Your AI FTP Detection Results

AI uses all power data for FTP estimation, just not progression levels. I think there are some caveats in the estimation for you aren’t doing some TR structure but I don’t think we know which one will produce more accurate results currently.

As of the last official statement (from memory) TR AI FTP Detection needs a minimum of 12 completed rides/workouts imported in your TR calendar. There was some confusion as to whether that must be TR or other rides, but AFAIK it includes ALL ride types, not just TR ones at this time for the minimum of 12.

Other than that restriction, TR AIFTPD is supposed to use ALL workouts with power data as part of the Detection background data. All that said, if you meet the requirements above, using AIFTPD is what I’d suggest.

Essentially, look for the AIFTPD option in your Career page on the Ramp Test, and if the option is available, you are good to use it.

1 Like

Thanks Chad. I have used AI FTP detection a couple of times when I was doing predominantly indoor workouts and I was happy with the results. At this time of the year (in the UK) most of my riding is outdoors so I am not sure whether these rides have an impact on the AI FTP detection

  • Yup, they are included in AIFTPD, so you are good to go.
  • This is true, but ONLY related to the Progression Level side of AT, not AIFTPD.
1 Like

As of the latest update, you can run AI-FTPD from the Calendar page which is a nice little change.

1 Like

Yup, I saw that in the release notes last night after the update.

  • I refrained from referencing it here since I don’t know if that is just Win or if it’s available on all platforms.
1 Like

I only use Android so that’s two out of :rofl::ok_hand:

1 Like

I haven’t been on the trainer much in the last couple months (unusually good spring for Nova Scotia), and AI FTP D seems way off when looking at just my outdoor rides. I used it last week, and it gave about 15W less than what feels like threshold to me right now, and 10W less than before my riding switched to mostly outside despite my volume doubling (and performance improving). It seemed bang on when just looking at TR workouts.

I suggest contacting TR support for review if you are questioning the results.

I decided to merge this post with the prior topic that covers the same basic concept.

1 Like

I would “expect” AI FTP detection to do worse with just outdoor rides compared to having indoor rides, for a few reasons:

  • For indoor rides, it knows what effort level the ride is supposed to be at
  • it has feedback on how hard the ride was, and if you completed the intervals
  • TR is likely focusing a lot of effort making sure this works for indoor riders following their plans, and these don’t include many all-out efforts. The point of this estimation is avoid the ramp test, so those go away too.
    It has a lot more context to work with for the indoor rides than the outdoor ones, and I think the feedback of how hard the intervals were is important. If you did 20 minutes at 250 watts outdoors, TR doesn’t know if that was all out, or just sweetspot level effort. Indoors it gets that, and knows about what certain workouts ‘should’ feel like. Also, since the indoor workouts are generally shorter and mostly of the same interval type, the rating you provide is likely much more valuable than a rating of how hard a 5 hour outdoor ride was that had some hard sections.
    Most/all other FTP estimation is done based on all out efforts and the power/duration curve of those, and in that way they avoid needing feedback - it is assumed that all values are maximal efforts.
2 Likes

Grateful user of AI FTP Detection.

Been working my butt off and then I was recently sick for a few weeks, which hampered my training. My FTP went down 2 watts from 274w to 272w. Honestly that seemed about what I would expect.

1 Like

Great stuff!

I’ve been totally unsure what to do with AI-FTPD and sickness. I’m just starting to get on the bike again after a broken three weeks and then Covid. AI-FTPD has indicated a drop of 2 Watts, I didn’t accept it because to be honest, it currently feels like I’ve dropped more like 40.

Figure I’ll ride easy for another week or two and then see what AI-FTPD has to say before doing some Sweet Spot.

Managed to get AI FTP Detection going yesterday. Got a not inconsequential 4 Watt bump to 324W.

That’s following going 328W when AI-FTPD first came out, then declining 323W and 320W during SSBHV (I assume refinement of the machine learning). Dropped to 318W following Covid and finally bouncing back up.

PL dropped 1.0 down to 4.4 which seems ok if not a little low. I’m really keen to see how Build goes with the combination of Detected FTP and AT. I will still do some careful usage of the alternates to avoid digging a hole working above my MLSS too much. Hopefully the combo can help me balance that line between pushing my limits and keeping a lid on fatigue.

1 Like

It sneaked up another 5w for me last night, probably because all my TT’s are sub-optimal and I think that is what AIFTP uses (sub optimal efforts). I need to figure out how to get an optimal effort out :thinking:

1 Like

I think you mean sub-maximal. Or at least I hope your TT’s are going as optimally as is possible.

Great you had a bump. What was your PL drop for SS/Thresh?

2 Likes

I’m going to out myself here as a novice/untrained/non-competitive rider.

I’d just started doing in some indoor rides in response to the Oz winter when I stumbled upon changes to AT and FTP detection. So I started a new training plan and enabled AI detection.

Using detection in place of my initial ramp test put me at 144 watts. Up 3 watts (2%) from previous ramp test. This was based mostly on outdoor rides. I was initially concerned to see the increase because at the time I felt like my fitness was close to rock bottom. As I continued training I felt like the majority of the workouts were right on the money and a few were a fraction light. AT kicked in to up the intensity and my PLs nicely.

At the start of base 2 my ramp test wouldn’t let me do detection due to a technical glitch so I rode it to get -1 from my previous detected value. Later that day with help from the support team I was able to do detection to get 152 (+8) watts (5.5%)

I haven’t done enough workouts since then to be certain but I’m confident that AT will deliver workouts that make sense for my new FTP.

Does the AI give accurate FTP estimates? Don’t know, don’t care. Do the ramp test and detection give the same result? For me detection is 2-6% higher. Does TR set well targeted workouts using AT when detection is enabled? I think it has done a fantastic job so far.

5 Likes

I have been relying on the AI FTP detection after the initial ramp test or two when I restarted with TR after too much time (and beer and chips and cake) off the bike.
I did the MV progression from base I and II to SusPB with nearly 100% compliance and failed one workout (an ambitious threshold breakthrough almost immediately after a FTP increase). So I was pretty much satisfied with myself and thought that AI FTP

At the end of build, the FTP number I got from AI was lower than I expected it to be (or what my reading of the power vs HR on intervals.icu suggested). it was kind of disappointing.

I took a week off reading TR forums HV threads and polarization threads.

And then did a ramp test. I had gotten to doing 4 x 4 Vo2max workouts in the course of build so it was not all wheezy and winded and horribly uncomfortable. That delivered the 5% higher. And also demolished thoughts of doing any other TR plan will stick to the same stuff that just got me the 35% gain from Feb (newbie gains and I don’t really expect it to continue, but it is working) . I will do the ramp tests going forward I think. An additional advantage (?) is that intervals.icu then gives me a Vo2max score too