I’ve been seeing some ads for aiendurance.com which generates training plans based on AI. I’m not sure the nuts and bolts but has anyone come across training planners (not plans) that will generate daily workouts based on your goals, prior training stress (swim/bike/run), and most importantly, factoring in Daily Life Stress (sleep, HRV levels, etc)? For example, if you have a planned 2x20’ at FTP workout session but the planner sees you slept like shit the past two days and have corresponding bad HRV score, it backs the workout off to something more like a recovery session.
TrainerRoad does this for triathlon plans. Garmin does this for running-only. TrainerRoad is using less data, and it is built on a static plan as a starting point. Garmin uses everything but is only focused on a single discipline (running) for now. Suunto is also doing it (using the same Firstbeat analytics as Garmin), but it is a little less robust than Garmin right now.
My guess is that Firstbeat will emerge as the market leader. They already have a fairly mature side of their business giving guided analytics to teams & coaches to refine training. And they are building all the device-based analytics that companies like Garmin & Suunto are using. I do not think any other platform is using HRV, stress, sleep, etc., like Firstbeat.
I’ve been seeing some ads for aiendurance.com which generates training plans based on AI. I’m not sure the nuts and bolts but has anyone come across training planners (not plans) that will generate daily workouts based on your goals, prior training stress (swim/bike/run), and most importantly, factoring in Daily Life Stress (sleep, HRV levels, etc)? For example, if you have a planned 2x20’ at FTP workout session but the planner sees you slept like shit the past two days and have HRV score, it backs the workout off to something more like a recovery session.
I have yet to see any good, solid metric of your body’s ability to accept training stress at any given moment. And I’ve really, really tried to find one. Things like HRV, sleep levels, TSB are interesting tools. But I would not trust them to automatically scale my daily workouts. Maybe as a “suggestion” - “I see you went on a bender last night and got in at 4AM, would you like a recovery day?”
Maybe the closest is TrainerRoad, where you can decide yourself to scale back a workout, and then let the system know why, and in some mysterious black box way it might take that into account going forward.
I looked at TrainerRoad but it looks specific to cycling workout suggestions and doesn’t seem to factor previous runs/swims along with any health metrics. It is cool to see some inroads have been made and was wondering if there was another platform/service that has taken it further than what TR has implemented so far
That is correct, TrainerRoad is heavily cycling-centric.
And you’re right, it doesn’t (yet) use health metrics (other than self-reported through a post-workout questionaire) as input, though I think it’s under investigation.
I’m not aware of any service that provides the full extent of what you want. What you describe is sort of the Holy Grail though. It’ll probably be a while.
TrainerRoad does adjust run and swim workouts. But it is limited in the information it can use.
Ditto to trail’s comments. I think that this will most likely emerge in the Garmin ecosystem. In order to really do it well, the AI training platform will need the additional data, like HRV & sleep, that are not in today’s workout files. So, the only companies poised to build this are the watch makers. Garmin owns Firstbeat, and this is exactly what Firstbeat has dedicated itself to. So, Garmin needs to buy something like a TrainerRoad or Training Peaks to integrate everything together and build the one platform to rule them all.
See the Garmin 955 watch that was just announced (look for DC Rainmakers analysis). It seems like Garmin is getting serious about training recommendations. I could almost see myself using it in a few years (except it would be telling me to rest more and train less so I would just ignore it).
Endure iq seems to be another AI app but fairly new.
As with all these things they are only as good as the data inputs. Look at whoop and DC Rainmakers review about accuracy. I’m a long time trainerroad user and the new adaptive training is cycling specific and pretty, well, meh. The levels you achieve drop off very quickly, so say you’ve done a level 7 threshold session then don’t touch threshold for a few weeks, it drops you back down, even though you know you comfortably do the same level session or higher/harder.
Being self coached I would love an app to tell me what to do, but I just can’t see it being accurate enough. The target market for these apps are the self coached who are generally spinning a lot of plates and struggling for time, and probably suffer more with the hrv scores. If you’re consistently told to do easy sessions you’d quickly get detrained. Sometimes those hard sessions go very well even if the metrics are telling you otherwise.
