I would say yes, if you have proven you can ride for X minutes at Y watts, then the system should know you can ride for X minutes at Y watts. If you bench press 200 lbs and then spend the rest of the month doing less, would you not still record your 1 rep max as 200?
Yes. Any good model should have built in decay if you are not maintaining however.
In the situation you outlined above obviously the ai should give you an ftp from the first effort but if you don’t do any work to maintain that for the next month it would go down.
2 Likes
evaluating after a few months AI ftp seems to work properly for me..
It kept rising in the base and build phase. Then I had the specialty phase up until my goal (AGR 150km). So 2 weeks taper leading up to it. And also a week rest after AGR as I was on holiday. So meant that in those last 4 weeks, my FTP went down 2% (from 249 to 242). Which is fair. Because there was far less volume and a week rest included. +it was the specialty phase.
I had a very similar experience recently. Completed build 2 and nailed each workout. I extended a couple longer and/or increased intensity a little bit. I thought for sure I would see my expected 2-3 percent ftp increase which is in line with my last 4 years using TR. instead, I was given a 1.5% decrease. I tested with a ramp test, unrested, and achieved a 3.1% increase vs my previous AiFTP. This was right in line with what I expected and I accepted this. I have an inquiry out to support so I’ll see what they say.
4 Likes
Would like to add to this topic. I found this discussion after coming to my own conclusion that AI ftp was malfunctioning and opened a CS ticket.
I came back from a several months injury break in January. Sadly, that is almost an annual thing. But that means I have lots of experience taking my off the couch fitness into a new season, and how that feels and progresses. I’ve been feeling really strong the last couple of weeks and increasing workout intensity/picking alternatives. Doing some outside rides that would bury me if I lacked some fitness.
Then AI ftp recommended I decrease 5%. That would take me actually lower than where I started back up in January. I asked CS, they told me the usual, fitness is more than ftp stuff. He said I should accept it and see how it goes. I pushed back on that. CS then got back to me saying they did find a “snag” in AI ftp and that they re-ran it and it’s actually a 4.2% increase. IDK what the details of the snag were.
Between what was initially recommended and then what was re-run, that is a full 10% gap in ftp. Which is roughly an entire zone shift.
So my point here is that it seems like a lot of people are questioning their AI results lately. I have no insider information, but the behavior of the app has significantly changed in my recent experience. And I assume a lot of people like me, at this point, generally know what their fitness should feel like. If you think it’s wrong, get TR to look at it. Don’t blindly follow our new AI overlords.
2 Likes
Define “a lot”. Specifically, how many are complaining and what is the complaint ratio vs. the number of users.
I’m not trying to call you out, but my point is that just because we see some users raise questions about their experience, even if they are all concentrated in this thread, it doesn’t mean there are “a lot”. Confirmation Bias is a real thing.
Conversely, just because users aren’t composing about their results doesn’t mean their results are accurate. They just don’t realize it.
I use think we need to be careful about making accusations based on anecdotal results vs. data. You will always have users of any product who experience issues. But the presence of an issue doesn’t necessarily equate to an overall defective product.
6 Likes
I don’t want to play TR’s advocate, but I’m not sure “a lot of people”. Again, if you look at my profile, you will see that I criticized TR quite a few times, but AIFTP seems to work quite well, particularly if you’re consistent.
I also track my FTP on intervals (eFTP), and when I put in the effort required to have a reasonable correct FTP assessed, the numbers are also pretty close.
1 Like
Except when it blatantly doesn’t. Which is the entire point of my post. And not an opinion, but validated by the engineering team.
2 Likes
No one is disputing your case….we are saying that just because you had an issue, it doesn’t mean that there is an underlying flaw in the program.
1 Like
Again, not saying TR is perfect, and issues might happen with products. AIFTP is a software feed by data. 1 bad data can skew the numbers. If you’re experiencing this, or a glitch in the software, you’re doing the correct thing on placing a ticket. But that doesn’t mean that “a lot of people” are questioning or having issues with the feature.
If you think the FTP isn’t correct, test it. It’s dead simple, and I don’t understand this resistance. Find your limits is part of the game.
