This just hasn’t been my experience. I was in beta for a week or so before it was released, but my prediction has been mostly steady. It went up a couple watts from it’s original prediction. And then it went down 5 watts after a threshold workout last Saturday that I marked very hard. But I understand why it went down - it wasn’t just that rating, my workouts for this week were adjusted to be just a little bit easier. So it makes sense if I am spending a week without as much stress. Either way, it is an estimate. It has only fluctuated +/- 4 watts total from the original estimate, and if you have enough experience with structured training you would know that expressing your FTP on any given day and workout is going to be well within a range like this.
I am new to the platform, 1 day in, how did you find the accuracy? Were you on track?
So I just started the NEW AI features last week. I have been using TR for at least 2 years now and I had loved it. However this new “FTP prediction” is all wrong for me currently.
Last week when I accepted my new FTP number (which was a few watts lower - 
) - it told me based on my workouts for next ~19 w/o my FTP would increase from 231 to 239 - which I knew that was correct and a huge jump! but I was very excited to say the least. Then I started going through planned w/o’s in TR without changing a thing and hitting the mark on each w/o and after each day or every other day my Predicted FTP keep going DOWN - now it’s at 235.
for someone who is doing everything “correctly” with TR (hitting each workout as planned, not changing the w/o, not riding outside yet (been snowed in) and responding to survey with mod/hard/very hard - just as TR has planned - It’s SO disheartening to keep seeing my predicted FTP keep going down when I’m not doing anything wrong.
username for TR is = ShellyFranke
Sounds like it’s just getting more accurate as it gets closer to the date to me?
I think at least part of the problem is that the prediction seems to be based on every workout being completed perfectly (no pauses, no fluctuations in power etc) and this is pretty much impossible to achieve. So it inevitably drops as the workouts are completed.
I am going to turn it off for my next block and just check in in four weeks. I think this will be better for my sanity.
Don’t forget the RPE plays into it heavily too. Rating one workout higher or lower than the center of the curve has an immediate impact.
I’ve not actually noticed power fluctuations and pauses making much difference at all - RPE surveys seem to be the biggest single contributing factor.
Along with workouts changing due to extra curricular activities.
I’ve been using it since it’s official release and gotta say it’s improved my motivation and consistency plus just how much proper recovery can benefit your future prediction!
I see and use FTP as a metric for how my training is progressing. It should be a small and steady rise as ride duration/intensity and ‘workout’ change week to week. So a steady few watts week to week is good as it shows steady improvement without over training/burnout.
Ten point increase is a lot and I’d want to see what has changed to cause such a prediction or change…increased duration/intensity/etc and does it continue or taper.
Especially during this period…it is past mid winter in northern ny, where I am…and I want to see steady incremental increases which indicate increased overall fitness and recovery so when I start hitting hiit training, ie hard intervals during late winter, early spring my fitness continues to improve and my overall strength watts and cardio have me properly prepared for the hiit sessions preparing me for the races/events I have planned for the upcoming racing/riding season.
The old adage…”January hero, July zero” is a proven adage…
Yeah that’s a fair point and I don’t necessarily have a problem with how it’s working, it’s more that there is something in my personality or whatever that doesn’t handle these constant fluctuations well. For example, I did a 2 hour hike on Sunday and deleted my planned 2 hour solo ride and my FTP prediction went up by 1 Watt. I would rather not be exposed to all this strange behavior and just check in at the end of the block.
In my experience, the AI is very optimistic. By that I mean that it seems that the prediction is based on us gravitating towards the lowest rating that it expects.
I can use my workout today as an example. A 5.9 WL VO2Max workout where the AI gave 57.6% chance for a hard rating and 39.1% for very hard. I nailed the workout, even got 6.0 WL credit for it. But it was very hard, and I rated it as such. As a concequence my AIFTP prediction dropped by 2 watts. But it didn’t only drop because I rated the workout very hard. It dropped because my VO2Max workout next week was adapted down from a 6.7 WL to a 6.0. As it should, as I would probably not been able to finish the 6.7 workout. If I manually switch it to a 6.5 WL workout I get 1 watt back, but any higher and it’s not reccomended and my prediction tanks.
I had an almost identical experience this morning. VO2 6.7 with a 54% chance of Hard, and a 29% Very Hard. I rated it Very Hard. Nailed all of the Workout Levels and even got a .1 higher on Anaerobic…and it dropped my Feb 6th prediction by 2 watts. My next VO2 session is also rated slightly easier.
I think a lot of people see stuff like this negatively, whereas I see it as doing exactly what I would expect since I rated it on the pointy end of the curve. I think a lot of people react emotionally and see it where “anything trending down is bad”, rather than thinking about it analytically and seeing that it’s adjusting to where your future workouts (and “FTP”) help you hit the expected targets.
Yes, ideally the system converges to something reasonable. Honestly, I view any prediction over a small handful of watts with suspicion, unless we are talking about an early beginner.
With 2 endurance workouts left my predicted FTP is + 3.8% based upon a coin toss of RPE selection on a threshold workout where hard gives me 3.8% and very hard gives me 1.9%.
From experience I know 3.8% is just too much for the work done over the month so will select very hard on that workout.
Im just very surprised the impact of these selections is so large and the nature of basically self selecting my ftp for the detection assuming it wont change now.
The previous ramp result AI FTP was more a statement of fact of my abiloties albeit an estimate or maybe it never was.
Workout selections have been good. I just feel like the predicted FTP is not accurare as it should know from historic rides I will mark at least one threshold workout in a block very hard at my current ftp.
Exactly this, but I think it’s inevitable that having FTP prediction available will lead to this behaviour in some part of the user base. Most people will have multi-month plans designed to prepare for some goal in the future, and progress towards that is not always going to be a straight line.
What I expect and want the AI to do is make sure the plan remains achievable and realistic, so that I can maintain consistency towards those goals. That’s not always going to align with a higher FTP number at some short-term arbitrary point along the way.
I feel like the new workout picking is helping with ensuring the plan is achievable. The prediction feels like a fun toy. It’s interesting to see what impacts the predictions.
I’m coming to the same conclusion but I don’t think it’s necessarily a bad thing.
I’m starting to regard a high FTP as a stick rather than a carrot.
Remember that you are going to have to actually use that FTP to validate it 
I would be ok with “realistic Watt changes” if we know the WHY behind it - even is something high level.
another question - TR has always shown a 1 watt reduction “while i’m performing a w/o” every w/o. I have checked my Wahoo power meter and will watch it on my app while riding in TR and wahoo is on point with the targeted watts - however TR keeps dropping my watts by 1 watt during the workout and I didn’t CARE that much before - but now with AI it’s reducing my results upon completion of the workout.
Anyone else experience this? maybe It’s a Wahoo thing but seems like Wahoo is correct and TR isn’t getting the correct signal.
I was thinking about this as well. When doing TR workouts in Zwift in ERG mode I am always 1-watt down. It doesn’t seem like it would change the AI much, but maybe I should just use the Zwift menu to bump the workout up a percent so I stay above the target. My predicited FTP dropped today after an easy endurance ride that I rated Very Easy.