AI FTP Detection Update

Think of it more like a constant progression where we are either moving your power up and/or out depending on your goals and place in your plan.

Then there are occasional small jumps after you recover and we detect an increase, but not such a big jump that it crushes you.

I like the weight training analogy. You squat 135 at 3x10. You then put 145 on the bar and you might get 1x10, 1x8, and 1x6. Then you work it back up to 3x10 but are at 145 lbs this time.


Family Feud Good Answer GIFs | Tenor

1 Like

TR and its athletes need better metrics. Too many of us are chasing power. Even though I am aware that FTP is not the only performance dimension that matters, it is really hard to quantify progress in the other dimensions. E. g. last year I did a power PR (117 % FTP for almost 7 minutes and an average heart rate of 159 bpm — and I had energy to spare). But how do I quantify that? Should I repeat the climb to assess certain aspects of my power?

I can agree with that, even after three years of consistent training my threshold power is comically low compared to my anaerobic power, that’s why I would like the added metric Nate proposed. I know it’s not for everyone but the whole three day test does help to paint a good picture of where your fitness really is, e.g. day 1 sprint/anaerobic test, day 2 5 minute test, day 3 20-minute test.

1 Like

I’m probably the wrong person to game this with because, honestly, I’ve been training for several years and I’ve NEVER had an FTP that I thought I could hold for 60 minutes (whether it’s from an 8 or 20 minute test or a Ramp). I just see FTP as a number to base workouts off. I was more speaking to the constant debate I read on the forums.

I do very much appreciate that TR listens and tries to solve these issues though!


Nate probably has the numbers on this, but there are outliers and they’re probably the most vocal about setting FTP correctly. If you’re not an outlier or if you fit well in that bell curve, you’re probably not super concerned about the number that AI FTP or the ramp test gives you because it’ll probably be just right.

Regarding the TR team, I am VERY thankful they take these issues seriously and actively communicate with the user base!


Alternative Tours GIFs - Get the best GIF on GIPHY

Family Feud GIFs - Get the best GIF on GIPHY

Yes, but first we need outside PLs, red/green, and more engineering work to do that but that’s where we are headed.

In the meantime, if a forum user is sure they can do a higher PL then use the workout alternative and try a stretch or breakthrough workout.

We also know that workouts that are adapted are half as likely to be failed than workouts that are not adapted.

We also know that as athletes do higher PLs than what is prescribed, their failure rates go up.


Yes, this is the entire point of TrainerRoad; to make cyclists faster.

AI FTP Detection takes unstructured outside rides into account. PLs do not take it into account yet, but we are working on it.
Sweating GIFs - Get the best GIF on GIPHY

We do take failure reasons into account, although some of them are the same/similar as others.


AI FTP Detection takes unstructured outside rides into account. PLs do not take it into account yet, but we are working on it.

Perchance are you guys working on AI FTP for anaerobic outliers? :pray: :pray:

1 Like

Thanks, I forwarded this to the team to look into.

1 Like

This is true.


I’m liking tFTP right now. I know it’s similar…but honestly it’s mostly the forum people who want to see it change.

Zwift educated tons and tons of people on what FTP is, even if they aren’t using it as a true 60-minute power. Us changing the name too far away from FTP would cost millions and millions of dollars to educate people about the change, and the benefit would most likely just be so forum people can sleep better at night. :wink:

Like why not spend that money on some more data engineers and data scientists!! :smiley:

In fact, we can probably just have a support article calling it “Training FTP” that can be linked in the “?” by FTP and discuss the who debate about threshold and FTP and how long you can hold it. Then keep having the words “FTP” everywhere else.


You make a very good point!!

We have this internally and we call it “Fitness Score”. It’s a combo of your FTP + PLs + Weight.

Then you get an arbitrary number that you can see go up/down. It has very fine grain sensitivity and can be used as a metric to know your performance is improving and you’re getting faster!

IE for me, I had a high Fitness Score of 1,226. And before Cape Epic I was around 1,149, or about 6% off of peak fitness.

We need outside PLs first, but I think this could be a pretty cool way to show small fitness increases and motivate people to stay consistent.


How far of an outlier are we talking? The PLs offer a pretty big range…


Whenever this gets released (can’t wait), will it be able to show us our historical scores? Would love to see it over time… as least as long as PLs have been out.

1 Like

Someone who can do a level 7 anaerobic workout with a 6/10 RPE but struggles to do a level 3 threshold workout at a 9/10 RPE.

1 Like

your PLs cannot contain me! lol



Regarding the name change, though, I have an idea. If you want to improve upon FTP, then you should name it what it is: estimated lactate threshold. This disambiguates it from wrong interpretations of what FTP supposedly means (e. g. 60-minute max power), it is clearer and most importantly, it is scientifically accurate.

And I assume it is internal for good reason: it is not clear for whom this metric works well and for whom it doesn’t.

The difficult thing is that I can easily see several layers of metrics, e. g. metrics to compare seasons or so. E. g. it’d be nice to have an end-of-season report where e. g. I’d see stuff like total hours, total hours in zones, average hours per week, etc.

For example, the metric I’d probably want to know in addition to FTP is endurance/repeatability. I know that this isn’t very easy to quantify or score, but endurance is essential and typically forgotten by beginners who tend to focus only on FTP. But you could first create an absolute weighted score that measures how many intervals and how many minutes you manage e. g. at threshold or VO2max. Then you rank all athletes. You don’t actually expose this absolute score to the athlete, though, but you compare athletes relative to one another. E. g. you could compare athletes of the same gender with either the same FTP or W/kg with one another; perhaps you throw in age as well.

Once you have these metrics, you could classify and rank athletes. Not that you’d want to expose that I am in the bottom 20,000s of sprints or so, but you could categorize people and build a profile. You could identify relative strengths and weaknesses, and then suggest e. g. two training plan alternatives, one to improve your strengths, the other to address your weaknesses. I am sure you have thought of this as something you’d like to have in 10 years or so.

Thanks for sharing, that’s pretty neat. Did you feel the score was accurate and informative? (I remember that you had a training hiatus before Cape Epic and didn’t feel you were anywhere near the top of your game. Just subjectively from how you spoke of your fitness, I would have expected a larger drop in fitness score.)

Yeah, that’s the big one for sure and really needed to get an accurate picture of the fitness of an athlete. Especially when it comes to endurance rides, TR has me way too low, because I don’t do long trainer rides in the summer (I don’t have more than 2 hours in the morning, 2 hours is stretching it, and on the weekend, I wanna be outdoors).


I know you said in previous that you don’t use TR workouts currently, but if you were to feed the Machine information through the surveys it would customize a plan around this. So your PL level would be higher for anaerobic workouts while your threshold workouts would be lower relative to PL.

It takes sometime to get acquainted but will build itself to best fit you.

1 Like

@Nate_Pearson I would find an AI generated hour power approximation extremely useful. I think that is a great idea.

1 Like

Why don’t you call it TTP (Training Threshold Power)? That way it. Is divorced from FTP entirely