4 Likes
I really don’t understand so many users refusal to do an actual ftp test when they don’t trust the number ai is giving them (it’s always too low right?).
It’s pretty easy to go do a proper test, get a good threshold stimulus in and verify the number you are getting.
5 Likes
There has been alot of data provided, and the number of independant observations has been growing. The people that keep saying its working and believe that questioning the aiftp is not done, remains however limited to a few select individuals that are vary passionate about defending their stance. These individuals havent added any data themselves, and only repeat their anecdotes, and quoting training generalities without verifying if they even apply. Some in this topic dont even pay for trainerroad… unfortunatly these repetitive “additions” derail the discussion that some members want to have about the ai ftp.
In my case btw, trainerroad admitted the aiftp was wrong.
1 Like
As I asked above, please quantify those numbers. Because if you can’t, you are just talking about anecdotes.
As for your particular case, no one has made the statement that AIFTP is perfect. As already noted, every product will have issues of reliability, whether physical or digital.
By your definition everything is an anecdote. A clwar trend has emerged here period, with users providing objective data. You havent provided any data of any sort to support these observations are false. You stance tbh is anecdotal, and based on your own bias.
Again no1 is arguing aiftp is perfect or needs to be perfect. The point here is that something has changed and people are getting unexpected numbers, that deviate from previous experience and we want to understand why. Going back to anecdotes, only trainerroad has the database, and we dont have access, so yeah… moving beyond anecdotes of own experiences is near impossible, for me AND you, the next best thing is discussing it openly in a forum and see if users share experiences, which seems to be the case.
Twowkg, i had the aiftp overestimate too, that wasnt fun either. So no not everyone is looking for increases. Btw have you bought a tr subscription yet? Or are you still an opinionated person, advicing others on a product that you have never used yourself?
Ha, TR implemented a ramp test because people didn’t like to test and now they implemented AiFTP because people really really don’t like to test including the ramp test. 
3 Likes
Give me one hard number to define this “clear trend”. Don’t respond with “a lot” or “many users”. If there is a “clear trend”, you should be able to provide a hard number related to it.
1 Like
If you are going to passively aggressively mention me at least have the decency to tag me…..
My only advice has been to do a test if you do not trust what tr is giving you, that hardly requires any in depth knowledge of tr lol. You can either do that or keep giving anecdata.
Also to correct another assumption you have made I have used TR in the past, for many years actually 
.
2 Likes
Many of contributors to this and many threads on these forums dont follow TR plans (not sure if they even pay for the product).
They seem to follow empirical cycling who advocate long form tests, increase TTE, easy days easy, hard days hard etc, big volume. And bang the same drum in most posts, do a test, 3 * 30 threshold shouldn’t be a very hard workout.
The reason I pay for TR is because of AI FTP, PL and the trainnng plans.
Until recently (coincided with the new custom plans changing) i could guess prior to running AI FTP the result within 1 or 2 watts, i would say 100% of the time I would know the direction of the result.
Personally I am at a stage where 5 watts is a big movement for me, normally 1-3 off a good block. Over periods of 6-9 months this is how I get to peak fitness so it does matter to me for the events I ride.
Ai FTP historically worked very well for me in conjunction with PL. The point is not about the absolute accuracy, its only an estimate afterall. Has something changed though, at least users could then reconcile why FTP goes down when adhering to plans.
2 Likes
TR should just not call it “FTP”. And even better, also not “AI” (but guess that´s a no-go nowadays).
And all would be good.
And no, an “AI” can´t ad-hoc guess your FTP out of a few rides with unknown RPE, same is the most advanced human coach couldn´t. It needs some try&error to swing itself in.
So neither 100% correct nor completely wrong. Which is easily confirmed by the fact that they say they are gonna improve it further.
If you want to know your FTP, do a 40km time trial once every couple of weeks. But even there the first few results might be wrong as you have to trial pacing out of nothing.
The numbers are right here in the thread. Alot of users have provided experienes. You have provided 0 data. Why dont you show us the data that we have gotten slower after completing a plan with 100pct compliance?
A few here have the need to keep “attacking” these weird aiftp observations, rather than listening and trying to understand what is going on.
2 Likes