Ramp test vs AI FTP

[burnett2k]

Hi there! Been using TR for all of this year and have seen amazing gains! Started at 234 and recently got to 296 on a ramp test which I am super thrilled about.

My post is to discuss my experience with AI FTP detection. When I began training, it seemed to work well, and I was seeing pretty good gains. After a while, not too surprisingly I saw my gains starting to slow. This logically makes sense. Each increase in power is harder to achieve and potentially requires more discipline and time on the saddle.

The last month or two I have been doing many more outdoor rides as well, many of them the prescribed workouts. Although I felt that I was getting stronger, my AI FTP detection had really stalled out. I was seeing almost no gains for two months straight.

While taking a week off to visit family, I decided to come home fresh and do a ramp test. My current AI FTP was sitting at 274. I have done ramps in the past and do not believe I am one of these people that are really good at them. I can push pretty hard, but they have definitely kicked my butt in the past. Was hoping that I’d get closer to 285 FTP, but I found I was still able to push it and get to 296. This is around a ~9% difference between ramp test and AI detected FTP. I even felt like I had about 1-2% more in the tank and could’ve pushed it for another 10-20 seconds.

In this situation what should I do? Take the ramp test value or AI? I am guessing my FTP lies somewhere between the two, but I’m not sure where.

I did do a VO2 workout yesterday using the new FTP of 296, and it felt “very hard” to me in terms of RPE, but I was able to complete it.

Has anyone had similar experiences? I am not a huge fan of doing ramp tests, but I am starting to think they are a useful data point that I might want to continue throwing in to see how it compares to the AI recommendation.

[eddie]

Hey @burnett2k,

Good question!

This is what makes Ramp Tests a bit tricky to decipher and employ compared to something like AI FTP Detection.

A Ramp Test is a single, relatively short effort (~15 minutes in your latest case) that can have different outcomes for a pretty large number of reasons, whereas AI FTP Detection is based upon many efforts of all different types and durations.

While your FTP has an effect on the workouts you are prescribed, it’s not the only factor. Your performance during those workouts has a major role in your overall progression. So while you may be able to continue training with an FTP set anywhere between those two figures (274 & 296), there is certainly an optimal place for it to be, which is based on your current abilities and recent training history.

The magic happens when everything is really dialed in and we use your FTP alongside lots of other data to help us pick the right workout for you. If you change your FTP, the outcome could change, and even though the workouts you get might still be doable, they likely aren’t as optimal as they could be with the right FTP figure.

A couple of notes I pulled from your calendar that seem relevant to me are your Picuris +3 workout on July 15th (5x12 at 254-265), and Wright Peak -5 on July 17th (2x30 at 257-260), which left you really fatigued for the rest of the week. This tells me that your FTP is likely closer to the AI estimate than your Ramp Test results.

I typically try to sway our athletes to avoid taking Ramp Tests, as they don’t bring much to the table given the tools we have available now. The system we have in place works really well, and we believe that it’s currently the best option for all athletes.

If you’re the type who really likes looking at data, double-checking things, testing tools, and using tangible data for peace of mind, you could always take on a few 20-minute tests each year, but, similar to Ramp Tests, those have their own shortcomings and challenges as well.

I know that as you progress, fitness gains are harder to come by, and we’re always looking for ways to push that number higher and higher, but just know that you’re on the right track! Finding a different source of data that gives you a higher FTP might be compelling to follow, but that doesn’t mean that it’s going to be the right path to go down.

Let me know your thoughts on this, and if you have any other questions.

Best of luck with your training! We’re always here to chat if you need us!

[Pbase]

I think the key here is to be honest with yourself. Wright Peak -5 is 2x30 at about 95% FTP with a 5 minute rest in between, which is awesome and hard work, congratulations!

If you did that workout at 260 watts, do you feel you could do 298 watts for anywhere near 60 minutes? I know TR doesn’t necessarily use FTP to mean “hour power” like some others think of it, but I personally want it to be somewhere within the realm of possibility that I can do it on that perfect day when the stars align and it all comes together magically.

