This is probably a FAQ… but just curious… By performing a TR-generated plan, and if in general, you mark those workouts as Easy or Moderate… are those AI-generated and adapted workouts expected to bump up your FTP every 4-6 weeks? Using the FTP AI feature which goes back and views those workouts?
What you are asking is largely difficult (if not impossible) to answer with any certainty. There are many TR plans for starters, and not all are aimed at FTP increases. Even those targeted at FTP increase specifically may not give those results for everyone, while other plans may do so for other athletes.
Athlete history is a huge factor along with all the other details like nutrition, recovery and such that make the picture far more complex to define with any practical expectation. Sure, FTP is one fitness aspect that TR aims to increase in some cases, but it’s far from the only aspect to the fitness in their broader plan picture.
2 Likes
Ok that makes sense.
But IF an athlete executes all the prescribed workouts well, and flags them as easy… then those other factors don’t really matter. The AI engine adapts the workouts to optimize results, and the AI FTP feature recognizes that the workouts were “easy” and done well… Then it seems that the athlete should “improve”.
I get that FTP isn’t the only energy system or capability that TR targets. Which I love. Maybe a plan could explain that IF you execute this plan, TR’s AI engine is focusing on improving X, Y and Z? Or maybe it already does I just don’t know where to find that detail.
Thanks!
That’s a lot of ‘ifs’.
- I contend this situation is only temporary in your hypothetical stating it could/would be longer lasting.
We don’t know specifics, but presumably TR is aiming for a desired outcome related to power and RPE (the survey rating) in an effort to build fitness in whatever measure(s) they are tracking.
- Assuming a regular workout (not some recovery or lower level Endurance ride), if a rider rates workout as “Easy”, AT will likely increase the progression for the pending workouts in that situation. It will continue to do so until it gets Moderate, Hard or Very Hard ratings as a result of increased PL and AT changes.
What you are asking for has been mentioned & requested many times.
- How much will my FTP increase if I do plan X?
Ivy even joked a bit about it on a podcast a few weeks ago. It’s a question on par with “how long is a piece of string”, that has many answers.
TR has progression defined behind the scenes in their plans. It used to be “set” and essentially the same for everyone. In a way, it’s like ‘averaging’, fit for some but not everyone and flawed as a result. But with the addition of AT, the results are even more variable than before with that ‘simple’ solution.
In the past, Nate has mentioned the desire to offer some form of projection related to FTP and other fitness measures. Sounds great in concept, but I suspect the reality is far more complex than any of us would like to admit. Not to mention, even if they nail it for a majority of people, there will be some where it fails and I can only imagine the pushback.
At one point, the mobile app had some level of “this is what you will gain with this plan” type of graphing, but I checked and don’t find them. What remains are the text descriptions which are about as good as I think is practical vs offering projections that may be misleading.
3 Likes
I might screw this up but here goes… Tim Cusick was recently on the Empirical Cycling podcast, and essentially he thought this was a question that AI/ML/BigData could be expected to answer in the short term. Something like “if I did a certain block of training, what results might I expect.” Keep in mind we are talking probabilities and mapping onto regression curves, so I think it should be viewed more as high-level guidance on selecting training blocks to go from where you are now, to where you want to be in 3-6 months. There is more to discuss than that, however at a high-level that makes a lot of sense to my ears.
5 Likes
Yeah, I do think there is potential here but it will require some notable asterisks IMO. As you note, the duration of coverage matters here, where shorter terms are likely more manageable than longer ones.
In a way, TR may be well poised considering the apparent wealth of data they have. Maybe that is part of why Nate mentioned it in the past? Maybe they can/will blend background goals along with actual training results from their stores.
At this point, this may well fall into the “Feature Request” tag, since nothing really exists now. I searched and don’t see one at this point, so this might be the first official request that I have seen on the forum at least, despite being mentioned many times over the years.
1 Like
Correction, I did just find one related but it is in a different category at present. It includes lots of comments from Nate too. But I am still reading to see if it crosses over to this thought.
Upon further reading I don’t think this is related, doh! On with the searching…
ETA: found a related one from a while ago:
1 Like
The notion it takes a mountain of data is an interesting topic to debate. Not here and now. Just saying.
I never said “that’s what it takes…”. I referenced it as one possible resource in addition to other resources.
Ok, the notion it takes a mountain of data to be well poised. I think we don’t need to parse words. The point being the continuous big data drum beating vs what’s actually required. Another topic, another thread.
1 Like
There are two ways to think about this. I think the other commenters have thought about it the practical way, but not the way to directly answer the question. That is to say, their answers are more meaningful where mine is more about the ML.
To me, this answering whether an individual will actually make improvements. That is much harder to determine. However, I think the question
This to me is more about the ML. I would argue that if everything else the model takes into account stays the same (I have no idea if takes in things like volume and TSS), you have 100% compliance, and you mark every workout as easy or moderate; your AI FTP should increase (at least it should not decrease). I would also argue that if your AI FTP went down, it is a sign that their model is overtrained.
