AI FTP detection offered me 8w decrease, did a ramp test and got a 20w increase

I have been back training indoors since October 1st. Starting with 264w FTP (AI detected). At that time, the estimated FTP felt spot-on. On November 20th AI pushed it to 271w. Since then though, the workouts suggested by TrainerRoad have been starting to feel a bit on the easier side. Using a medium-volume training plan produced by plan builder.

On December 19th AI FTP detection suggested that my FTP had fallen by 3w (271w → 268w). This was quite disappointing. But perhaps fair as I had been skipping quite a lot of workouts as I had been replacing them with cross-country skiing. I decided to not accept this FTP downgrade as 3 watts doesn’t really make any practical difference, and I had not been feeling that the workouts are too hard. Rather I had started to feel the opposite.

To improve my consistency, I decided to switch to low-volume. Now, with less volume, the workouts started to feel even easier. (except for the breakthrough Anaerobic+3.1 workout, which really had me on my limit). I padded the plan with some endurance volume plus one end-of-the-year race.

This takes us to today, January 16th, when AI FTP detection offered to lower my FTP to 263w. I did not believe this. How can my FTP be decreasing when my workouts feel easier? My progression levels have been going up, why is my FTP coming down? This made no sense. So I decided to do a ramp test instead and got an FTP of 290w.

And I wasn’t even in a particularly good state for a ramp test. I had had a horrible sleep and a extra-stressful day. At the start of a ramp test, riding at recovery pace, my heart rate was already 150rpm. Normally I can ride with steady 140rpm at endurance pace for several hours.

I have been pretty happy with the AI FTP estimates so far, but now I’m quite a bit more skeptical. There might be several factors that contribute to the AI FTP detection not working well for me:

  • most of my workouts are classified as outdoor rides (well, they’re really done in Zwift, but TrainerRoad probably doesn’t tell the difference).
  • I’ve done some cross-training about which the AI knows nothing about
  • My survey responses about workouts might not have been good enough for AI to take action. For the endurance rides I almost never rate them anything else than Easy or Medium. For the higher-intensity workouts my standard answer is Hard, with a rare Very Hard. I think I’ve never rated them Medium. But I feel like there’s a wider spectrum within this Hard category. Like, how should I express that while the workout was hard I’m confident I could have done several more sets of intervals?
1 Like

I think that AI FTP detection only looks at the data of the rides you did, not the post workout survey. Might be wrong, my impression was that the surveys are only/mainly used to alter your training plan. E.g. if you mark workouts as Really Hard it’s more likely to dial down your progression and give you some Achievable or Productive workouts, whereas if you’re marking them as Easy or Moderate it might dial things up a bit and throw you some Stretch workouts.

What kind of plan have you been following and what kind of ski workouts have you been doing? That would be one source of missing data which could throw AI FTP off. Especially if, say, you’ve been cruising in TR workouts but pushing things on the skiing.

Also just possible that you’re somebody who ramp tests really well, and the number you get from a ramp test might be one you struggle to do workouts at. I’ve had this in the past, I seem to get a higher number from ramp tests than I do from 8 or 20 minute tests. The ramp test number is fine for me doing VO2 and Endurance workouts, but I really struggle with it for SS and Threshold workouts. To the extent that last summer I was ignoring TR power targets for those zones and just doing them by RPE outside (including just entering some road bike TTs instead of doing threshold workouts!). Switched to using AI FTP a couple of months ago, it dropped my FTP by 5% which was fairly dispiriting at first, but on that number I’m now able to nail SS and Threshold workouts and I’m making good steady progress in those PLs which is mentally great after a couple of years of dreading them, and will hopefully pay off in terms of actual fitness as well since I’m getting better quality and consistency in those zones. I haven’t done any VO2 workouts on the new AI FTP, there are some scheduled in my next block starting in February and I suspect they’re going to feel pretty easy at first, may have to pick some alternate Stretch or even Breakthrough workouts to bump the PLs up a bit more quickly if so.

If you haven’t already seen it, the TR blog on this might be useful - Post-Workout Surveys – TrainerRoad

Basically “Hard” is how hard it was to complete the workout, not a measure of RPE. RPE for VO2 and Anaerobic workouts is always going to be pretty high, they’re always going to get your HR and breathing rate up high, but those workouts can still be “Easy” or “Moderate” if you feel like you’re on top of those efforts and could handle even more, you’re recovering quickly between intervals, you feel like you could do multiple more intervals at the end, etc.

1 Like

Hmm… there seems to be contradictory information about how I should really answer this survey.

In the linked article it says:

I also remember from the Ask a Cycling Coach podcast it being described as:

The descriptions of each intensity in “How to rate your effort” chapter seem to be much more clearly actionable though, especially when concentrating on how large part of the workout would you be able to repeat:

  • Easy - could repeat this whole workout again
  • Moderate - could repeat several intervals again
  • Hard - could do one interval with some mental effort.
  • Very Hard - would not be able to complete one more interval.
  • All Out - barely managed to even complete the workout itself.

