Comparing actual FTP Test to the AI FTP Detection Feature

I’m a bit cautious by nature and though my ego had a bit of a boost with my recent AI FTP Detection results, I wanted to verify with an actual test, as I felt that the most recent detection was a bit inflated given my recent threshold efforts out on the road.

I did a 20 minute FTP test today and my FTP was roughly 2.4% Lower than my estimated 207 AI estimated FTP.

Got me thinking that it might be interesting to see what the difference has been for others that have looked to verify results with any other FTP tests; wether Ramp, 20 min, mult 8 min. I know the AI is only attempting to estimate the ramp test results and not the others, but i’d just be curious to include all types.

Poll: My actual FTP Test was ___% lower or ___% higher than what the AI FTP Detection estimated:
[Note: the “spot on” is +1% or -1% from the AI Estimate]

  • 1.1-5% LOWER
  • 5.1 - 10% LOWER
  • 10.1%+ LOWER
  • 1.1-5% Higher
  • 5.1%-10% Higher
  • 10.1%+ Higher
    • or -1% from Estimate (spot on)

0 voters

By the way, I am a huge fan of of the AI FTP Detection Feature. Think it is a great addition to the tool set TR provides to keep us both motivated and on track with proper training metrics. Poll is meant more as a curiosity of what others are seeing.

Huge can of worms here. There’s already two threads about AI-FTPD results and I think you’ll end up with a lack of consensus about the definition of “actual FTP”.

2.4% seems pretty close at 207W. Nice work :+1:

5 Likes

Certainly wasn’t my intention to open up that sort of debate. I’m sure there are better threads to sort that out, and I can see some may view it as comparing apples to oranges sort of thing.

Was just curious if others are continuing their FTP estimation protocols and how they compare to the AI generation, regardless of definitional differences.

1 Like

Yeah, it’s a fair curiosity. I think though, you’re going to get a mixed bag. You’ve done a 20min test, lots of people have compared with a Ramp, some still stick to the hour or KM test.

Just wanted to give you the heads up about where your thread may head.

1 Like

I used the new feature when it came out in Feb, six weeks ago. It estimated 311 which seemed pretty reasonable based on what I could do in longer intervals at the time. I used 311 for an FTP block over five weeks. I used the AI feature again last week and it estimated 322. I used that to guide pacing in a TT at the weekend and did… 318 avg (319 NP) for 40 minutes. So I’m sticking with it, for another FTP block at least! We’ll see…

7 Likes

I just tried out AI FTP today and I think I am in the ballpark. Intervals.ICU has me at 207, AI has me at 217 and I haven’t done an actual Ramp in quite sometime because I just don’t find it that necessary anymore. So much can affect my performance on the day of the test.

For example, a regular Ramp will have 1 week of low intensity or maybe even a rest week prior and last fall I tested at 237 and intervals.icu had me at 230. Then Covid got the better of me in December and my FTP crashed and burned. When I started to train again, after a few sessions, Intervals.ICU had me at 186 but I found all the sessions were too easy so I set it myself to 210 and was doing all the workouts very nicely. Hard but not very or extreme.

Today, I did the AI FTP and it showed 217 and intervals.icu had me at 207. Ok, so I set all my TR, Strava, Wahoo, etc. to 217. I thought they should be more similar because this was one block of indoor sessions, no out of saddle and all structured.

Like I said, so many variables can affect how I ride any given day so FTP in a range is good enough for me. Sleep, what I eat the day before or that day, temperature, work stress, sun moon stars alignment…

2 Likes

FTP won’t be static day to day. 2.7% could easily be any combo of factors like what you ate the day before, sleep or lack thereof, hydration, humidity, general feelings of motivation that day, etc… All you can do is all you can do, some dude used to say

1 Like

spot on today, and disappointing.

SSBMVII complete @ 314 - AI detection 308. Actual results 308. Not was I was expecting for 6wks of effort.

1 Like

how much time off have you had, and how many workouts did you fail/adjust?

