athletes will respond positively or negatively to this. As an add on to an existing training platform, it may offer some novelty initially and may provide some useful feedback for athletes that may have never had a coach. As with LLM-based systems, their interactivity is more a way to get questions answered rather than having a true conversation.

AI Coach: Your overall recovery is fair and your cardiovascular fitness is improving. Based on your Rate of Perceived Exertion after yesterday’s exercise, you can continue training your cardiovascular system today. This will increase your capillary density and improve mitochondrial function.

Machine Learning

Machine learning (ML) is a technique often used in AI systems where data is analyzed for patterns (i.e. “trained) which are associated with outcomes. For example, power data training history could be analyzed and associated with an outcome such as an associated FTP. ML could also be used to take user-input of Rate of Perceived Exertion (RPE) and with a set of encoded workouts and then based on that user input, predict the RPE of a workout that the athlete has not completed before.

ML, although a useful tool, doesn’t do what many coaches already do with data. Predicting a FTP using AI seems like a good use of ML technologies. From a coaching standpoint however, assessing data and establishing an estimate for FTP is something many coaches have learned to do themselves through the use of existing tools or analysis of data directly.

Coaches use their athlete’s comments to assess their readiness to move to the next level of intensity in their training.

When in doubt, coaches will incorporate testing as part of their athlete’s training to reinforce or provide the necessary feedback they need on their athlete’s progress. What a predictive AI for FTP does well is perhaps allow a coach to possibly skip a test for their athlete. What a predictive AI for FTP doesn’t do for the coach is offer an explanation as to why their athlete’s FTP may have changed. A good coach will comb through their athlete’s data, from power, to heart-rate, to documented RPE to see where the source of the changes are due to and adapt their athlete’s training accordingly.

AI-based Training Plans

Other applications of AI techniques lend themselves well to the autogeneration of training plans for athletes. As described, many times these AI tools encode aspects of coaching and applied exercise physiology as rules. Example rules could be how training should be periodized or the adjustment of training based on athlete feedback / readiness data. These rules enable the system to pick training for athletes based on their goals, availability, current training status, readiness, etc.

From a coaches perspective, it does overlap with what a coach will offer their athletes. However, a good AI-based training system can make a coach’s job easier as it can handle the complexity of juggling availability, athlete choices and alignment towards a goal. However, many commercial AI training platforms will offer a turnkey approach for athletes which undercuts some of the services offered by coaches. As these platforms mature and become better automated and integrated into the athlete’s training systems, it will disrupt how coaches will interact with their athletes and they will eventually need to adapt, offering services in concert with these AI training plans and help athletes through the maze of information and options available.

Xert Breakthrough Training

At Xert, with the latest development of Forecast AI (currently in beta) we’ve taken an agnostic approach to training. This applies to the physiological exercise principles that we use as well as to the application of our method by athletes or coaches. It’s more of a tool rather than an “AI Coach” where it behaves like a coach, following a particular training philosophy or applying established coaching best practices. Xert is driven by what information lies in the data and to make that information accessible and actionable by athletes and coaches. This is a key difference between Xert and all the other platforms.

This type of AI is the most challenging since it aims to discover what the athletes did that affects the guidance and predictions rather than simply providing predictions and basic sentiment analysis using ML techniques. Users of the system will get insight into how a plan should be laid out in ways they would not be able to do without it. Xert doesn’t attempt to do what coaches do. It’s doing something completely novel that benefits coaches and athletes alike.

For example, Xert introduces the concept of a fitness signature which is made up of 3 fitness variables rather than just FTP alone: FTP (or just Threshold Power or TP in Xert terms), High Intensity Energy and Peak Power. These 3 values align with each athlete’s aerobic, anaerobic and neuromuscular systems respectively. Xert tracks the strain (as Xert Strain Score or XSS) placed on each of these from historical recorded activities and quantifies them. Then each of these is used in a separate Impulse-Response model to track individual fitness and freshness in 3 dimensions. All this is complex mathematically and not something that an athlete or coach could do simply with their own spreadsheet or data analysis platform. Xert then uses this information for each athlete to look for patterns between training (strain on each system) and fitness changes (how the fitness signature changes as a result of the strain on each system). Xert examines the athlete’s data to look for information that reveals why improvements occurred, not just what the improvements were.

With the new Forecast AI, having learned about the athlete, it applies the patterns obtained using an AI-based optimization algorithm to map out a plan on how the training for an athlete needs to be applied in order to get progressive overload to all three systems towards the outcome desired – an outcome could be to reach a certain FTP or power capability, or it could be towards having enough durability for a particular event, or both even. It is this insight and assistance that coaches and athletes would greatly benefit from and something unique to Xert.

Coaching with Xert

One of the unique aspects of Xert that is particularly useful for coaches is how Xert lays out the training needed but doesn’t prescribe the actual workouts to do. Coaches get to choose what training their athletes will do. Xert will then adapt the athlete’s training accordingly (much in the same way coaches will juggle around their athlete’s schedule). The new Forecast AI feature offers coaches a way to accommodate their athletes and assist them with planning what training they can do rather than using something that is tossed out by another plan that they then have to figure out.

Coaches can prescribe workouts they often will prescribe for an athlete they’ve been coaching and see how that has influenced their success. Using “what-if” scenarios, it can also offer ideas on other approaches they could take that they didn’t realize. Again, Xert is a tool that shines a light into aspects of endurance training coaches and athletes have never had before.

The Future of AI in Endurance Sport

Athletes will always need help setting and achieving their goals, whether these are the modest goals of a beginner cyclist to having a goal to win the Tour de France.Athletes will turn to coaches to help them through the plethora of options, especially when these options become more complex. Not all athletes are prepared to invest the time to learn about all the approaches they could take towards training and many don’t have the capacity to appreciate all the elements – from training, to diet, to rest, to mental focus, to managing illness, to having confidence in oneself, to race day preparations. AI tools will help to some degree but the entire needs of an athlete is far more complex and a real-live coach that brings experience, guidance and motivation will most always be an essential part of the success of an athlete.

Author’s profile

Armando Mastracci is the founder of Xert Breakthrough Training. Armando founded the company in 2015, after having developed software that uncovered patterns in fitness data that had not been previously discovered. He is currently engaged with the University of Calgary in bringing these discoveries to the scientific community. He is a graduate of the prestigious Engineering Science program at the University of Toronto and has spent over 30 years in technology related industries.

The article was originally published in The Journal of Cycling Coaches #02/24. To whole journal is accessible to ABCC members. For further information, please take a look at

Contact: Hartmut Hübner,  ABCC Communications,

Camino El Colegio 46, Santa Brigida, Las Palmas, 35310, Spain
Contact the ABCC Today
Call us on +34 696 313 209