AI FTP Progression vs. AT seem to produce bad specificity

I need some advice. I just sent this to support but I’d also be curious in what the community thinks:

I have a plan builder plan and admit that I substitute a lot of outdoor rides to practice riding in fast paced groups and bike handling, so I understand that this isn’t optimal conditions.

I also travel a lot and therefore switch trainers quite a bit.

That being said, I have failed threshold workout after threshold workout over the last few weeks and only pass them when they get adjusted down to essentially vo2max durations.

I am training for a time trial. Despite all of this, AT keeps adjusting my intervals down and when I schedule a ramp test AI FTP detection keeps telling me that my threshold is going up!

The result is that I am two weeks from my target event and have no idea what my sustainable power is (well I do because I climbed a mountain a few weeks ago, but it’s nowhere near TR’s suggested threshold) and I am getting no training specificity because AT keeps shortening my intervals at something that’s quite obviously well above my threshold.

What’s going wrong here?

More importantly, if I set my FTP to the power I know I can sustain for the duration of my TT, would AT reduce my progression levels and therefore give me less specificity even though I’d be reducing FTP?

Seems like an odd training strategy for a TT. Basically the opposite of a TT. I think the best you can do is manually select a workout with a duration similar to your TT and pick an intensity using the intensity sliders that are something like 85%-90% of your current FTP. See if you can hold it. You haven’t been following AT or the TR plan to this point, so no need to try to salvage the “plan.”


What exactly is your power measurement situation? Please list any and all data sources, as well as general guess to their consistency to each other.

1 Like

All my power comes from my vector 3 pedals. I think that part is consistent. But because I use different trainers I do think there’s inertia differences and I understand that those kinds of inconsistencies might make AI FTP prediction less reliable.

1 Like

Sounds like you don’t actually know your FTP and thus end up using wrong numbers for your training. If you don’t want to do a ramp test or 20 minutes test, I find that a workout like Lamarck is a really good indicator if your FTP is set correctly. Do that workout with your most used power source and what you think is sustainable. Build of that as your baseline.


Hold on, I have substituted like one intense workout for a group ride most weeks. There have been some travel related interruptions but I think that’s within TR’s understanding of following the plan?

Oh I did a ramp test about four weeks ago. And that part is consistent with AI FTP detection. It’s just not consistent with even 40 minutes of sustainable power.

Then it’s just a matter of working on TTE. Build up the following weeks from 10 to 20min intervals. The only way to get better at them is doing them.

1 Like

I just think that it’s strange that AI FTP detection keeps increasing my threshold and reducing interval length. I suspect I overtest on the ramp test, but if training for a time trial, wouldn’t it be better to do the opposite? I feel like it scales down threshold intervals to durations that turn them into vo2max workouts. I’m good at those but that seems like the wrong kind of stimulus.

We would probably need a TR engineer to provide some insight into the algo, which they probably won’t do, but I suspect you’re right, the VO2 work is dictating your FTP, but you don’t have the base to back it up. Not a-typical for riders that go outside and smash with their friends on “a lot” of group rides. TR makes it worse by reducing interval length (as you’ve recognized) at lower PL’s. I really wish they would have some long intervals at lower PL’s so people can actually train their TTE. I think you realize in asking this question that doing more specific training to your event will probably have better results than what you’ve been doing. In this case I think you are going to need to override the AI FTP and set a more realistic FTP for yourself.

1 Like
  • Just to state the clear “nuts & bolts” aspect… Anytime a rider changes their FTP, TR will adjust down the Progression Levels. This is done to be sort of a “soft start” to the next phase of the plan. This matters here and elsewhere because lower PL’s will likely be shorter work interval duration.

    • It’s one of the characteristics that has lead to discussion about skipping FTP changes in order to push PL’s higher. There are several topics floating around that consider the implications of taking or ignoring FTP changes.
  • What was your general time between FTP changes? What were the increases in FTP from one to the other? Were they minor changes, or larger ones?

    • Depending on the frequency of your FTP changes (regardless of source via AIFTPD, testing or educated guessing), you may have reset PL’s often enough to essentially divert the desired progression in work interval duration.

All of these are important pieces to the puzzle. On the surface, none seem overly beneficial or helpful to your TT training goal.

  • Group rides are what they are, and seldom the type of workout I’d consider as TT specific.

  • Swapping trainers may well be altering the relative / effective FTP you have when using each one. It may well lead to over/under performance and not really hitting the intended goals for any given workout. Getting the most from the variation may well require some testing to determine the differences in the trainers, and how best to leverage them for your needs.

  • Depending on the failure, and your specific answers to the survey question, AT may be reducing pending workouts. If so, the is also counter to your TT goals. So addressing the reason behind those failures is necessary to reduce or eliminate them.

There are some real factors here that seem to be tripping up the AT system and it’s likely necessary to address those in order to leave AT at the helm. That, or you need to take the reins and make adjustments you think are more correct.


I think you can fix this yourself while still using AT.

Set your FTP lower. Something you think that on your best day you could sustain 30-40 mins. Dont worry about the suggested PL when it comes up.

When a threshold workout comes up, lets say it gives you a 3.0. At your new FTP this should be too easy for you. Look for workout alternates, find something that pushes the intervals out longer, and complete it. When done, rate accordingly and AT should take over with those new levels.

Eventually this will balance out.

