AI and Time in the saddle

The meme was a joke because you keep repeating yourself without answering the questions I asked you (which is classic bot and/or trolling behavior so I was seriously questioning). Sorry if you took offense, it was not intentional.

I appreciate the doubling down on the 5 hours of training is all anyone needs. I strongly disagree… but would love to hear why you believe that other than “AI says so.”

TR may have data that suggests people perform better with more intensity and less volume. However, “having data” is not the same as proving a conclusion.

TrainerRoad has always presented itself as a science-based company, so it’s reasonable to ask: where is the published evidence showing that intensity alone is the most important factor in training outcomes?

Platforms like Strava and Intervals arguably have access to far larger datasets from real-world athletes. In theory, they could use AI to analyse that data and develop evidence-based training strategies. Yet they don’t claim to have discovered a fundamentally different training paradigm, and there’s little evidence that TrainerRoad is doing that either.

Meanwhile, one of the most widely cited bodies of research in endurance training—the often misunderstood 80:20 model associated with Dr. Seiler—suggests a different principle. The idea is to train as much as you can overall, with the majority of that volume at low intensity, while adding only as much high-intensity work as you can recover from while continuing to maintain high training volume. Among Norwegian endurance athletes, this often works out to roughly an 80:20 distribution.

So simply dismissing that approach because TrainerRoad doesn’t prescribe it, on the basis that they “have the data,” isn’t a very convincing argument. Without transparent evidence, it risks sounding less like science and more like marketing and trolling.

Maybe all the world class coaches and world class athletes are wrong and the optimal amount of aerobic training is between 4 and 6 hours per week.

That doesn’t seem very likely, though. Not really.

What data does it have on people who follow a more volume and a little less intensity approach?

This. TR is a business and that business is getting arses on saddles, preferably indoors with their plans/interfaces. There is a significant risk of selection bias in their datasets. Nothing nefarious just the fact that for most of us the point of this is the outdoor, largely unstructured ride. Something TR is not in the business of providing. Nor can it parse such data as it has coming from those rides, by dint of the fact that they are unstructered. There is no “They did X and the outcome was Y” in there. The dataset that the AI is ploughing is all the indoor ( or outdoor structured [edit] It turns out they can’t handle outdoors) rides that we have completed and it will naturally glom on to ones we finished and discount those we didn’t. While ignoring all the unstructured stuff. Maybe there is some clever stuff on the fatigue models in there but I would be surprised if you can extract a lot of meaning from such stuff for unstructured. Any TIZ data will be fuzzy; amalgamating traffic light starts for some sort of anaerobic stat on an edurance ride is nonsense. The more volume people are going to naturally fall out of the TR database. It takes a particular type to strap a trainer onto their arse for more than a couple of hours. My limit is 1.5. Hence the selectional bias in the dataset. The LSD crowd are not in there.

There will be some slicing and dicing where you amalgamate data sets from “rider types” to give an enriched personalized insight. Then action the plan accordingly but to claim that the AI is just right when it is more intensity less volume seems to miss the point a little.

I don’t do TR to become better at doing TR. I do TR to become better at riding my bike IRL. Riding IRL takes more than just “fitness”. It takes bike handling skills under fatigue, It takes feeding strategies, weather management, ancillary adaptions such as I mentioned. None of these are served by a max ride time of 1.5 hours, unless all you ride are track or criteriums.

I thought they up the intensity because they know in the main their customers don’t have enough spare time/willingness to train longer?

Didn’t they used to prescribe longer endurance rides, but found their users didn’t do them.

Initially they did leave the “Sunday Ride” as a user outdoor ride but then we the users complained of the lack of progress/ desire for an indoor option/ compliance to plan. Probably brought about by all the town sign sprints, hill yomping, testiculating and general mayhem of a Sunday ride that was nominally slated to just be an endurance pootle in the plan.

Well the B race was last Saturday and not the one before as I mistakenly posted. After dwelling a couple of days, I think it is safe to say that my concern here was realised. It did not go well. It was by no means the worst I have had, but since I have been recording the metrics it has some of my worst numbers. Hydration and feeding were ok. The course was flat as a pancake (250 m ascending, mostly bridges) there were no redline surfing efforts apart from a couple of unsucessful efforts to hook on to groups. Cardiac decoupling was a whopping 12%. My base fitness was simply inadequate.

Now I am not going to dis the AI on this as I suspect it simply did not have good data (Last year was a fallow one on the bike for me). It went live around 2 months before the event which is not a lot of time. I, contrarian that I am to these times, absolutely take responsibility for my own choices. I went with its recommendations as an experiment because it looked “easier” than putting in the miles. I’m mindful of the fact that I am now deep in the masters world and fatigue management is a priority.

YMMV

I’m not trying to be a contrarian but I’m guessing the majority of TR users are time crunched so TR’s data would be skewed toward lower volume being successful in increasing fitness. Practically every coach in existence says volume is a key driver of fitness. If you’re able to do more and are able to recover from the increased volume, you should always do more. Even Jonathan’s volume has increased to 10-14 hours since he started training for Leadville.

Indeed I made that very point up thread. For me the issue is the confluence of the calendar “stoplights”, the attractiveness of the AI workout selection; the skewness toward time cruched (core demographic) subscribers. And my own personal circumstances: Coming back from a fallow year, gunning for an early season race and being an old boy. The AIFTP seemed to make some convincing aguments on the calendar that I might want to pay heed to (fatigue management). It was my choice to run with it, even when the ancilliary metrics did not agree: My TSB was not as low as I was acustomed to managing, the ramp rates were a little low and my Garmin metrics were also largely not where I was used to. All that could be a real function of my immediate training history. It could have been an artifact in the AI, time-crunched approach. I seem to be falling towads the latter as an explaination. In any case I am going to go off plan now for the next few weeks to mitigate. I am just going to put in the miles after this recovery week.

Well the A-race happened on Saturday and not great either but not as bad as the B-race. My AIFTP is up for grabs this week, a recovery week, and was showing a 1.5% decline. Well, that was until I deleted one of my recovery rides due to travel.. It was marked as a yellow day anyway. Now we are looking at a 3.6% decline. This has all gone in the wrong direction since it kicked off for me. Notwithstanding the glitchy start for the new AI (-9% on inception then +11% on he first block) it has been mainly declining. Certainly feels it on the bike. I wonder if something else is going on here…