Here is my own sheet, that came from my initial thoughts & discussion with TR reps:

And as mentioned, it’s about the actual experience & true feel in the moments, not expectations of what you thought it might or should feel like.


When you switched to a low volume plan, did your total volume come down? If not total volume, how about TSS?

I don’t actually know what model they use, but I could definitely see one of the two scenarios above causing it to think your FTP went down.

I wouldn’t expect the skiing to really directly benefit my cycling. I think it has more of an effect on my upper body and core strength, minimally on my legs. I’m not even doing any specific skiing workouts - I’m mostly just trying to suck less and stay upright. Though the skiing tends to be on average higher intensity, as I’m currently kinda incapable of just skiing slowly with little effort.

Though for the moment no more skiing… the snow has melted away in the last few weeks :stuck_out_tongue:

Tricky to say. Monthwise there has been a bit of a downward trend in TSS:

  • 1616 month 1 (med volume)
  • 1552 month 2 (med volume)
  • 758 month 3 (med volume, lots of skiing here)
  • 1376 month 4 (low volume)

That would explain why I got a minor FTP drop after month 3, but doesn’t really explain an even larger drop after month 4.

If you’re struggling on a pl3.0 workout that might be a strong indicator for AiFTP to lower you a bit to help you be more consistent.

I don’t know if this is right but my experience is that I don’t usually go up FTP until I’m completing 5-6+'s without failing.


It was a +3.1 workout, meaning it took me to 4.1 level.

1 Like

It seems XC skiing and Speed Skating are two highly correlated forms of cross training with cycling. I think you would 100% be influencing your fitness on the ski’s and as you say, TR can’t see that.

1 Like

XC skiing works your legs as well as the upper body unless you’re just straight double poling.

Skiing without poles is a decent drill for teaching that. I did that a ton while learning to (skate) ski. It’s slower than with poles but not by drastic amounts; though steep hills are considerably harder.

Well… I’ve been skiing exclusively in classical style with a good amount of double poling. I’m sure there’s also some benefit to legs, but I’d think it would be more in the area of improving some muscles I use less during cycling. But you might be right and it’s indeed the skiing that has improved me.

But even if my cycling fitness came as a result of skiing, it doesn’t really explain why the AI FTP thought that my FTP should be lowered after I stopped skiing and concentrated on doing all the trainer workouts.

1 Like

I think this is a great chart for reference, thanks for sharing it again.

It would be nice if it were connected to the post ride survey…like keep the same survey but maybe add a link to this chart for anyone unsure of what to fill in?


I’ve mentioned it before, but TR had a form of “Tool Tips” in the survey at one point. It was a bit flawed with some errors, and they removed it shortly after I saw it in use and mentioned the issues.

It’s not returned yet, but I really think TR needs to add a short supplement link with some form of their support doc there. Until then, I keep mine open in a separate tab on my PC and reference it for most rides when cooling down.


I don’t think anything is ever going to be completely definitive.

My Sunday ride was twenty minutes at z2/3 then three sets of over unders then another twenty minutes at z2/3.
At the end of all that it’s kinda difficult to remember exactly how hard the over unders were.

But your chart would help narrow down the correct response.

  • Yup, I end up waffling between ratings on occasion, even with my chart.

Subjective rating is subjective :stuck_out_tongue:


I agree with that, mostly…

1 Like

:rofl: :rofl: :rofl:

1 Like

Hey @nene! Good discussion going on here.

  1. For your Zwift/Outside rides: AI FTP Detection will use that data, even if these rides aren’t associated/paired with a TR workout. This data will still be helpful for AI FTP Detection.

  2. You are correct that AI FTP Detection can’t analyze your XC skiing. It’s likely that the decrease in cycling volume over the period of time when you were skiing more contributed to the AI FTP Detection you got in this case.

  3. Survey responses will have an indirect relationship with your AI FTP Detections only in the respect that they help Adaptive Training give you the right training and based on that training, your fitness will evolve. But, surveys are not directly used by AI FTP Detection as of now.

Additionally, it’s important to note that your Progression Levels continued to increase over this time period. That means you’re still improving the zones being targeted by your training! You mentioned a Breakthrough workout had you “on your limit,” which sounds about right – Productive or even Stretch workouts are meant to push your fitness along in a sustainable manner without burning you out or being overly challenging.

The good news is that all of the data from your recent training block, combined with the Ramp Test you just did, will give AI FTP Detection more information to make better detections for you in the future. Keep in mind that AI FTP Detection looks at all of your cycling training data over time, so the more data it gets and the more consistently we train, the more accurate each detection will be.

Hope this clears things up! Feel free to let me know if you have any other questions.


When I’m in the middle of over unders I often think about how I’m going to rate the workout as hard in the post workout survey, but then toward the end of the ride after the endurance portion and the cooldown I think ‘oh that wasn’t too bad’ and give it a rating of moderate :joy: . I’m a month and a half into this and haven’t failed one yet so at least there’s that (currently at Threshold workouts around 4.0 and V02 at 6.5).