It’s looking at the watts done. From there it can determine power estimates (and FTP) based on the specifics of the training among many other things.
We are still calculating FTP for our athletes. We don’t look at all the data in the same way as before, but ultimately, the watts you’re doing in training helps to determine your FTP.
You say this like i haven’t been doing Threshold efforts? I have just counted 9 Threshold efforts since the start of my plan, given AI schedules one per a week thats 9 weeks worth..
Yes i have scheduled the odd one or two session to SS, mostly to take into account conditions or similar this is the minority not the majority of the time.
So your suggesting that from Thursday I reduce my FTP down as scheduled? If I do that shall I then complete a couple of level 10 SS work outs on the bounce just to prove the system and yourself that its wrong? Again surely this is going against the point of AI training when i am having to tell it what to do but don’t get me wrong i will enjoy doing so.
come to think of it, even if I was to do that I would still have to wait then several weeks until the FTP detection comes up for renewal again, so a little pointless as that’s another block of training at zones that are not accurate.
What I’m suggesting is for you to use the tools required to get a workout that feels accurate for you. Then we can investigate further and help you with your FTP if needed. There’s no need to train for a whole month at the wrong FTP. I’d be in touch to help you along the way. 
I’ll be working on your case in the background, but again, getting a good hard ride in place will help serve as a benchmark for your current fitness. 
Understood, so do I accept this new AI FTP or ignore it in your opinion, because if i accept then SS efforts are going to be at around 305ish and that sounds painfully low. Given I cruised around just a few days ago for an hour at 330..
Either one works. Just check Workout Alternates for harder workouts that are closer to what you’re looking for and give those a go. Let me know what you find!
One thing that had me wondering after thinking about this a bit more. The AI obviously has some idea of the capabilities of @RenvaleRich, given how hard workouts it has scheduled for next week. But what is the plan after that? Lowering FTP and then immediately scheduling workouts at the very top of the range seems a bit counter intuitive. Where does it go from there?
I don’t know if it’s true, but my theory was that it would help the AI if your plan beforehand looked a bit more like what it ultimately ends up being. If you do 2-3 hour workouts than maybe your schedule should reflect that?
Come to think of this you have said that the power i did on Tuesday was on par with what you would expect from a V02 7-8 workout. Surely someone who is completing a 7-8 V02 workout, marking it as easy isn’t then the same person who is then recommended a 10w decrease in FTP the following day?
Yes very valid point, yet again it makes no sense..
The long and short of it is i have been completing training either as or over what has been prescribed, i have been marking key session as easy when doing them and yet time after time AI FTP wants to continue to decrease FTP.
Feels to me like there has been a shift from rider focus and looking at the data and what’s happening there to gamifying the AI. This is a easy mistake to make with AI…
I’ve done some more digging, and I’ve been chatting with a few others from the team about this one, and I’ve got some ideas and questions for you.
First, how do you decide how to rate your workouts? Also, are you rating the work you’ve done, or are you considering the prescribed workout? What constitutes easy, moderate, hard, etc.?
When was the last time you had a Sweet Spot or Threshold workout that felt hard, or very hard?
It sounds like our workout recommender can get a bit confused when it doesn’t get expected RPE surveys for an extended period of time. If all the workouts it’s recommending are easy or moderate, when it expects them to be hard, very hard, etc., then it loses track of what should actually feel hard for you.
Looking at that Long Suprathreshold workout, I think there could be a few things going on, but you did come up a bit short on some of the power targets, which could have an impact on how we’re weighing that workout.
The question is then, are you stuck in that loop because we’ve never given you hard enough workouts, or are your workouts not hard enough because historical survey responses have led the recommender down that path? That’s what I’m trying to figure out now. There’s at least one other athlete in this situation right now, so I’m learning as I follow up with each of you.
Regardless, depending on how you’ve been answering your surveys, I still think a good option moving forward would be to get you to knock out a hard workout that feels appropriate to your abilities. I’d be happy to help you find one if you’re willing to give that a go. Just let me know.
Let me know your thoughts here and we’ll find the best way forward. 
first things first, you said I came up short on Tuesdays session? The prescribed zone was 387, I hit the following: 407, 410, 404 and 412 so not sure how you have drawn a conclusion that I came up short? I am keen to hear what has lead you to believe that.
