I moved your post under the existing one that covers the topic and has many other examples for comparison.
3 Likes
question - how does this feature work? if I set my ftp 30w lower and then I ride a workout at 90% ftp, how would TR then know that I could way more? So if I always ride workouts below my “correct” threshhold and underperform, how will TR know that I would be capable of doing more?
A lot of the “how it works” is what we like to call “secret sauce…” In other words, we can’t give away too much info on that. 
We used Machine Learning (a subset of Artificial Intelligence) to analyze millions of rides to understand how FTP changes over time and in response to training. Our system looks at your personal biometrics and the unique details of your recent and past cycling data to detect your FTP without the need for maximal efforts or specific workouts.
The feature works best with accurate data, so making sure your FTP history in TR is correct is important to make sure you get the best results with AI FTP Detection. Setting your FTP 30 watts lower than it should be would skew your future AI FTP Detections to be unideal.
Hey. Thanks for the reply explaining it. I should have elaborated more on why I did ask this.
I came back from an injury so I did not ride a lot for one year. So my ftp was kind of at 230-250 before but now after almost one year of break I think it’s more likely to be around 185ish. TrainerRoad calculated it with 230 based on my past rides. But they/you did not realise they are about 1 year ago and some of the better performances even longer. So you should include breaks and low volume for longer periods into your calculation.
1 Like
Got it. In some instances, it may still be a good idea to do a Ramp Test or 20-Minute Test – coming back after about a year of less training due to injury would probably be one of those cases. That year acts as a pretty big gap in the accurate data AI FTP Detection needs to function properly, and, of course, the algorithm won’t know if you’ve been recovering from an injury.
Once you reset your FTP with a benchmark test and get through a few weeks of training, AI FTP Detection will get you dialed back in without the need for further fitness assessments!
As some say; FTP is a range of 20 watts up and down
No way.
As with other physiological determinants of performance (e.g., VO2max), maximal metabolic steady state (for which FTP is a surrogate) is subject to both day-to-day variability as well as longer-term changes due to, e.g., training. Nonetheless, as with VO2max, such variability is quite small, i.e., only a couple percent.* You should therefore be able to readily determine your FTP to the nearest 5 W. If you can’t/aren’t at least that confident in your estimate, you should revisit your test protocol and/or question your equipment.
*This is partially where people go wrong in holding up lactate testing as the “gold standard”, as in fact it is more, not less, variable than just measuring performance.
4 Likes
I’m one of those people where AIFTP is pretty spot on. But, I also do a lot of trainer work. I’m probably going to be exclusively indoors on the trainer and in the weight room until April. I think workout execution is easier that way, and I think execution of the workouts correctly plays into the accuracy of the ML models that calculate AIFTP.
I will say, having done MMP tests from 30s out to 20 minutes, and some of the Kolie Moore test protocols, and some high% of FTP hour efforts over the last year, I think there’s a ton of benefit to doing them regardless if you rely on AIFTP. They’re good workouts done in the right place, and the ability to pace and “feel” your limit and FTP is a skill that takes a little work / experience.
7 Likes
AI detection told me 234, did a ramp test and got 237, pretty damn good estimation by the AI, think I’ll just use that from now on instead of doing the ramp tests
2 Likes
Finally had a chance to compare FTP detection vs a ramp test this morning, despite swearing I’d do it shortly after it was released.
Barely rode the past year, toddler, work, moved etc. So when I started up a week ago on TR again it dropped me down to 237 womp womp… but fair. After one easy endurance ride where I had taken a few breaks due to not being used to saddle on trainer bike dropped me to 235.
Scheduled new training plan yesterday to start this morning and for some reason it would not give me a swap out ride for the ramp test. Obviously I could have hit train now but said whatever lets just try a ramp test, 236. I also started doing squats again last week so I didn’t think it would go well a few mins in, but having 235-237 in my head helped me hold on until it was the lungs that gave out not the legs.
Of course when I got to work and went to check something on my TR calendar it said hey want to skip ramp tests and suggested an alternative, but it was a few hours late and glad it forced me to actually compare.
4 Likes
Nice @tmw444 and @mrtopher1980 
!!
2 Likes
This is an interesting article: Hour of power: Do we need a better way to benchmark fitness? - Escape Collective
Not sure why, but It seems as though protocol with the 5 min effort before the 20 min test is forgotten. Thats the way I learned it 20 years ago, and thats what im sticking to.
@The_Cog correct me if im wrong, but didnt you always say 95%, but could be lower
Estimating FTP from 20 min power is Hunter’s approach, not mine.
That said, on average 95% of maximal 20 min power agrees with the WKO4 model prediction.