The target market for these apps are the self coached who are generally spinning a lot of plates and struggling for time, and probably suffer more with the hrv scores.Are you sure about this? Firstbeat’s clients are professional athletes and teams. Their analytics and coaching assistance engine seems to be way ahead of anyone else in the market. I believe that Firstbeat’s and Garmin’s target market for these watches and analytics are high-performance athletes.
Has anyone tried or used Athletica AI…same premise in using predictive technology to adjust your ongoing plan based on subjective markers and I am assuming some sort of fatigue / fitness relationship.
Created by Prof Paul Laursen who is the Prof in the Prof and Plews relationship.
I think this style of programming will be the norm in a decade as the wearables we use get better at detecting true physiology
The target market for these apps are the self coached who are generally spinning a lot of plates and struggling for time, and probably suffer more with the hrv scores.Are you sure about this? Firstbeat’s clients are professional athletes and teams. Their analytics and coaching assistance engine seems to be way ahead of anyone else in the market. I believe that Firstbeat’s and Garmin’s target market for these watches and analytics are high-performance athletes.
Not sure, no! I agree analytics for the pointy end, to assist the coaches and athletes, but not in actually prescribing workouts which is what some of these apps do and I think what the OP was asking? Professional and serious teams/athletes have first rate coaches that interpret the data in conjunction with athlete feedback, and won’t take it as gospel. I don’t think AI has reached the point at which it can beat human experience and insight (yet). I think some of these apps are still targeting those who don’t have a coach but want to be told what to do.
Does TriDot count as AI? I found it had too many rough edges, needed refinement. Potential, though.
Has Garmin added this feature to their actual training plans or is everyone referring to the ‘recommend workout’ that comes up on the Run app?
I used Run w/ Hal app to train for the Jacksonville 15k. Not very sophisticated but it did seem to review my completed paces as part of building future workouts.
Has Garmin added this feature to their actual training plans or is everyone referring to the ‘recommend workout’ that comes up on the Run app?
I used Run w/ Hal app to train for the Jacksonville 15k.Yes, this is part of Garmin’s recommended workout analytics. For example, if you had a new Garmin and entered your Jacksonville 15K race in the calendar, then Garmin would give you daily recommended workouts that target that race. Further, it uses all your body’s metrics and recent activities to scale each daily workout so it is optimized for your performance.
Interesting, I didn’t realize my ‘old’ fenix 6 was lacking something for training plans.
Interesting, I didn’t realize my ‘old’ fenix 6 was lacking something for training plans.The Garmin Forerunner and Firstbeat teams worked closely together to brainstorm and develop a lot of new and dramatically improved analytics. They are rolling these out to all the latest generation of watches (255, 955, 945 LTE, fÄ“nix 7, epix, etc.) but the prior gens are not getting them (945, 745, fÄ“nix 6, etc.).
As others have said the health metrics are really not reliable enough as things stand. Too many people have set PB’s with poor metrics and had awful training sessions with good metrics. I’ve heard a few pro coaches on podcasts saying exactly this, that the health metrics are interesting but just part of the overall picture and you probably shouldn’t modify training on them alone. (There is also the placebo/nocebo effect - does your watch telling you are feeling good/fatigued affect performance regardless of reality?)
Even if the health metrics were perfect there are still problems. Firstly how do you even scale a workout? Let’s say I’m only at 80% so AI decides to scale the intensity my planned vo2 max workout down. Well now it’s no longer a vo2 workout and I’m doing threshold intensity and getting a different stimulus/adaptations. Let’s say I’m so fatigued the AI says no intervals just easy zone 2 today. Great but to keep to my weekly load I’m going to have to do a longer session (e.g. my 45m interval session now becomes a 2 hour zone 2 session) or create more load later on in the week which will most likely have to come from additional duration. Even if AI recognises this most of us don’t have the freedom to massively shift our time around at such short notice.
You also still have to prove that this is more effective. For example I don’t have AI training and am fatigued but struggle through a hard workout, then have a great sleep/recovery that night and am fresh the day after. Am I in a worse position than if I’d had AI training and done an easier session? I might even be in a better one!
As much as I love the science and data driven training part of me thinks this isn’t so complicated. Simply go off feel. If your feeling bad and then start training and everything feels harder than normal and heart rate is higher than expected cut the session back appropriately.
