Yeah looks good now - but should have been right on day one like you say.
I wonder what it was about your training history that tripped it up? 
Hey there! I took a look at your account.
You don’t have a ton of training data for the AI to analyze. Your oldest activity is from August 2025. Additionally, looking at your power numbers before your FTP detection, I don’t see much to indicate to the AI that your FTP was higher than it thought. Your all time highest 20 minute power is 226w from a Zwift race in November marked Max Effort. More recently, the week before your FTP detection, you did another Zwift race marked Max Effort where you did 218 NP/ 207 Average power for 58 minutes.
Based on this relative lack of data and the wattage of what look like maximal efforts, the AI did the best with what it had available to it. Now that you’ve got that Zwift Ramp test and some other structured workouts under your belt, the AI is rapidly adjusting its understanding of your capabilities, as you observed.
You’re scheduled to have an FTP bump in a few weeks. In the meantime, if your workouts feel too easy, you could certainly choose harder alternates or manually increase your FTP a bit.
17 Likes
I had a Sweet Spot 5.2 i think and bumped up difficulty into the 120%+ range as it was way to easy and wrong. I completed if fine at level 10.3 , and marked it as easy.
Now the metric for difficulty has dropped in the detection down from 241 to 226, this is really really stupid like what AI slop is in the backend, as i’m now have 2 options. Either I have to sit for another 28days with a wrong FTP (bear in mind we are already doing that now and have been for weeks) or whatever the mteric is which sets difficulty and manually set each and every workout to the right level…….pointless being subbed to this fitness tool.
Or I can manually change the “number” and then not get the beinifts of the AI tool and product i’m paying for.
can you just set my number to 243 in the backend and we can get back to working out at the right intensity to drive fitness gains, and not have to spend time showing you it’s broken.
@eddie @SeanHurley- see above
1 Like
I’m also in the situation that my FTP was estimated too low. I have tried three approaches:
- High progression levels (9.7 Threshold)
- High intensity (110%)
- Normal workout, low RPE
I have the feeling that 3. affected the predicted FTP the most. Although there might be confounding factors at play, so this can also be incorrect.
2 Likes
Also it’s odd you get a red day tomorrow despite rating today’s workout easy.
1 Like
Looked into your account and there’s an edge case that was causing that 226 FTP Prediction. We’re actively working on addressing this.
Our AI model believes your FTP is 241 today. That seems like a reasonable FTP to me given the Sweet Spot workout you completed today, so I applied that change to your account and you’ll see it used for your upcoming workouts. If you want it to be 243 you can manually set your FTP to that, but since it would be a manually set FTP, FTP Predictions wouldn’t work until you used AI FTP Detection again. With AI FTP Prediction enabled now, the model is predicting further gains over the next month as you continue training.
My suggestion would be to go with the 241 as 2w is a minimal difference and your workouts will be better calibrated either way.
This should smooth things out after the bumpy start you’ve had. Sorry about that!
13 Likes
Survey responses are incorporated into Fatigue Detection, but they’re far from the only factor it considers. In this case, this workout set a number of recent PRs, went way above target, and was triple this athlete’s average daily TSS, so it’s not surprising it’s anticipating some resulting fatigue.
7 Likes
Thank you for the change to my account, i’ll leave it at 241 and can continue to use the platform now without having to manually adjust anything. Thank you.
8 Likes
You’re welcome! I’m glad we got things sorted. Let me know if you need anything else going forward.
4 Likes
I have been surprised with all the comments about too low Ai ftp. Mine was the opposite I had an increase to 356 but that was at start of three week off period so I held out once back I set up a new calendar training block and it dropped my start fto to 333 I think it was. I new I didn’t loose that much with three weeks off so re set to 346 figuring would just work up from there until next detection. Then Ai ftp came out two weeks later and it put my ftp up to 362. I struggled on workouts having pause time to finish but with short breaks could finish. I kept it at what it was just to see how would work. It has adjusted me down and workouts are getting to be more accurate and pushing me right up to my limit. Looks like will adjust me down a bit at next detection to that 356-358 range which should be pretty accurate.
To me it’s been fun to see how it interacts with my workouts. It would be nice to have the option to add a note into it such as last week I was fighting off some illness junk my kids brought home and could feel it was affecting my workouts. Like the note if you have to adjust intensity or max effort.
Sorry you’re having problems with it. I think how you’re trying to “cheat the system” isn’t actually hurting you.
