AI FTP detection vs Ramp Test vs. 20 min test

nope, just ignore it and manually set to 285 I think. Still felt that was a bit too high. Then follow mFTP and eFTP and set it to 280.

It was a bit of a surprise that number. And I belive it’s due to last 30/60 days activities when I had good numbers before stoping for a 1 week break. It was around 295 FTP with 312 for 20 min. Maybe some residual from these numbers.

  • That is the reason I asked. It’s at least useful for you to understand what lead TR AIFTPD to give that value. It may well lead to TR learning about an edge case or actual problem that they might be able to address.

  • Main point is that I think it’s always good for users & TR to share these odd results so there’s at least a chance of it getting improved or corrected.


But why would people expect Ai FTP to be correct after breaks? It’s not magical. You have to feed the model just like with WKO/mFTP or any other modeled FTP.

TR can’t program for every case. They don’t know that you took a complete break from working out, just rode outside, switched to mountain biking for a while, or rode a Peloton for a month.

It’s just another tool. Taking a break is like starting fresh so you probably need an FTP test, a good guess, or the 10 Ai FTP rides to get it in the ballpark.

1 Like

I agree that there are going to be limits on where AIFTPD can work. One thing that these cases may point to is a need for TR to filter out these types of issues and directly state to the user that AIFTPD is not possible in that instance. Likely good to give at least some background reasoning so they have a picture as to why, and possibly know what steps they can take to get it back in working order.

GIGO stuff here, but we’ve seen numerous times where AIFTPD will give a rather poor value without warning or exceptions stated (once the 10 inside workouts and whatever are the current requirements are met). Meaning I think there is room for improvement in TR cutting out some examples that seem to repeat in giving questionable results.

OMG are you serious - do you see the Auto AI detect symbol on it? I did not accept the AI value and chose a value other software recommended - ie I manually entered it. Check your facts before you choose to take aim at peoples ‘agenda’.

In my very limited time back on TR and using AI FTP, I’ve found it within 5w of a 20m test.

I don’t use the ramp test any more as it consistently overestimates by 10-15w.

FWIW, coming onto a forum full of strangers and making comments like you did (dubious, wrong, I don’t think so… etc… TBH I don’t recall EXACTLY what you said and haven’t gone back to reference) can VERY MUCH be taken out of context UNLESS you provide some context to go along with those words… IMHO.

I’m not condoning what @MI-XC said, but without some context, the rest of us (as readers of your comment) are left to speculate and in his case, I suspect he looked at your calendar to find that FTP number you entered manually.

That said, as @mcneese.chad and I believe others suggest above, I would talk to support to see WHY AIFTP increased your number. If TR can pick apart outliers like this (i call it an outlier based on my own personal experience with AIFTP… which has only ever gone DOWN when I take extended time off OR even this past summer, when I spent the majority of the time outside and riding for fun and not following much structure AT ALL), they can only make the software BETTER for all of us.

I wouldn’t get too hung up on AI FTP as it’s only one measure of your fitness which prescribes workouts based on your overall aim.

I’m happy with my modest progression 2.5, 2.9 & 2.4% respectively. My progressions levels reset, albeit quite high which means I’m doing 16 minute thresholds rather than a higher FTP and shorter intervals. Vo2 is set low so I use the alternatives option and aim to find a workout close to my previous watt output.

So the harder workouts are providing the adaptations required, the ftp is just a number, used as a datum, for the training on here.

I am expecting a good bump if I return to a manual test but I currently don’t see the need. That is until I return to club rides in April. And that’s just for my ego.:wink:

1 Like

I looked into all your online data on multiple fronts and stand by my assertion. Either AIFTP completely bugged out and you are a one off scenario or you’re being less than truthful. The below appears to be a disingenuous jab.

Maybe you misread my message. I didn’t “ask” for an AIFTP, it prompted me. Plus, I did a week of tests consisting of 5s, 1m, 5m, 20m.

I didn’t “want” a detection, I knew my FTP. Others maybe not.

I don’t think I replied directly to your post. I was replying mostly to the general tone claiming that AI FTP is “dubious” (dhi67540) because he got an inaccurate ftp after a 4 week break. That is exactly when you should expect a software model to be wrong. It’s just a tool (good or bad) but at least you have to use it correctly.

In your case, you can just turn off AI ftp detection since you use WKO.

1 Like

I just did a TR ramp test after doing the festive 500 and on the back of a solid recovery week. I also have an updated eftp based on some longish efforts during the festive 500. Going into the festive 500 my AI FTP was 236. During the festive 500 gave me an updated ftp of 239. My ramp test at the end of the rest week following the festive 500 gave me 240. So…n=1 I believe for me they align very well.

1 Like

AIFTP doesn’t rely on training either. I didn’t do any training activities for a while, and still get numbers back when they are available. I may or may not use them, depending on how I feel.

Ramp is pretty accurate for me. I sometimes use the ramp to get a rough idea of how much I have lost and then maybe to a more Kolie Moore style later, which I prefer.

For those curious folks: began the 2024 season December 18, 2023. Began after a month or so of lower training load followed by a week completely off the bike. My process this season is to run AIFTPD on the Monday it is available, not accept it, and perform the 20 min test the following day. I like the test, it seems to give me very good zones to train around.

So, AIFTPD on Dec. 18th gave me an FTP of 313 - pretty much the same value I’d been training on the lower volume with. 20 min test the next day: 296.

Yesterday, Jan. 15th AIFTPD: 303
20 min test today: 308.

Seems to be quite accurate when detecting growth, but a bit lacking when detecting decay. I’ll follow up as the season progresses.


I like your method - I know the 20 minute test and even ramp test are hard, but on the other hand if we are competitive athletes then executing a test to find out how strong we are is going to be tough. To me it’s part of the process.

1 Like

Re-using this thread not to open a new one…

AI FTP got me at 275W @71kg
I don’t feel that is accurate (a bit too high) but i want your feedback based on last two days performances.

Saturday: 30 min TTT (4 people) - my first ever TTT on a TT bike and second time ever on a TT bike. Here I felt I could push more, but i was not confident at all on the bike.
269W average
281W NP

Sunday: 1:07 HH Crit race - of which 1 hour spent in 4 guys breakaway
251W average
283W NP
(manage to get the point classification jersey :smiley:!)

1 Like

Nothing there to suggest it’s off.

Maybe try 2 × 20 at 270 watts and see how you go? RPE of 7 or 8 should seal the deal.

Or try one of the @empiricalcycling FTP tests.

1 Like

Update after latest block:

20 min Test: 318

Pretty impressive. Next update in four weeks.