Hear! Hear!
4 Likes
Cool, so the solution is to completely ignore ai FTP and do a ramp test then. Glad we got there after ~140 posts.
3 Likes
Or… we can investigate the issue and optimize the AI/ML learning model, or you can keep trolling posts for a TR feature you have never used youself per your own words 
. And derail a thread.
2 Likes
You are best off working with TR directly in that case, they are the only ones that would have the actual knowledge of how the model works.
All we can do is verify it on an individual basis.
1 Like
I think their forum is a good place to start.
Seems to be working excellently so far.
1 Like
Even got some 0.02 from people that are not with TR, and never used the features to chime in for free!
1 Like
Well how about TR take the necessary steps to rectify a clear problem with their product from a paying customer.
The entire point of AI FTP is to avoid tests, thats what paying customers are being sold.
3 Likes
I heard that too, and personally I’d be perfectly willing to accept a new “better” model.
2 Likes
%
Anything less than negative 1% FTP drop should be removed from the algorithm! It’s meaningless and does more harm than good. It’s not a reflection of reality and well within the margin of error.
2 Likes
If you did 4x20 at whatever you think your ftp is, I think you already have done enough to be confident in your estimated FTP for training. I do think you maybe misunderstood the “take an FTP test” comments and thought people were recommending a Ramp test though. I saw some of us recommending something like “a real, long form FTP test”, but not many recommending a Ramp. There are some people who a Ramp works great for, but a lot of us have issues with the results, so I’d be surprised if there were lots of recommendations to do one. I could be wrong though, and I haven’t gone back through the thread to count.
Again though, if you can do 4x20@FTP, I think you can pretty confidently say you know your FTP better than the AI does.
Side note: I understand your frustration. Really, I do, and I hope my previous comment shows that. I do think you’d be better off not arguing with folks who are trying to help you though, even if you don’t like the advice or the tone you “read into the comments”. Some folks come across more aggressive than others, but the intention is usually good. However, it’s the internet, and things escalate quickly when we think people are being rude. You can always block or ignore people if needed, if nothing else, for your own mental well being.
4 Likes
I can see why some in this thread are confused/upset about AIFTP. FTP is metric put forward for success, and if you’re “hitting your marks” it should continue to improve, especially if PLs are increasing. Maybe showcasing other metrics in the app, something like a modelled PD curve, or net improvement points per PL (accounting for drops with FTP increases)?
3 Likes
284 to 288 is a 1.4% drop. To me that’s model error. AI FTP is a model after all and not perfect. It’s good but not perfect. My general thought with AI FTP, if it’s within +/-2% of my current FTP i consider it model error. To think of it another way, it’s day to day variation. As in, one day you might be able to do 284watts for 60mins and another day you may be able to do 288 watts for 60mins just by chance holding everything else constant. There’s just natural variation.
What I do is I focus on my goals and if my progression levels within those goals are improving. As an example, i’m trying to do more TiZ in Sweet spot. If i’ve gone from 60mins TiZ to 90mins TiZ then I don’t care what my FTP did b/c i’m accomplishing my goal and i can see that in my sweet spot progression level.
2 Likes
Again, anecdotes vs. data. Has a “clear problem” been established? There are outliers for any FTP test….see above comments re: ramp tests, etc.
A handful of people have questioned their AIFTP numbers….heck, I have a couple of times myself. That doesn’t mean there is an issue. If there is, then yes, I would expect TR to address it.
2 Likes
I stopped replying once TR said there is nothing wrong with the detection but I will give my final 2 cents.
The problem: TR says it will raise your W/kg and you will not have to do FTP tests because the AI is super accurate. It is the basis of their marketing. User follows program to a T, PLs improve, user hits longer intervals, better power averages every week. AI FTP detects LOWER W/kg.
That is a clear problem in MY opinion Other may not agree and that is okay. Goal of the platform did not work as intended for me (and others actually). I raised a concern and TR came back saying there is no problem, be happy because your 1 hr power improved. I will look at other options going forward.
You cannot have a marketing platform that preaches raising W/kg but when it does not, its okay because its not all that matters. This is only MY opinion because I am more of a black and white person and people will dissagree with me but you also cannot say that there is nothing to see here.
Thanks for all the chats and sharing of your own experiences and thoughts 
10 Likes
I reckon your ramp test results are wrong. Only way to be sure is hold that FTP of 280 at least 40 mins.
1 Like
I don’t follow much of their marketing so maybe I missed it but I’ve never seen them claim to raise your w/kg. Most of the marketing I see are claims to make you faster. Being faster is comprised of multiple variables including W/kg but also repeatability, TTE, etc
3 Likes
I mean they have no control over how much pizza I eat…
9 Likes
I also appear to have no control over how much pizza I eat. 
13 Likes
I mean…does anyone? 
1 Like