I also think you need to take a close look at your actual performance. Looking at your NP vs AP curve, your NP is being artificially inflated based on a KILLER short power. This is a known issue with NP with the 4th power raise that’s in the calc.
On Dec 14th, you set your FTP to 245.
Your best 6-week AP before that was 225.
Your best 20-min NP for the 6 weeks is hard to find. I see 264 and 251. The first one being a “recovery” ride where you did a monster sprint and it put your watts high. I believe 251 is also overestimating your abilities.
You’ve been having problems with sustained power (I’ll show below), and I don’t think you should rely on NP to find your FTP if you’re doing big sprints because your impressive anaerobic power will override it.
Dec 14th - FTP: 245 (Self Selected)
Best 20 min AP: 225
Best 20 min NP: 251 (Inflated)
Dec 22nd - FTP: 245 (Self Selected)
You self-select Monitor -1 2.5 Sweet spot, and you failed it. That’s 5x6-min @ (88-94% FTP) with 3-minute active recoveries.
The AI would see this as evidence that your FTP is not 245.
Your next workout with real sustained power is:
Jan 12th - FTP: 235 (Self Selected)
You self-select Geiger 4.1 sweet spot (3x12 @ 88-94%)
You fail that workout and score a 2.9/4.1 and mark it “hard”.
The AI would see this as evidence that your FTP is not 235.
Jan 21st - AI FTP Detection 212 FTP
Jan 22nd - You do a ramp test
I’m assuming you didn’t like your 212 FTP
You score a bunch of all time power PR bests during this Ramp test from 2:45-10:00 mins.
Jan 23rd - AI FTP Detection 212
You self-select Crested 2.6 threshold and overperform to a 2.9 and mark it easy. The crazy thing is that we would have given you Sill, 4.0 threshold!
Jan 27th - AI FTP Detection 212
You self-select Mogollon 8.0 sweet spot 2x28 @ 94%.
You fail that workout but say it was due to “equipment issues”. I can see that you jumped ahead during the last interval and then turned down the second interval. The HR where you bailed is also similar where you bail on other workouts.
Jan 27th - AI FTP Detection 212
You self-select the Spickard 7.1 threshold, and you fail. You end up turning down the intervals during the first set.
We would have prescribed you University 5.9 Threshold that day. You mark it is moderate, which I think you’re trying to trick the system with, but you’re also not completing it and your HR is pegged for you.
Feb 3rd - AI FTP Detection 212
You do AI recommended Galena -1 5.2 sweet spot. You GREATLY outperform this one and do 240 watts for 20 minutes, which is 17 watts ALL TIME power PR and mark it EASY!
Feb 3rd - New AI FTP detection of 241
So here’s what’s happening:
AI sees:
- Has problems with 213w for 6 mins
- Has problems with 214w for 9:30
- Has problems with 223w for 13 mins
- Has all-time 20 min power PR of 240w and calls it EASY
It also knows that even if you have problems with a ride, you’re not likely to mark it hard/very hard; so it’s predicting that you’re going to say rides are easy/moderate, even if you are also likely to fail them.
I actually think that’ s a cool thing about the model 
.
The ironic thing here is you’ve failed all the rides whe
You’ve literally only done one workout we recommended, and you have an obvious spindown or calibration issue with your power PR on Galena, given your 20 min all-time power PR (by a lot) that felt “easy” to you.
I suspect the slope of your trainer might be off and that’s why you’re also getting gigantic sprint numbers. I looked at your account and your power PRs for outdoor rides are a lot less than when you’re indoors (it’s normally the opposite).
I recommend getting a new indoor power source.
41 Likes
Could this be a case of the athlete erroneously basing their RPE on the original workout - not the actual work that was done - or is it an obvious power measurement problem based on heart rate?
A good example of stick to the plan? 
5 Likes
I don’t think so since it’s a big all time power PR and he does Zwift racing often.
2 Likes
I think you may have an unintended double negative here?
The rest of the post implies that they should follow the plan and not “cheat the system”…. and to get a PM that works….
Or are you saying that the system is still working in spite of all their messing around?
EDIT: and shouldn’t you be asleep now?
Hi Nate,
Firstly, thank you so much for looking into this. Your insights have been really valuable and genuinely appreciated. It’s cleared up a lot of confusion on my end.
I’ve also now come to real realize that I am a tricky profile for the system to work comfortably with. And that’s down to my inexperience with structured training when cycling, i’ve not been a good candidate doing actual maximal efforts deliberatly in a strutured manner across the different time periods which woould have left a useul data trail for the system to work with. Up until January i was just doing lots of Zone 2 with some thursday team time trial races on swift.