Said differently, I could arbitrarily lower my FTP in order to meet these conditions. If the model further predicts my FTP went down after marking everything easy or moderate, something is wrong with the model.
Just to be clear, I am not commenting on whether actual FTP would increase, just the models prediction of FTP.
2 Likes
Agree - assuming the AI interprets the “Training Plan” as being able to raise the FTP with perfect compliance. If the Training Plan was poorly designed this could happen, no?
1 Like
This is super difficult from a product perspective: imagine if TR forecasts a range of FTPs and you fall outside of it (on the lower side) — a lot of people would come here and complain even though perhaps it has nothing to do with TR. Life stress has a huge influence on one’s training.
Also, the more experience you get and the higher you get, the less gains you will see. I work hard every season to gain back 25–30 W, and forecasting those small gains is probably harder than bigger gains. Am I disappointed? Nope. It is just part of being close to my personal limits with the lifestyle I lead now.
5 Likes
Except to a certain extent TR already has the data to deliver this. If it didn’t, AI FTP and adaptive training wouldn’t work. These algorithms need to be taught on data (which they have) to make future decisions about the variables that they can impact: AiFTP, PL ramp rate, and workout substitution.
Therefore, all it requires is to bench test the algorithm with test cases. Or, get someone who plays with AT on a regular basis to estimate the question.
The point here is that AT could offer two different solutions to the circumstance of: ‘all workouts easy/moderate’. One: raise AiFTP until the workouts that ‘should’ be hard are hard. Two, ramp PL until the workouts that ‘should’ be hard are hard.
5 12 minute over under blocks are harder than three. But three at 10% higher FTP are similarly more difficult.
Some people will tend to find one type of progression easier, others another. But AT doesn’t really know the difference between these two unless it has a lot of data about the individual.
In short: I’m pretty sure TR know exactly what ‘one block of ssbmv all votes moderate’ should do in terms of AiFTP.
1 Like
Yes, probably. But the issue is whether or not TR should surface the data to its athletes? Or if AI FTP predicts a value that some people think is too low for them (“But I want a 300+ W FTP!”). I think it can create wrong expectations and set the wrong incentives.
In my opinion, TR should focus on making its athletes more consistent, not focus on a single number.
FTP progression and PL progression are not equivalent. Currently, AT uses solution 2. In the future, TR could combine that with FTP progression (by simply changing the FTP to whatever value AI FTP has determined). But I think you want to have a combination of both, increase in power and increase in duration and/or repeatability at a given power.
However, FTP progression takes into account that your lactate threshold changes with training. PL progression aims to increase time until exhaustion or repeatability.
It isn’t about what is easier, but about what is right.
You seem to be misunderstanding me. Sorry you had to invest so much time in a detailed reply. At no point did I suggest TR should ramp FTP rather than PL. Nor that a higher lactate threshold is more important than repeatability for winning races.
Simply: they have the data to answer OP’s question.
However: at some point TR / AT is making an active decision about this.
The question being asked is: 'at some point AT is making a value judgement about which best provides an athlete with the correct stimulus: increasing intensity or increasing TIZ.
I’d be interested to know where the curve sits for this. Eg if you’re at pl 3, 6, or 9, when does AT view the value proposition leans towards increasing FTP vice TIZ. It also might depend on training phase; TIZ during base, intensity during build?
3 Likes
Our of honesty - after 4 weeks of ssb, I found my threshold workouts were feeling too similar to my SS workouts, and honestly too easy. I was hitting moderate on over unders. So I chose to AiFTP rather than continue to increase PL. It jumped me from 329 to 334. I could have pushed TIZ and aiftp’d after the rest week, but I was worried it would overestimate going into ssb2. I was also influenced by knowing I am returning to training, and have previously 20 min tested in the 360-370 region.
Great discussion! I imagine a quick query of the data lake TR has accumulated, would answer that. They could run all kinds of queries. For example, they could tell us:
87% of the time, when athletes perform the prescribed TR plan, and marked all the “end of workout” surveys as Easy or Moderate, and accepted the TR AI workout adaptations… that by week #8 of the plan, the typical FTP increase (vs the start of the plan FTP) was +5%.
Or whatever the data tells us. That is obviously no guarantee. And there are diminishing returns as you approach your personal capability limit. But that would be interesting. And would be a great testimony to the effectiveness of the TR AI engine.
1 Like
I think this probably falls within Nate’s “secret sauce” territory. I also think that the answer or prediction will fall within a wide bell curve and therefore, as mentioned above, it will cause all sorts of complaints and issues when some people don’t see the outcome they desire.
1 Like
If that feature ever comes, I think it’ll be time to leave the forum! All the threads giving out.
I am 100% sure we’re going to see the same with PL v2.0 taking into account unstructured/ outside rides, as it’s not going to show what people hoped from group rides. We already saw it pre-AT, when they changed the plans - years of complaints TR plans are “too hard”, suddenly became “too easy”
1 Like