Let’s say I’m only at 80% so AI decides to scale the intensity my planned vo2 max workout down. Well now it’s no longer a vo2 workout and I’m doing threshold intensity and getting a different stimulus/adaptations.
you can scale a vo2 session down by volume but maintain some vo2 intensity stimulus. whether that is the right thing to do when the athlete is fatigued is debatable
Let’s say I’m so fatigued the AI says no intervals just easy zone 2 today. Great but to keep to my weekly load I’m going to have to do a longer session (e.g. my 45m interval session now becomes a 2 hour zone 2 session) or create more load later on in the week which will most likely have to come from additional duration.
is it necessary/desirable to maintain weekly load if the athlete is excessively fatigued? maybe your volume is too high and you’d be better off to ease back a bit
As much as I love the science and data driven training part of me thinks this isn’t so complicated. Simply go off feel. If your feeling bad and then start training and everything feels harder than normal and heart rate is higher than expected cut the session back appropriately.
in principle i agree, however many of us are not good at listening to our bodies so some objective metrics/advice are helpful. even with a coach, they have limited ways of knowing how you’re really feeling
its a really interesting area. i haven’t seen anything yet thats really there but i think there will be big steps forward in the next year or 2, if platforms are willing to share their full datasets
Let’s say I’m only at 80% so AI decides to scale the intensity my planned vo2 max workout down. Well now it’s no longer a vo2 workout and I’m doing threshold intensity and getting a different stimulus/adaptations.
you can scale a vo2 session down by volume but maintain some vo2 intensity stimulus. whether that is the right thing to do when the athlete is fatigued is debatable
Let’s say I’m so fatigued the AI says no intervals just easy zone 2 today. Great but to keep to my weekly load I’m going to have to do a longer session (e.g. my 45m interval session now becomes a 2 hour zone 2 session) or create more load later on in the week which will most likely have to come from additional duration.
is it necessary/desirable to maintain weekly load if the athlete is excessively fatigued? maybe your volume is too high and you’d be better off to ease back a bit
As much as I love the science and data driven training part of me thinks this isn’t so complicated. Simply go off feel. If your feeling bad and then start training and everything feels harder than normal and heart rate is higher than expected cut the session back appropriately.
in principle i agree, however many of us are not good at listening to our bodies so some objective metrics/advice are helpful. even with a coach, they have limited ways of knowing how you’re really feeling
its a really interesting area. i haven’t seen anything yet thats really there but i think there will be big steps forward in the next year or 2, if platforms are willing to share their full datasets
Yep, you could just do less intervals for a vo2 workout. However there is some suggestion time above 90%HR is optimal stimulus for vo2 max adaptations. It tends to be the later intervals where the HR is most elevated (if I was scaled back to 3*4min intervals at 120% of FTP my time above 90%HR would be pretty low, others may vary), so the question becomes are you actually still getting sufficient stimulus for target adaptations with lower reps? Or are you just adding extra fatigue for relatively little or even no gain? I don’t have the answer, but it is just an example that scaling workouts might not be so simple.
If sick or sustained overtraining, yep cutting down weekly load may be a good thing. If you’ve just had one bad night of sleep or a single day of slightly overreaching and can realistically “catch up” there seems little reason to lower weekly load - but perhaps moving sessions around may be optimal. There is also the issue of how the load is made up - I don’t think I’m alone in saying a load with 90% zone 2 is a lot easier to recover from than the same load with a ton of high intensity work. AI could make this switch to a lower intensity workout, but it’s going to require a way longer duration which is not particularly useful for those with time limitations.
HR alone is a pretty good indicator. If your resting HR is up, HR during exercise is up, and your not feeling good that’s probably enough to say pull back. I suspect the people that ignore these things to push on will be just as likely to ignore AI.
I agree it’s very interesting area. I just don’t think we are currently close to an AI optimising training. That doesn’t mean there’s not things we can learn and take away, but a ton more research is still needed. Being able to reliably predict performance based on health metrics is probably the first step.
PKvitality are bringing out a real time lactate monitoring device, which could mean real time training modification i.e. you increase/decrease pace continuosly to maintain correct lactate measurement which would be an interesting training experiment. (Of course first thing will be to see if it’s skin based lactate measurement is actually accurate and reliable!)