Try it now. The new model is significantly different, so it might work “better" for you.
When was that FTP test? What training were you doing before and what training are you doing now? If it dropped that much I would think it means you have not been doing much Threshold Training. Share screenshots of your recent training if you want feedback.
Are you saying you changed the entire plan to easier? What I would do is just set the one zone to easier, not the entire plan. Leave the rest of the plan where it gives you harder workouts in the zones you want to be harder.
No, I never changed anything to easier. I changed from balanced approach to demanding training approach as Trainerroad was cutting back too much on the hard workouts based on outside rides. I never set anything to easier.
hm I was using TR for a few months leading up to an ultra event in the Fall but my progression levels never increased because I didn’t like the plan it created for me so I never did any TR scheduled workouts. I would look at what the AI suggested, create a version of the workout in intervals.icu and then do it and sync. I doubt it’s that common, but I can imagine there are other users who use TR for other features like the traffic light system, workout library, etc and just sync from Garmin or Zwift so the progression levels never increase. I have ~200 or so rides on my calendar.
i’m confident in being able to hold my current ftp for 60 mins (i did it and more in 2025). i in no way expect to hold my predicted ftp for 60 mins in 28 days time. in fact i’d say 20 mins would be it. unless its miracle training from TR
I think this will also be super useful for the crowd that wants to use FTP for pacing race efforts. Not to mention I think it provides a more complete picture of where your fitness is predicted to be and how that changes over time.
Great. So you don’t need to use the TR AI FTP number for pacing, because you have more data, and it makes sense to preferentially rely on that. Superb.
So why are you saying that you’re hoping to use the TR AI FTP number for pacing? You don’t need to.
Or are you just wanting to point at the number that’s too big and say “that number is too big”?
Post your recent training and your future training.
I was in the same boat and now that I’ve got a few workouts under my belt the prediction has settled down to a far more realistic….. 0% increase in the next 15 days.
Which I’m fine with to be honest 
Yes, will do!
Is there any chance to recalculate prior AI FTP detections using the new method to see if it better reflects progress?
I don’t believe so … although perhaps someone from TR could try that for you?
I had a similar drop (-21W) when I was in the beta and was skeptical at first, mainly down to relying on my FTP figure for pacing on long events. However, after accepting the new number (and it is just a number) I’ve had some really good workouts, challenging and just the right level of watts over time. I’ve now refocussed on the monthly increases rather than anything else.
That’s fair. I only do my workouts within the TR platform and follow a plan 95% of the time. I’m probably more of an “ideal” user from that perspective versus the triathletes and ones that mix other platforms.
I’m with you on not really caring about the number, as long as power curves increase, etc. In the end, it might not matter, but I can understand the heartburn and gut reaction to seeing the sudden change when when SO MUCH discussion around cycling performance is tied to that value. My “FTP” went up 12% (freshly back on the bike and at high PLs due to recent workouts), so my perspective could be skewed a bit versus seeing a 5-10% decrease.
So far I like the update and new UI. Question… since I have a bit of flexible time the next weeks, I played around with the workouts/lengths to get the highest FTP prediction. Why not just show this as „better alternative“? And is it really possible to get 10% better in four weeks? The hard workouts in four weeks look daunting.
(FTP has increased with new algorithm, but should be on point.)
I guess because sometimes when people say they have 60 minutes to train, that means getting out of bed at 0500, and getting up at 0430 to do a 90 minute workout is a step too far.
Actually in the end it was longer hard workouts and shorter/less easy workouts. So not more time in total. (Edit: I kind of agree with it, I was doing lots of long rides and not hard enough.)
From an ML perspective it looks a bit like the model is overfitting and acting too much on small changes. But let‘s see how it reacts to the workouts when done.
This is purely a guess, but did you somehow struggle or not follow the workout structure? It shows you did less time than the assigned workout and lower TSS/IF that expected.