The story (& some of the tech) behind the Ithlete Coaching Platform

About a decade ago now, I started the daunting process of learning to code. My motivation, from the get-go was to develop the skills to be able to better handle, & make sense of the mass of training data I had accumulated over the previous decade of coaching endurance athletes. That decision eventually led me to the inevitable question that I think many of us face - OK, now that I can access all of this yummy data, how can I actually use it to make better decisions as a Coach?

This question, ultimately, led me down the path of studying Machine Learning &, more specifically, Prescriptive Analytics, i.e. using data to answer the simple question - "What is the best prescription that I can give the athlete based on the data at hand?" As it turns out, this is a much more complex question than it might first appear! The trickiness comes from the fact that we're not selecting the best option to lead to a high performance today or tomorrow, rather the day that we want the best performance on lives many weeks, sometimes many months into the future. Furthermore, in order to arrive at this highest level of performance, we'll have to be OK with lower levels of performance along the way. These long term optimization "1 step back to get 2 steps forward" problems are mathematically tough to solve. As it turns out, our bodies are quite complex :) Additionally, this drive for higher fitness on race day must be continually balanced against the risk of overdoing it today. The athlete must get tired, but they can't get *too* tired or it's "Game Over", so we have to first answer the important question - how tired is *too* tired for this athlete?

Undaunted, I went about putting all of my data to the task of solving this complex problem over the better part of the 2010's. A part of which included taking on a role as CTO for a start-up that was looking to apply this "new, hot" field of Machine Learning to sport. Towards the end of the decade, I'd arrived at a workable solution that consistently converged on a computationally optimal solution to training prescription for endurance athletes. I set about employing this solution to assist me in making decisions for both the the professional & top amateur athletes that I was coaching at the time &, after some refinements, I was impressed by how useful it was in guiding me to that "just right" workout progression that treaded that optimal balance of risk and reward.

Shortly after I resigned from my role as CTO of the ML start-up, Simon from ithlete approached me with a new and interesting task: What if we took a different approach & used the mass of heart rate variability data on ithlete to build models that use the individual athlete's recovery data to inform even better, HRV guided, prescriptions? And so, the Ithlete Coaching Platform was (excitedly) born!

So, let's dive a little deeper into how it all works..

The greatest strength & the unique differentiator of the platform is that we utilize a HRV guided approach to planning the training. This approach has been studied extensively & generally shown superior to "traditional training", however no commercial solutions that really utilized this approach were previously available. As one of the longest standing HRV apps around, ithlete has a massive amount of HRV & wellness data that can be made use of in making training decisions in accordance with the recovery state of the individual. In the platform, we make use of not only this heart rate variability, but also resting heart rate and morning wellness data to inform our model of how generally "fresh" or recovered the athlete currently is. Importantly, this assessment is made specific to the long term data from that individual, i.e. every individual has a recovery model specific to them &, as more and more data is collected over time, this model gets to know you, as an individual, better and better.

The other side of the model comes from our partners at Training Peaks and Garmin. You can sync your ithlete account to either of these providers to give us access to your workout data so that we can not only assess your current "freshness", but also your current "fitness" as an athlete. We pull all of your session data from your workouts to get an accurate assessment of both the training load that you are currently under as well as what that training load is doing to your fitness. Similar to the recovery model, over time, our fitness model gets to know you as an individual athlete and is able to learn what makes you better.

The tricky part comes next - we want to use both of these models to come up with a sequence of workouts that is designed to make you better without taking your recovery too low and putting you at risk of injury or over training. For this, we send both your fitness and freshness data to our optimization engine and we have it run many "what would happen if?" simulations: What would happen if I gave this athlete a long ride today and then a hard interval run tomorrow, knowing what I know about his/her current state of fitness and fatigue? What would happen if I gave the athlete a rest day today? Would the increased freshness be worth the trade in fitness over the long term, knowing what I know about this athlete's current recovery state & their dynamics? These simulations are run with one objective in mind - to come up with the sequence of workouts that will give the athlete the highest fitness on race day without exceeding a safe level of fatigue. We run these sequences all the way to the end of the season and then return your individual best one to you, the athlete.

Amazingly, our Cloud engine is able to run this computation in the ~4 seconds it takes for your app to say "We're just spinning our A.I. wheels" :-), as the computer is coming up with the optimal options for you! We then serve the top 3 options to you on your app, giving you the athlete final call in which of the top 3 options you are feeling that day. Importantly, depending on which route you take, which option you select, the "branch" for tomorrow's options will change. Consider it your very own "Choose your own adventure" :) i.e. if you pick a long ride for today, tomorrow will offer different selections than if you choose an interval ride for today. The system has a memory of what you've done to this point and will ensure that we choose actions that keep you balanced. You can also "nudge" the workouts in the direction of a specific event by nominating a "Sprint", "Olympic" of "Half/Ironman" specialty within the user menu. We aim to build this out over time, with run only, bike only and even a row only model for single sport athletes.

From this simple concept of moving away from the usual fixed, traditional "training plan" to an adaptive approach that stays flexible & tweaks the plan daily in accordance with your measured recovery & your feeling/preference, while keeping an eye on maximizing the long term performance goal, we have found there is a surprising amount of power, both in terms of avoiding those "iffy" sessions that potentially lead to injury or illness, while making sure that the key workouts are both ideal for you & placed on the days that your system is set up to get the fitness benefit from them. We're excited by this whole paradigm shift in training and excited to keep developing it. We hope you'll join us in this new approach to training! We kinda think it's going to be big! :)

You can get more info on the Ithlete Coaching Platform here

Train Smart

AC

  Tweet