The “on average” part is one reason why I have never been a big fan of this approach, because why go to all the trouble of a formal test when you can (mis)estimate FTP just as (in)accurately using other, informal approaches?
The other concern I have with the approach is the 5 min “blowout” effort. The one study of which I am aware that tested the effect of this surprisingly reported that it makes no difference. However, if you truly do go all-out for 5 min, it’s hard to pick yourself up off the floor, much less follow it with a decent maximal 20 min effort. So, how hard should you really go for those 5 min, and what if the quality of that effort varies from one occasion to another?
Finally, re. the link you posted: at least based on the title, it appears that the authors don’t really understand the concept of FTP. I therefore wouldn’t put much stock in the actual content.
ETA: Sorry, I couldn’t help myself…since when does multiple regression count as “AI”?!?
4 Likes
This is exactly my experience; the only time I tried the 20min FTP with a 5 min blow-out I was so full of lactic acid afterwards that my legs were all wobbly and totally useless for a long time afterwards.
2 Likes
I have always wondered about this and basically concluded that if you’re doing an “all-out” 5 minute effort, you shouldn’t be able to do a representative 20 minute effort within several hours of it. Consider: racing a mile and 10 minutes later racing a 5k. That 5k ain’t gonna be fast.
2 Likes
This has always been exactly my issue with the 20 min test. For me, 20 min test done shortly after a 5 min TT has nothing to do with maximal 20 minute power. For the 20 minute test workout, TR sets a recommended level of that blowout effort at 110%. I can hit 105% of my FTP for 20 minutes after that. The problem is that I can do more than 120% of my FTP for five minutes. Shortly after that, my 20 minute test results in an “FTP” that I can hold for a couple of hours when fresh.
1 Like
Yeah… last couple of times it has been way too off for me. Yes, I’m more of an anaerobic athlete (puncheur/sprinter). Last time for me:
AI FTP Detection (5 days ago, middle of rest week):
TR ramp test:
zwift ramp test:
so yeah… “5.1 - 10% LOWER”
one month earlier it was something like this as well but proportionally a bit lower
So, sorry, but I don’t trust at all in AI FTP Detection Feature. Also, I believe that is the main reason why some people burnout with TR plans. Probably I’m an edge case…
Nonetheless, Love 
TR and recommend it to all of my friends. Also, probably one of the persons that listens to the podcast more often.
edit: maybe I just suck at ramp tests and maybe I just got unlucky with the timing, both times (bit under the weather). This was probably only my 3rd ramp test ever.
Hey there!
We just took a look at your account to see what might be up.
It seems like you have a lot of life stress going on around your workouts that may negatively impact your training sessions. You mentioned yourself that you were sick for your Ramp Test, which almost certainly impacted the results of your fitness assessment.
This is actually a big reason why we recommend AI FTP Detection instead of Ramp Tests or other fitness tests. AI FTP Detection looks at all of your training data over time, not just one single point in time like the Ramp Test does.
If you have an off day when you do the Ramp Test, it can have a big impact on the FTP value derived from that test. AI FTP Detection, on the other hand, doesn’t run into this issue as it can see the entire picture of your training.
If you haven’t tried any sessions at the 246W FTP value you got from AI FTP Detection, we’d encourage you to give it a go! You’re probably more capable of completing the work at that FTP number than you’re giving yourself credit for.
5 Likes
Yup! You guys were spot on! 
Yes, I have a lot of life stress going around my workouts. I think I’m just constantly hoping that I have better times. It’s called “denial” 
I would like to thank you guys at TrainerRoad for reaching out here and on my email to the support team with all the attention to detail on my current circumstances and for correctly analyzing what was going on. Especially since life is far from anything so “black and white” situations. I was pleased to find that you took care to all the detail of my life surrounding said ramp tests and giving the best advice that could have been given for each topic addressed in such details.
That being said, I have indeed tried the sessions at the 246W FTP value I got from AI FTP Detection and I have found it was just right for those sessions: productive threshold workout → hard/very hard; endurance workouts → easy.
For this reason, I have changed my vote above accordingly and I am looking forward to my next workouts and AI FTP Detections
Like discussed by email: The best approach is to try the AI FTPs detected and answering to the post workout surveys as honestly as you can so the workouts can be best adapted.
Also, be very mindful of life stress, sickness and calorie restriction (even on recovery days/weeks).
“only train when you truly feel up for it, and don’t hesitate to use Workout Alternates if you’re not 100% up for the workout that’s in place for the day. People who carry a ton of life stress like you and don’t recover well often need to know when to throttle intensity in order to continue to train sustainably.
(…) Knowing how to adapt to your plan helps as well. This means not taking on workouts (or, even worse, FTP tests!) when you’re sick, tired, or stressed out/unmotivated.”
4 Likes