I hadn’t really touched a bike for about a good eight years, since August last year and i lost 33 pounds before doing that while quitting smoking in June 2025. So i was about as primed for newbie gains as I could possibly be which again i don’t think, makes me an easy profile. My first ramp FTP test was 185watts in August and then on 22nd January it was 249watts, so i’m not a stable profile.
I also have only done 13 rides outside since August and 1/3rd of them was with a 4iii power meter which wasn’t great, 1/3rd was done without one in spain on a hire bike and since demcember it’s been with a Magene PES 515 which is within 2% of the Tacx Neo 2 trainer so again different data feeds.
All the way up to January I was just doing base with the odd TTT which wasn’t maxiumal stuff and a very ocasional other race but nothing in a overly structured manner, and I wasn’t really focusing any effort or commitment into threshold, sweet sport or even VO2 level workouts. I think it was the perfect storm of a large amount of newbie gains due to getting back into cycling, stopping smoking and spending a lot of time in zone two with the odd sprints here and there to relieve boredom. The only real maximum effort stuff i was doing was ramp tests every 6-8weeks.
2025 training
2026 training
That caused the issue when I did go into my build phase in January and start to work on zones above 2 and get into the harder workouts, the AI FTP detection which was setting the intensity level was quite a lot below what I required as stimulus. it was a kinda perfect storm of cool new feature release and me changing my training to ramping up into intesity at the wrong time.
I can now see as you’ve shown based on the data from the reply, from the riding i’ve been doing the model would’ve assumed correctly an FTP for setting the workout intensity to be much lower than my actual capabilities are at the moment. NP is useless for tracking actual realised power, I’ve basically been sandbagging my fitness because the data points the model needs like an actual 10 minute, 20 minute, 40 minute maximum effort I haven’t done, or even an hour maximal effort which I have never done, doesn’t exist for it to use as a data point.
Even that last Galena-1 workout I wasn’t at max for that 20mins, my max HR is 198 in december 9was 200 in September) and i came of the bike thinking i should have gone harder and could have done 255+ it wasn’t near exhaustion at all.
I also think you’re right and how I was trying to "cheat the system” although I wasn’t completely trying to cheat the system, I was more trying to trigger the next AI FTP detection to be at the level I feel is required for the right intensity workouts. And my goal was to make sure when the AI FTP detection triggered, it reset correctly for my annoying profile. As I didn’t want to get locked into another 28 days of workouts that wouldn’t have been quite right with only 8 weeks till the first important event.
Again, thank you very much for your reply. I wasn’t expecting you to reply as the support team did solve the issue, but I am very grateful for your engagement and time on this. I like what you guys have built and I am excited now for the next couple of months of structured training on trainer Road.
I’ve signed up for 7 events all mostly all 127km or 164km centurys so I am putting all my eggs in your basket to be ready from them this year!
7 Likes
Just to add some detail and I think confidence in TR AI. I’m an ex-300FTP person on TR 20min tests of the past. I had worked up to a 250FTP in early Jan this year but was reduced to 235FTP with the new AI upgrade. I was a bit unhappy obviously but read the TR comments about this and accepted the new reading. What I can say is that yes, I’ve completed all my Hard interval sessions but they were ‘Hard‘ and some just do-able. So I think TR have got it spot on and I have a predicted 241FTP in a few weeks. Yes, I’ve a 300FTP goal but the fact that I’m close to my training limit but still able to progress is goal really and that 300 figure will come and or I’m getting fitter.
8 Likes
I have a very si.ilar problem and it is not “solved” as noted above, wasting a months training because AI can’t generate correct workouts is a fundamental issue caused by awful software design.
My AI determined FTP is around half the correct value. The app won’t create workouts based on a manual FTP input. The rides I’m offered are totally useless. I could easily have done the so called “Sweet Spot" workout -supposedly just beliw FTP - without breaking into a sweat. I had to manually increase the effort by 175% to get a workout but after completion it nothing changed. It was totally ignored by AI. Why should we pay for this - I don’t want to spend a month training with the wrong workouts.
Most riders will end up with a wrong FTP at some stage e.g. not completing a ramp test due to some issue, illness or injury. Please don’t tell me to do another ramp test, I’m fed up with being given excuses or suggested workarounds. Please get your software team to actually FIX THIS FUNDAMENTAL FLAW. It’s not acceptable to waste a month’s training just to allow AI to catch up.
