Part way through my month of ftp detection. Changing the RPE of a single recent workout in the month (from hard to very hard, and vice versa) results in a 8% difference in my predicted ftp change in just over a week’s time (hard = 0%, very hard = my ftp will be 8% lower than before the month (so motivating….). That seems a tiny bit unrealistic to me. Can you confirm if something is broken with TR or my login? I am trying to follow a training goal here, but am fast losing confidence.
Why have FTP prediction at all though if it swings around so much based on single surveys and single workouts. Through all of 2025 (apart from a month break due to an operation) I found it was all robust. I did my month of training, I’d look forward to my FTP AI update, and I’d also see small but consistent gains. It really motivated me and over the year I improved my FTP by about 50 (first full year of training so some good early gains). This year I’m training more structured and more consistently but Trainer Road AI is letting me down. It starts off my year by basically changing the ground beneath my feet (adjusted my FTP by 20 watts). Then it keeps telling me that if I do my workouts for a month I’ll get slower or stay the same. TR response is “dont worry just carry on” basically. Sure, but then what’s the point of all this AI FTP detecting and predicting if the answer is that the ftp doesnt mean anything. How do I know if I am improving based on data TR is giving me? It is ruining my motivation to use TR now it’s FTP detecting is erratic and doesnt seem to track day to day, or month to month anymore. I like the ideas they added, but the core function has gone from awesome to meaningless to me and I really miss that
Nate mentioned that in the future they may release some form of individual power curve data (maybe just 60 minutes and 20 minutes?). That might help clarify some of this for people looking for concrete ways to measure fitness improvement beyond the new version of AI FTP.
For me I’m turning off the display of the AI prediction as I find it messes with my mind and I end up over thinking my survey responses and whether I need more/less rest etc. I’m not (yet) convinced that it’s accurate, but so long as the workouts I’m given seem right (based on 7 years of TR use) I’m happy enough. So far (about a month) so good on this front.
AI detection is different, I’m trusting but verifying. I race a lot, so it’s easy to check my power curve after race to see if it’s trending in the same direction as my TR FTP. I’m only checking the trend, not the actual values.
Hey Luke, I can see how this would be frustrating. I looked into your account and here is what I found:
First, your fitness didn’t change overnight on Jan 23 when you accepted an AI FTP Detection value using the updated version of AI FTP Detection (266 to 247). I can see how this would be a particular sticking point for you as you had the goal of raising your FTP up to 280 be the end of 2025, but with TrainerRoad AI updates, I think you’ll have a better chance of that now than you ever did before! In terms of the FTP values themselves, think of it like getting a new scale that measures differently. You now have a scale that is more precise and is data-backed to give you better training. Just as it wouldn’t make sense to compare the values from the old scale to the new scale, and the same applies to your old vs. current FTP values.
When you did that 30min (seemingly) all-out effort at 253w on Dec. 31 and got an AI FTP Detection value of 266 with the old version of AI FTP Detection, our new updated version would have given you an FTP value of 257.
After that you took a bit of time off and a break from structure, then started back in with your FTP at 266 but you were finding your hard workouts more difficult than they should have been, which is good evidence that it was harder for TrainerRoad to give you the right workouts at 266.
Once TrainerRoad AI was released and you accepted the 247FTP value that the new version of AI FTP Detection gave you, your workouts were better calibrated.
Now, in terms of why your FTP isn’t going up in the future, there’s a bug that’s affecting this that is evident on your account. It’s a high priority for our team, so I’ll update you once we get it solved.
Thanks Jonathan for your reply, I appreciate it a lot. I’m relieved to hear there is a bug and an issue with my account. I can get behind the ftp recalibration as long as the reliable and consistent monthly gains I have enjoyed with TR continue (I realise that these gains may slow down a bit as I get fitter and closer to my limits). Thank you for recognising and prioritising the bug and I will now continue with my training while you look into it.
Great news! The team has a fix in for the bug. This bug fix doesn’t change what the FTP Prediction is–it’s actually fixing an issue that was not giving you workouts that were progressing properly.
Looking at your account, your current FTP Prediction is 250 and by the end of your current AI Training Simulation Window I can run an admin tool on the backend and I’m seeing an increase to 255.
Since you replied to give me the promising news of the bug being found and fixed back in Feb, I have a bumpy ride with TR. First of all I encountered a subsequent different bug (verified by your team and communicated promptly to me from Evan) that was causing TR to not properly ingest some of my non-TR ride data. This was triggering a premature AI detection and leading TR to not be able to find workout for me. Again my account was “fixed”.
We’re now at the end of the training window you looked at back on Feb 10th. I’ve had 4 good training weeks with 5-6 solid rides a week, and I set some good PRs on my benchmark ride routes without pushing myself overly hard. Steady progress.
Yet the AI prediction just steadily trends down as time progresses through the window. That 255 became 254, 253, 252, 250, and……yay.. on the last day 248 (so worse than at the start of the month). Though for a short period of a few days in the middle of all that it had a moment of positivity and gave out a 258 and 257.
That’s an awful lot of variance, to the point where I don’t feel I can take it seriously or get any value from it, and it doesn’t feel right to me compared to the subjective feedback of how I feel, and the results I’ve been seeing on the bike.
So my question: Is this working as intended or is my account still affected by bugs? It is hard to know at this point what is and isn’t what I should expect.