Sessions haven’t felt truly challenging for some time hence riding outside of the zone in some instances, but yes I haven’t always been diligent in completing ride surveys, I was led to believe that was a “nice to have” rather than it being essential for the AI to actually work.
Sadly had to cut todays ride a little short due to a ill child at home but plan to get out tomorrow morning and complete a truly challenging ride all going well so will report back tomorrow.
I think the reason he asked about the ratings is because you should rate the work you DID, not the work you were assigned. So, if the assigned workout was super easy, which led to you ramping it way up to make it very hard, but then rated it easy, it’s going to confuse the system, which will think you said the very hard work you completed was easy.
That sounds confusing…so bottom line, rate the work you did, not the work you were assigned, and rate every workout so the AI learns. Hope that helps!
Ok this is fair enough, i thought the survey was in relation to the set session not the session you did (granted they should be the same) but that’s a valid learning in that case.
This makes a major difference. 
If you’ve been rating the workotus based on the prescription, I’d go back and change any surveys that don’t accurately reflect the work you’ve done. This will help quite a bit moving forward.
As for that suprathreshold workout, that was what one of our support leads shared with me. I just double-checked, and you’re right – the power targets were 110% of FTP. 
I’ll follow up with them and let them know that they led me astray. 
I have a similar-ish issue. All of my workouts are done outside. I have been terrible at following them for a long time, not bothering or just deciding to do something different. But since this new machine learning stuff came in, I decided to go for it after it dropped my FTP by 30 watts. Since then, I’ve nailed pretty much everything - it’s hard to get things exactly right when working outdoors, and today I did my 2x20 sweetspot intervals in the zone, but randomly had to move the car for 15 mins (about as long as the prescribed rest…) inbetween the intervals. My AI FTP updates tomorrow, and in spite of going over the TSS and duration a little bit (or perhaps because of), the prediction has dropped from 325 to 316.
I had a similar issue yesterday with a VO2 session where I had to cut the last interval due to childcare, but still was in the ballpark for TSS and duration, where it went from 331 to 325. So that’s 16 watts or 5% of my total FTP that’s been cut after two workouts. I have half a mind to delete the completed workouts and mark them as manually completed, it just fundamentally doesn’t make sense to weight these two workouts so highly with the updated FTP being so close.
I think there’s an issue with outdoor workouts in general whereby the new ML system doesn’t register work being completed successfully with the level of chaos (in the data, at least) that outdoor / real life scenario training features. How much of TRs data is based on people who do all their sessions outdoors?
I have to say it feels like a bit of a kick in the teeth to have performed pretty well this month and been headed back towards 330 FTP (these numbers have value to all of us, let’s not kid ourselves) and then it’s cut to 315/316 instead.
They’ve been telling us for years that our outdoor workouts often weren’t as effective as we thought they were. I think these are prime examples. You took a 15 minute break from pedaling in the middle of your workout. It seems pretty clear why your actual workout wasn’t as effective as the prescribed work.
Thanks Conrad, good to hear from some one else who is looking at the data and how they are feeling and knowing this isn’t corrosponding to what the AI is doing.
Let it be a lesson for life…blindly following and believing AI despite the facts not supporting what its saying or doing isn’t wise…
And yes i get it, the system probably wasnt designed for those who train outside, indoor training creates clean data, as you say with outside data there is to much interference that its probably struggling to cut through, but id rather TR just said that rather than suggesting that because i swapped a threshold session to a SS session two weeks ago then that’s why my FTP has dropped by 10 watts!
Don’t get bogged down in the granular detail, do you really think one session where you stopped pedalling for 15 minutes is affecting your FTP…
Its a little like being on a very strict diet for 10 weeks, sticking to it very well but then one day eating a cup cake, then getting to the end of the diet and saying i think the reason i didn’t loose 4kg was because i ate that one cupcake… the reality is that cupcake made no difference…
It really depends on if it’s one cupcake, or a few cupcakes a week.
I’ve seen some athletes who can complete their outside workouts with robot-like precision and adherance, and others who really struggle to get the intended stimulus from the prescribed workouts when riding outside.
One workout won’t likely make a major difference, but trends over time certainly will.
I think this is the case with RPE surveys. One or two misrepresented efforts might not hut too badly long term, but consistently inaccurate surveys over time certainly will make a difference.