Perfect Example of why the Ai Prediction is so incredibly frustrating

FWIW, after sharing similar frustrations with what you’ve described … I actually quit using plan builder and have totally flipped to enjoying the AI FTP prediction feature. After an AI FTP assignment, my predicted FTP will drop like 17%. Then, over the course of the next 4 weeks, it keeps inching up further and further as I complete rides - which I find strangely thrilling.

I’ve been riding whatever I want. If I feel like a workout, I use TrainNow. I commute to the office by bike when I want. I do a punchy group ride when I want. I skip days when I don’t feel like riding. It’s been really freeing. And I’ve managed to still increase my TR FTP in the process. It’s a fun inversion if you want to give it a go :slight_smile:

Yeah that or a ramp or whatever metric you are trying to improve. Pick one and go with it. The best “predictor” of performance is the actual performance so if you need to do 40 twenty second efforts for your crit racing or 55 min steady state for your 40k or whatever, you could just see how that improves over a few months. The idea of “my goal is for aiftp to give me the number it said I was going to get 4 weeks ago” is a weird goal, isn’t it?

Now having said that, it is surprisingly frustrating seeing that number come down when you are busting your ass to nail your workouts. Like…I don’t know why I care but I do and I can’t seem to embrace the idea of “just turn it off”

But….ive been doing 20 minute tests for a long time and those numbers are looking pretty good and the workouts seem right so it is working, probably better than ever.

So….how do your actual performance numbers look?

Joe

I still don’t understand the quest for no testing. I do a long form test 3 or 4 times per year (30-40 min. Kolie Moore style test). With the long test, you don’t suffer above threshold like you do with a ramp or even a 20 minute test.

People train hundreds of hours per year, if not a thousand. People are so resistant to four 30 minute threshold efforts? TR has been on this quest for years to eliminate the need for testing. It must have cost a small fortune.

People used TR workout with 4x10-15 min at ftp with 5-10 mins rest to validate ftp. Forgot the name of it. It isn’t that hard to try riding at what you think your ftp is. You will get feeling of it immediately. Try few intervals at 10m 100% maybe 2x15 or 2x20 with 5-10m between intervals. You will notice if it’s your training number or not. It works for me, but we are all different. On the first maximum second internal you will get general feeling if you are close to your ftp. It is trainable to ride more and more at your ftp.

I think you’re example workouts of validating ftp makes sense and I feel like there are various methods/workouts to do that. Nero cycling gave an example workout he uses to validate ftp.

I feel like most people who have trained a lot get to know what the different zones feel like and if they’re pretty far out of whack. And if that’s the case you can always up or lower the ftp a bit and see how things go.

Especially when you don’t have a target goal you need to progress towards TrainNow is a good option.

Yes, that’s a very important point. What I wrote about works for me because I’m just training for general fitness and doing well in group rides. I don’t cope well with “A” events. The mental stress of building towards one big event is too much for me haha.

My pathetic and simple monkey brain likes seeing rewards for my efforts. If I crush a hard workout and see my AI FTP prediction climb 8 watts, it’s as if someone patted me on my snowflake head and told me “good job”.

Almost makes me wonder if there’s a good feature request here. Like maybe a ‘princess’ mode or something that flips the script on AI FTP Prediction. Where it shows you what you’ve added after a workout instead of what you’ve lost :thinking:

Last year I had an AIFTP bug and support recommended Marion for me to validate a 20 min test that I had preformed to get things back on track.
It should feel Hard if the AIFTP is right.

I like this idea. I’m struggling lately with life intervening, and every time I have to miss a workout or change to a shorter one, etc. - the perfect-future predicted FTP drops (as I would expect it to- I didn’t do the work) and that isn’t exactly motivational. I have to think they didn’t quite anticipate the psychological side of this change… or at least I didn’t. :slight_smile:

Nothing wrong with that. And if it keeps you consistent that’s a big part of the battle. It’s nice having TR be able to feed you an appropriate intensity level workout with TrainNow

The AI has some quirks for sure. I’m near the end of a 4 week block, at the beginning of it, it gave me a lower FTP than I thought I should be at based on recent race efforts (256). First workout I did in Zwift up the alp, when I got to the cool down I turned off erg and did the last 20 minutes of the climb at 308 W normalized… that should probably not be doable at a 256 W FTP. I then turned up every workout the rest of the block by 10%, and have successfully completed every workout. This last Friday I did a free ride on Zwift on the Alp again, and did 290W average 291W normalized for 64 minutes, so I don’t see how my FTP could possibly be lower than 290, I even had enough left in the legs to do 320W the last ~3 minutes of the effort. As of this morning, it was predicting me to be a 287W, pretty close… I’d take it, but it dropped it to 265W after I changed out tomorrow’s workout for a slightly harder one. I’m guessing it will go right back up after I complete the workout and mark it as what I expect will be moderate. But strange nevertheless. my profile if anyone cares enough to check it out.

Quick follow up on this:

So as you can see in my calendar I had my last workout before my trip today. I put in some extra time during the weekend and tackled the workout today. Completed all the VO2 blocks and then realized I was running a bit late. So I decided to cut the cooldown short and rated the workout as Hard (which is also what the model expected).

The result? The FTP Prediction dropped again and now tells me that after a month of working out hard my FTP will go from 327 → 327 :rofl: Long story short: the workouts are great (most important part), the FTP Prediction feature is a complete joke and should be scrapped entirely or at least not made the centerpiece of the APP/Website.

My AIFTP was detected at 254 watts, I’ve looked at the Lamarck workout, which is 4x10mins at FTP, the AI says that ‘This workout is predicted to be too intense for you’. How can my FTP be 254w if I can’t ride 4x10 at that wattage? If FTP is set correctly, then this should be a doable workout. It’s things like this that just take away faith in the system.

I currently have a blank calendar until my next AIFTP detection, this morning I added an AI VO2 max workout and it dropped the prediction 4 watts, so I’m better off not doing any training for the next week than actually getting on the bike. :joy: I simply can’t get my head around this, unfortunately, I’m of the belief that they’ve added a layer of confusion and stress to a system that was simple and worked well enough.

I have also interesting case. I did train but not by TR structured plan. Other plan. TR just “watched” me doing WO. At the last day, end of a more than 50 days block of training it predicted AIFTP on thebsame day, as 120. Ok. I did a ramp test and ended 50% more. :slightly_smiling_face:.

Absolutely 100% agree with all you say here. TR is just plain crazy/stupid now. I had thought/hoped they’d take on board all the frustrations in the feedback, admit it was just s wrong turn and they’d fix it. But it seems they’ve dug their heels in now and its staying.

Just curious how you are determining your “actual physiological FTP”?

What if he weighs 290 pounds? You can’t say someone is advanced just by their FTP.

Is it though? If you set TR to accept the FTP prediction then it will adjust your workouts below what you can actually achieve because your only measure is RPE which some people struggle with.

My wife thinks every workout is hard because she doesn’t do well with discomfort. If she rates workouts that are below her ability to perform as hard then the AI FTP will keep downgrading her FTP and therefore her workouts which will prevent her from progressing as well as she could because they will become easier and easier (based on the downgrade of her FTP).

It seems the issue has become that the primary indicator of our FTP with AI FTP Detection is RPE. If I wanted to use RPE as a measurement of fitness isn’t that going back to the dark ages?

Kolie Moore?