I’ve been on TR or a year now, using AI FTP the entire time.
I’ve consistently got increases of ~1% a month ending up at 241w early October.
A few days later I did 252w for 48 minutes. A few days after that I did 262w for 46 minutes and a few days later 278w for 20 mins. Albeit, all outside but all several hours into a ride.
All of which seemed to point at an FTP more like 250-260, so I manually set my FTP to 250 but AI FTP continued the course on ~1% increases, only just reaching 253w today, three months later.
Which got me wondering, is AI FTP capable of finding breakthrough increases? Or can it only manage gradual, subtle changes? I’d love to hear other people’s experiences.
This isn’t a complaint as such, progressively increasing PLs continue to be a good source of challenge on the trainer. I’ll likely soon do my first ramp test in a year out of curiosity too.
I have used TR for a month on a calculated FTP by AI. After just a week holiday, TR recalculated my FTP down 10 watts…
I felt my FTP from before the holiday was a little on the low side, so was thinking about adjusting it manually, or take a ramp test, but the guys at TR recommended me to follow AI FTP detection.
If I would accept the new AI FTP, I think my training would become very little effective.
I really cant see that the AI FTP would be correct for me at this moment, so i will stay on my FTP as before, and do a ramp test after my coming rest week…
I think the only reason I saw breakthroughs with a ramp test was down to two things:
Longer between testing as most people avoid testing if they can. So if you use AI FTP every 28 days and go up 5w each time it looks less like a breakthrough than ramp every 90 days with a 15w jump.
Consistency, I can have a shitty test and gain very little then the next one I’m on fire and it looks like I’ve improved massively but in actual fact it was a poor test previously.
Obviously, this is just my experience and people who test better might feel otherwise.
AIFTP does take into account unstructured outside rides. My N=1 is that it doesn’t seem to weight them that highly though. Early last summer following a crash and a break from training AIFTP dropped me to 314W from 331W. The crash injuries meant I had to ease my way back into training gradually, so since the race season was already a write off I decided to take all pressure off and just not have a plan, go largely by feel and really enjoy my riding. Built up to a really good volume and consistency by late summer, was doing hard group rides and climbs and felt like my numbers were back to at least where they were pre-crash, probably a bit higher. Backed up by doing a 40 minute climb at 338W fairly comfortably (I.e. Felt like I could have carried on at that power for 50-60 minutes if the climb was longer) and a number of shorter 15-20 minute climbs at >350W. In that 3 month period AIFTP only bumped me from 314 to 316 to 319 when I’m very confident I would have tested at around 340W.
Not overly fussed as if I do need to do a long threshold effort in training or racing I will always do it by feel not numbers anyway. So AIFTP is very much a number for setting training intensity, and it continues to do that job well. Probably because if I’m doing them indoors in erg mode RPE at any given power is always a little higher so slightly lower targets are still hard. And if doing outdoor workouts there is a power range and I can just aim for the top of that range if feeling good.
The first two gains are as I was coming back from injury, so it makes sense they would be slightly larger than typical training. When I consistently follow and complete TR workouts , ~2-2.5% is what I typically see for AI FTP increases every 28 days. I generally choose Productive or Stretch workouts, only ever choosing a Breakthrough workout if I know my progression level is low for that workout or the workout looks particularly easy.
Will AI FTP Detection consider unstructured outside rides as well?
Yes. AI FTP Detection takes into account all rides with power data, whether they’re indoors or outside, structured or unstructured. This can include group rides, races, and even rides on other indoor training apps synced to TrainerRoad.
After the prerequisites are satisfied:
What AI FTP Detection Needs to Detect Your FTP
In order for AI FTP Detection to have enough data to accurately detect your FTP, you need a minimum of 10 completed TrainerRoad indoor workouts. If you haven’t completed enough workouts, don’t stress—you’ll see how many more you need when you click the button to use the feature. Once you’ve completed these 10 workouts, AI FTP Detection will analyze and incorporate all indoor and outdoor rides with power data, whether or not they’re TrainerRoad workouts.
Funny how much attention we all pay to FTP numbers when with AI FTP we really don’t know what goes into that number.
My feeling is that AI FTP is throttled somewhat, i.e. it tends to add or subtract smaller and more conservative percentages of FTP. I wonder what one would do to maximize AI FTP jump to test this? Like, if I set my FTP 50 watts low, then did all workouts at 120% of suggested power, how much would AI FTP correct?
If someone is 100% brand new to training with power, how does TR suggest getting an initial FTP? Ramp test? This is a situation where I could imagine someone having an FTP that’s WAY off, so AI FTP might need to make a big correction.
I came to TR on September 18th 2023 - so pretty recent. I am not a new cyclist, but am/was completely new to training with power and structured training. I was also completely new to an ERG trainer.
Initially the onboarding set my FTP at 136 at 55kg bodyweight which, looking at the stats, I knew was a little low. I am not a pointy-end XC racer by any means, but I am a reasonable climber in my social groups, both female only and mixed. So I set up the new Kickr and my first ever ride on it was a ramp test with no fan, which set my FTP at 158. I figured that was probably a tad low too, considering I was too hot and had no idea how to ‘ride’ an erg trainer…However I was pretty burnt out from lots of unstructured training, and settling into what kind of structure I could cope with/plan on top of my weekly rides so was happy to settle with that rather than redo the ramp. Since then, its been bumping it up bit by bit each month and I’m now at 181 at 55 kg = 3.29w/kg (actually most of the time I weigh less than that - that’s the upper level when I’m hormonal and bloated), but I still think it’s probably a smidge under. I haven’t failed a single workout, even when I have done a massive outdoor ride before, or I am carrying a lot of fatigue from a 700+ TSS week. I hate ramp tests though and would rather have AIFTP than ever have to go through one of those ever again. It probably isn’t a million miles off, and it’s more motivating to know I should be able to complete a workout than be overestimated and dreading sweetspot.
So I don’t know how AIFTP works, but I do think it is moderate in the increases. I don’t think it would ever actually bump someone up a huge amount. Onboarding said I could just miss the initial ramp test and rely on AIFTP, but I had an A race to prep for and couldn’t risk being that much undertrained. I was surprised it bumped it yesterday, I am actually now finally tapering for my race this weekend.
There may be a retention aspect to gradual FTP increases too. A slightly underestimated FTP where you’re crushing workouts and soaring through progression levels is more likely to keep a user engaged and consistent than an overestimated FTP where you’re failing workouts and getting burned out.
FWIW I seem to get a higher FTP from ramp testing than from AI FTP. But in the past I’ve had ramp test FTPs that I think were overestimated; at least, RPE of workouts based on these FTPs was VERY high and I felt near failing a few workouts. With AI FTP this has never been the case, and I’ve actually built up to an AI FTP number higher than my previous ramp tests yielded.
At a minimum I think AI FTP has less condition-dependent “noise” than a monthly ramp test, i.e. solves the issue that you might happen to be feeling great/lousy on ramp test day and over/under-perform relative to your usual ability.