First I want say I love the new AI training features and plans as I definitely felt stronger since the last few weeks. However, I don’t really understand how the FTP prediction works, and it felt very disheartening. My FTP detection day is tomorrow 2/6, and one week before I felt really strong in the legs but the AI doesn’t give me any Z4 workouts above 4.0 level (my Z4 WL was 4.1). I started tweaking the workouts but anytime I added a “not recommended” workout, the FTP predication would drop significantly. One workaround I found is to add Z4 workouts back to back to progress up the WL to eventually hit 4.0. So I got the screenshot below on 1/31 and the AI planned all these sub/near WL4.0 rides with a 297watt FTP prediction.
Then, something weird happened (see my calendar screenshot below). Although I completed the workouts perfectly on 1/31 & 2/1 with no pause/drops and RPEs matching the AI expectation, the AI reduced the WL on 2/2 & 2/3 from 3.8 to 3.5 with no reason. In fact, these Z4 workouts felt like SS to me as I did them back to back feeling no sore in the legs. Then on 2/4 I was tired of the AI and pulled the trigger to do a WL4.5 workout, and in fact I got 4.6/4.5 out of it and I rated it easy. But the FTP prediction dropped to 294 from 297, almost 1%.
So essentially I was following what was proposed by the AI (except yesterday) but the FTP prediction dropped without a reason. This is seriously disheartening and I want to under how to avoid this. @eddie or @Nate_Pearson , please help.
Maybe it doesn’t trust you after rating threshold workouts with power records as “easy” ? Or maybe it’s the back to back to back threshold workouts causing a little fatigue?
You are doing too many “hard workouts” and the AI is trying to stop you from burning yourself out. Sweetspot and above counts as a hard workout, you have 5 hard days in your training plans.
But I wanted to get more Z4 WL4+ workouts yet the AI refuses to do it. It kept me in the 2.8-3.5 range which is all like 3x9 o/u workouts. If I just change one workout from 3.8 to 4.0 it became “not recommended” and would drop the FTP prediction. I wanted to show the AI that I can sustain 3x12 easily.
you make a good point with the power records…but considering that I can make power records after b2b2b2b o/u workouts, that means I am stronger. Even if the AI doesn’t like it, shouldn’t it at least hold the 297 prediction instead of giving out low numbers?
They were all indeed easy even carrying over the “back to back fatigue”. The 285 FTP was also set by the AI detection when beta was released…maybe the AI recommended workouts were so effective that my FTP grow out of it in less than 4 weeks?
Totally agreed…so the 297watt prediction should hold instead of dropping, right? Also I would be interested in what the actual AI FTP detection will be.
I don’t look forward to the next 4 weeks as I did 4 weeks ago. I felt the AI is just playing with the numbers - boosting a high expectation but then cut it down for no reason when near the detection day.
I had the same thing happen. Tomorrow is my prediction day, and it was supposed to be 316. I had a sweet spot that I nailed today, marked it as “Hard”, and AI predicted it to be so. Hit the levels I was supposed to and the AI FTP detection for tomorrow dropped 2 watts anyway. If I hit the workouts at the level it predicts I will, why does the AI drop the FTP prediction? If I change it to “Moderate”, the AI FTP goes back up to 316, where it was before the workout. I think most of us just want to UNDERSTAND what’s going on. I get it that it’s going to do its thing, but it seems incredibly arbitrary.
I think the missing piece of the explanation is heart rate. Did you have heart rate data for that ride?
If so, then despite hitting the power targets and expected RPE, your HR might have been higher than expected. The AI FTP detection and prediction model uses heart rate (if available) in addition to power*time and RPE. Higher HR for a given power means that you are less fit.
It’s also not clear to me if fatigue factors into the FTP prediction. @Zizhao_He fatigue does affect workout selection by the AI. It will not always give you a workout with a WL at or above your AL because it’s trying to avoid overly fatiguing you.
I reckon you’re spot on here. We know that the model now looks at HR, whereas it didn’t before - but we really don’t know how it’s assessed, or to what extent it influences model predictions compared to, say, power and RPE.
An (warning: n=1 !!) example from me - I had a 1hr endurance ride yesterday, before which I was predicted a 5w FTP bump today. Predicted RPE was something like 70/30 easy/moderate. Felt a bit tougher than normal (after a hard SS the day before) but I rated it ‘easy’ anyway, in the interests of ‘protecting’ my predicted FTP……. Regardless, my prediction changed from +5w to +3w. When I changed the rating to ‘moderate’ it stayed at +3w.
So I compare the HR to previous times I’ve done this workout, and both my average and max were a few bpm higher than they have been. It feels like the AI looked at my HR and decided I’d been sandbagging the RPE, and knocked me down anyway!!
Updates after the detection - I was so skeptical about the system that I didn’t do the Z4 workout yesterday, and as expected the AI dropped my FTP detection from 294 to 292. Not sure if missing one workout can truly affect 2 watts of FTP.
But what’s predicted seems even crazier. AI now predicts a FTP boost from 292 to 313 for the next 4 weeks. That’s 21watts/7.2% which is something I have only seen in beginner gains (I have way passed that phase). With my weight at 68kg I would jump from 4.3 to 4.6 w/kg. The number just doesn’t seem real to me. AI could drop 2 watts of FTP for one missed workout, but then boosts 21 watts in 4weeks?
Very good point - indeed my HR was high during the over part of the Z4 workouts, hitting something like 175bpm max. My max HR from Garmin auto detection is around 187 so no way I have hit my HR ceiling. Also I didn’t feel anything about the heart rate stopping my legs from spinning. But this HR thing does explain why it adjusts my WL for Z4 workouts. I wonder where we can set HR zones in PR? I don’t see such an option. Or the AI doesn’t know about HR zones either, it just knows the Bpm?
In short, it’s probably lowering your FTP because like @JoeX mentioned, you’re working against the system + you’re digging yourself into a fatigue hole, so doing more doesn’t always mean more watts.
You applied 3 back to back Threshold workouts to a Recovery Week, hence why you were seeing all the Endurance workouts. Recovery Weeks are meant to be low intensity so you’re body can absorb all the training, recover and be ready for the next block. I see you’ve also manually adjusted all your future Recovery Weeks to include almost no rest.
Training 7 days a week with no Recovery Weeks on end is a recipe for disaster I would try to follow what the plan is giving you instead of forcing changes to it. We can see the system raising your FTP (even if you see watt decreases) and already predicting a higher FTP in the future, so it’s ramping you’re fitness.
I really would be mindful with all the training, and labeling the post workout surveys accordingly. Prior to the new AI FTP Detection of 285, you were sitting at 280-284 and rating the workouts with Intensity or Stress:
Honestly it makes some sense, if you follow the plan. Your training has been less then optimal as you are doing multiple back to back to back threshold workouts. Follow the plan. It may not be exactly those huge gains but probably better than what you accomplished by trying to confuse the model
Regardless, my prediction changed from +5w to +3w. When I changed the rating to ‘moderate’ it stayed at +3w.