I became much more relaxed about FTP when I started seeing it as a range. A range defined by several variables. Sleep, nutrition, bike fit, power meter accuracy and the FTP value in itself and so much more. No way there is a fixed number somebody can hold for one hour. A gift from the past when they defined FTP: “what someone can hold for 1 hour”. It isn’t. It’s mostly a range, not a fixed number.
I understand that TR needs a number to calibrate things on but it would be good if they communicate so, that FTP is or can be a range. It would relax people more. It’s something human that box thinking.

So don’t feel down about the difference between 314 and 308 watts. That in itself is my power meter’s accuracy for example.
Be thankful you could ride your bike and enjoy every workout you “can” do.
I use TR for some structure and simplicity but I never worry or let my emotions flow because of some arbitrary number. I don’t lose a race because my FTP (or w/kg) is lower than my peers.

3 Likes

yeah more bummed it didnt go up given the work i put in. Still happy w my current fitness (4.3wkp), and always glad to just go ride. FWIW, I hit all the workouts in SSBMVII without any additional time off, although the threshold workouts were struggles each week…AI dialed them back so i could at least barely complete them, but still very challenging.

Used the AI to estimate and then did a ramp test immediately after. Got the same FTP as the estimate. Was very surprised.

1 Like

So the 314 number was in fact too high. That’s why your ftp went down. It wasn’t your ftp to begin with. Now it’s dialed back to hopefully allow you to get more productive workouts in.

I’m an anaerobic athlete compared to the average cyclist. TR’s ramp test ruined me by overestimating my FTP by 10% compared to the 20-minute test.

Flash-forward a couple years and I tried out AI FTP detection which put me at 303 watts (insert skeptical Fry face here). I went with it for a modified 5-week block and hung in there though it felt very difficult; I was positive I would not be able to hold a threshold effort for more than 20 minutes at a time.

The WTF moment came when AI FTP bumped me up another 7 watts to 310 after the block (insert unsettled Tom face here). I’m two weeks into the next block and I’m calling it quits on AI FTP. The first workouts of the last two weeks were 30-30’s which I blew through relatively easily, but the threshold workouts were absolutely brutal. I barely completed the first workout and I completely failed the second one.

I ran through another threshold workout and I’m fairly certain my threshold is around 270 watts or about 13% less than where AI FTP put me. I realize the TR folks say that FTP is not your “hour power”, but I’ve been training consistently for almost three years now and I know what FTP feels like. I can hold 270 watts for 40 minutes at a 7/10 RPE but holding 310 watts for 10 minutes is an all out effort for me.

I know AI FTP is very new and will get better as time goes on, but I think anaerobically-strong cyclists are better off sticking with the 20-minute test or even Kolie Moore’s test to get their actual FTP.

2 Likes
  • Have you contacted TR support directly, about your experience? Maybe you have, but I don’t see it mentioned in this post. Part of the “get better…” requires us relaying info to them, especially when things don’t go as planned or expected.
2 Likes

I agree with @mcneese.chad: this is the sort of feedback that would be very beneficial to pass on the the TR team as they refine the program.

Thank you for reminding me! I was going to do that but got sidetracked on the forum :stuck_out_tongue:

1 Like

Good deal. This sounds very interesting and is something I think they really need to review. Thanks in advance for getting it to them :smiley:

1 Like

I do think part of the challenge is TR workouts are based off of a percentage of FTP. Regardless what that number is a number of the workouts do have an effective range for us to work in. I dont think it is always the same for all of us for all workout types. It works for the majority but not all.

My own experience with ramp tests has always been I do the workouts a lot better if I decrease my ramp test FTP by a few percent… For me to get the right power for threshold and sweet spot is often just a 1-2% lower amount then what it is set as. VO2 workouts though I have no trouble do a high percent of FTP. I do think that is just how we are built. If I didnt use Erg mode it may not be such an issue.

You’re probably right if “actual” FTP is important to you - but AI FTP might be better at setting training zones for anaerobically strong athletes.

1 Like