Don’t use a ramp test or a machine estimate… go do a proper FTP test. Warmup well, then pick what you think threshold is, start 3-5% below that, ride as long as you can building the effort until you find threshold, then sit there for as long as you can.

Great, specific workout for you that will give you exactly what you’re looking for. I think the TR workout Lola is a decent proxy for this, but maybe a bit short? Not sure on the name.


What you might be finding is that your TT bike FTP is a lot lower than your road bike FTP. This discrepancy will turn your threshold efforts into VO2max efforts on the TT bike, which you are very likely to fail. Probably the easiest way to see if this is the problem is to look at your heart rate at the end of the first couple intervals and compare it to something like the estimated threshold HR you get from something like a Garmin head unit. If you are over the estimated HR your are probably way off on your TT bike FTP.

If you want a FTP value that reflects what you can do on a TT bike you really need to ramp test on the TT bike or only train on the TT bike for long enough that AI FTP detection understands what you can and can’t do on the TT bike.

I think the lack of information on bike type is one of the biggest flaws in the current AI system. Machine Learning/AI isn’t magic, because without key data inputs (I.e., explanatory variables) it cannot magically determine a failed workout from stress or fatigue from one where you tried to do 5x10 minute intervals at 115% of FTP because you have a lower FTP on the TT bike.


I also have concerns about TR and what seems like hand waving “don’t worry about it” when we talk about things that likely impact “FTP”. Tons have been mentioned, but to list a few common ones:

  • Bike & related rider position (higher rider often leads to higher FTP, while lower rider leads to lower FPT in most cases)

  • Power Data Device Variation. Multiple power meters and/or smart trainers in a rider’s data circulation

  • Inside vs Outside power variation even when the equipment is identical (bike with rider position & power device)

Those three separate aspects and worse yet… combined cases can lead to HUGE variation in power data. It’s hard to take the general statements by TR that seem to say “We have it handled… don’t worry about it…” without some level of reassurance based on some data or clearer statement about how they are dealing with this.

I’ve never heard anything that covers my personal concerns as someone that uses many different bikes, all with unique power devices and rider position. It’s more than a few variables and judging by what we see reported here between data devices, there is reason to be concerned about how TR is leverage that range of data.


Why not try making some custom workouts? This is how I work on time to exhaustion (I am not on plan builder or using AT at the moment but do use the platform). One issue is youre super close to your goal event, so that’s going to make it challenging.

But you could design some workouts that have some sustained threshold efforts (after a warm up of your choosing) and then each time you work on threshold increase the time in zone. You will be able to tell very early on if the bock is sustainable or not. Also, make it at least 10 minutes to begin with, shooting for 13-15 minutes in length.

If you flame out on the first 10-15 minute interval that is not your FTP (or at least +/- 5% of it). I tend to do threshold intervals at 95% of threshold instead of right at it (***but let’s just keep myself honest here. My guess is I am doing these workouts AT my threshold, but I like the higher number for my ego because I hate testing and AiFTP says its 5% higher. But these 95% intervals FEEL about right. Hard but doable, and my heart rate is stable at around my threshold heart rate by the second or third 15 minute block).

Part of this is a mental game as well - doing long sweet spot or threshold intervals is difficult mentally. Get comfortable with the discomfort. I think even beyond ramp testing being inaccurate, aiFTP has a confidence interval - and you might just be on the lower side of it (like me. I am NOT an anaerobic athlete at all, but can hold sustainable wattages for a while).

Thank you for coming to my Ted Talk


I agree with Chad about the hand waving the part that scares me the most right now is how failed workouts are handled.

I recently was not feeling it for a VO2 Max Workout and bailed on the final interval. I know that it may be a shot to your ego but I think in these cases it would be a better idea to decrease the future workouts to avoid consistent failures that may demotivate people further.

1 Like

Remember “TR FTP” =/= functional threshold power, and has no duration component to it at all. TR is very clear this is the case, and that it should only be used in their system to set their training percentages. Follow @kurt.braeckel advice if you want to actually know what you are capable of, and use a capacitive algorithm such as eFTP to estimate your threshold capabilities using a longer duration time input.

1 Like

Agree with a lot of the above. It sounds like your FTP is set too high and you’re doing shorter intervals but not just VO2 length, but possibly VO2 intensity.

What have you been giving as the survey response to those short interval workouts that you can get through?

If you reduce your FTP manually, one would hope it would increase your PL’s, but I don’t know that to be true. You can always just use the “alternate” function and pick something a bit tougher/more appropriate and reset them to the correct number (Lamarck is a good example of something achievable and good as a tester for your FTP).

Given your Power measurement is consistent it seems that should be pretty reliable. Definitely different trainers can feel different, but that shouldn’t change the measured Power in a significant way I don’t think?

1 Like
  • This heads in a direction deserving of a separate topic, but to at least touch on it here since it was mentioned…

  • I firmly believe the function of the flywheel (weight, size, trainer drive ratio, etc.) all can impact the feel AND the function with respect to rider performance. Testing by myself and others with high vs low speed gearing has shown strong RPE differences and some apparent “FTP” differences as well.

  • Depending on the differences in the trainers in use and gearing used on the bike for each, I believe there can be differences in performance on par with things like cooling, motivation and such for FTP and training based upon that. Testing on one trainer, and doing workouts on a different one may well lead to a rider being “out of zone” in some instances.

  • With respect to this rider, even with the identical bike and power meter in use, he may be getting just enough difference in demand from the trainer at the same reported power data, that the functional effect on his body is different.