This is about the only science of decoupling that I’ve found:
1 Like
I’m not much of a runner but installed the app anyway to have a look. It’s very slick, and I can easily see this style of app transferring to cycling. But behind the scenes I assume there’s just a complex series of generic rules for the adaptive planning rather than anything smarter that is actually “learning” what is working for the particular athlete.
Well I’m sure I’ll get pummeled (and I’m heading out of town for a Ragnar Ultra Relay today), but as I’ve always been a fan of TR and Nate + Crew I thought I would jump in and try to answer questions. I’m the lead(ish) on the TP Adaptive training effort. I’ll try to go through and respond to people before I head out but first some general stuff.
Me: Ryan Cooper
Roles: Founder Best Bike Split, VP at Training Peaks, Math Nerd
Background: In 2012-2013 my Business partner Rich and I started a company called Optimized Training Labs (OTL) which was a virtual coach / adaptive scheduling concept for triathlon. It had some mild traction but we had some fundamental flaws in the way we wanted to bring it to market. The biggest being that we were trying to be the experts as opposed to working with the experts (Hal Higdon, TrainerRoad (Coach Chad), etc.). By 2013 we had launched BBS in beta, originally as a funnel to drive traffic to our virtual coach. Needless to say BBS took off and after being able to kind of ride the fun of that growth to multiple tour teams, gold medals winners, Ironman Champs, and of course all those looking to make course cut offs and cross the finish line with smiles on their faces, I still had an itch to do build something in an area I really care about and it’s rare to get a second chance to create something.
I’ve always thought that if it was done the right way Adaptive Training (not really full on virtual coaching) could be extremely useful, especially people like me who are busy, have multiple kids and commitments making following a strict schedule difficult. The philosophy I prescribe to is that don’t fix what’s not broken and try to build and enhance upon that foundation with an eye (and tech hooks) into future functionality (be it AI inputs, human in the loop, micro or macro adaptation loops, etc). Ultimately we wanted to build a platform where the partner coach / persona could choose the how and at what level to allow the adaptive nature of the plan. Speaking of busy I’m off to drop the kiddos off at school, but will be back to try to help answer some questions.
As a quick side note I’ve been using the Hal app to train for the last 4 months in beta testing. Was happy to take the Ragnar Relay Black Loop overall with my buddy (31.5 miles of pair trail running split over a couple days) by just over a min. We’ll see if the fitness can carry over to a 4 person 125 mile race tomorrow!
7 Likes
I know you don’t want to disclose future plans, etc…but it would be nice if you could reveal which ecosystems you’ll be prioritizing gathering data from. For instance - if rest and recovery data is tracked - will you pull from Apple, Garmin, Whoop, or others? What will the priority list be? A prioritized list of which you’d plan to implement for each major area (recovery, diet, weight to name a few) would be helpful to know ahead of time, particularly with the holiday season coming up
So behind the scenes is a set of different things actually. There is a set of complex optimization rules as a starting point but those are flexible to a persona (Hal in this case, but would be different for different personas / periodization schemes etc). The big thing is we wanted to the plans to work for your daily life and availability constraints, and if you miss a workout for it it to handle the impact of that not just with a narrow view but within the scope of the entire progression towards your goal. We don’t let you move workouts wherever you want because Hal would tell you not to, but missing a key workout may or may not warrent a reschedule (depending on the holistic view of your past and future training).
Other adaptations in place or in work are also persona driven. Some coaches might care about sleep trends where others want to make decisions based on HRV or subjective so the framework for those micro trends can be accommodated. Thinking about bigger picture you have macro adaptive (things like compliance to the plan, fitness progression, etc). The base structure is there and the backend API to start to incorporate these things either internal to the platform or could be driven by an outside source (say HRV4Training or Whoop…).
3 Likes
We see adaptive as spreading structured training to a wider audience by helping to remove roadblocks to entry (Step one get more people out the door errrr on the trainer). IMHO it can not and should not ever replace a good coach but could definitely help automate tedious tasks and remove some of the span of control issues that coaches face as they scale. Let machines do what they do well so we can free up humans to do more actual coaching.
1 Like
I’ll throw out that we really do mean it when we say we aren’t out to replace coaches. The coach is always the best and we have no plans to do a TP bot. We have always been built by coaches for coaches with the goal of connecting athletes with the experts. Being able to scale coach personas was definitely at the forefront when we built the underlying adaptive platform and API.
er hi 
That’s fair! I have the perspective of somebody who would never hire a coach, so a platform like TrainerRoad is my substitute. Adaptive training could close the gap between the uncoached and coached experience even further, so I have even less incentive to maybe explore what I’m missing.
That said, since I’m in the “will never hire a coach” bucket, the coaches using TP have better things to do than convince me otherwise. I can absolutely see how an adaptive system would take a load of busy work off coaches so they can do the things that require real expertise.
2 Likes