[burnett2k]

Thanks for the thorough explanation. I would love to see any research whitepapers that TrainerRoad has published on their AI FTP detection, and how the “magic” works and accuracy was assessed. Overall, I think it is an awesome piece of technology, but I’m not satisfied with the black box format of it. As an athlete, I want to know a bit more on the ‘why’ in terms of ‘why’ does it think my FTP hasn’t increased. What are the factors that go into that decision? There’s definitely a psychological factor as well where I’m thinking “how am I not improving when I am adhering to my schedule and getting in a lot of volume”. These topics I know you’ve covered ad nauseum in other podcasts and posts, so we don’t have to get into carbs, recovery, etc.

Do you have any idea what will happen if I do another AI FTP detection when the counter resets? I suppose it will just give a new recommendation which I can accept or reject…

Thanks for looking into my recent rides. The wright peak you mentioned was made harder by me by the fact that we had a heat wave that day, I did the workout outside, and I added 60 minutes to the end of it.

Picuris was hard, but I felt ok for that one.

[burnett2k]

I do think it’s possible, yea. I looked back at that workout and it was on a hot day, and outside. There were a few factors that made it harder than it needed to be and that was my fault. But hey, when the weather is really nice and you have the time, I opt to go outside. So although that was a fatiguing ride, I do think I could potentially ride 280-290 for an hour.

Realistically, my FTP probably does lie somewhere in between the AI and Ramp figures. Obviously, I’d like to believe it is the higher one.

@Pbase Do you do ramp tests at all or do you rely on the AI FTP detection?

[Pbase]

For me, personally, Ramp Tests always end up too high and I end up burning myself out trying to sustain the FTP it produces in the following workouts.

I mostly do the Kolie Moore FTP protocol. It’s been too hot for that lately though, so this summer I’ve been using AI numbers. The FTP Test: Physiology and New Protocols

[John_Koch]

A whitepaper would be very interesting, but I doubt we’ll ever see anything like that. I’d speculate that it’s using a critical power model or something similar, and then updates it when you get a new mean max power record for some duration. As far as I know there’s no good model that uses submaximal efforts to predict FTP or anything maximal, so it’s probably not looking at something like you did 280 watts for 30 minutes after you already did 260 for an hour, so your FTP must be 300. I’d be shocked if it were something like that. Garmin’s model must use HR or HRV in conjunction with power, and you can see how easy it is to get garbage results. If I ride a super technical trail on my MTB, where my HR is really high but average power is low, it will try to cut my FTP in half or something ridiculous like that.

If you’re not already using intervals.icu, sign up for a free account there. It will tell you that your FTP estimate increased because you did xxx watts for y:yy duration, so you know exactly what’s driving it.

[burnett2k]

@Pbase

For me, personally, Ramp Tests always end up too high and I end up burning myself out trying to sustain the FTP it produces in the following workouts.

What were the main indicators that you were burning out? Tired legs? I definitely want to avoid that!

[burnett2k]

Thanks for the recommendation. I do use intervals.icu and it’s estimate is consistent with my ramp. I suppose you mean comparing to other non ramp test rides though?

The hardest part about relying on those eFTP estimates is that you have to go all out and I don’t typically do that as it causes excessive fatigue.

[carytb]

I don’t think there is much downside to understating your FTP as the workouts and your Progression Levels will increase quicker on the basis of the post workout survey. You will also not become too fatigued. On the other overstating your FTP could leave you very fatigued before the system sorts you out.

[AJS914]

@burnett2k

While taking a week off to visit family, I decided to come home fresh and do a ramp test.

snip

This is around a ~9% difference between ramp test and AI detected FTP.

Something like this would be the expected result - ramp test which favors anaerobic capacity while totally fresh.

[John_Koch]

@burnett2k

The hardest part about relying on those eFTP estimates is that you have to go all out and I don’t typically do that as it causes excessive fatigue.

There’s really no way to get a good FTP estimate without going all out.

[mrtopher1980]

@John_Koch

I’d speculate that it’s using a critical power model or something similar, and then updates it when you get a new mean max power record for some duration. As far as I know there’s no good model that uses submaximal efforts to predict FTP or anything maximal,

I don’t know what they do either but I have been slacking the last year due to some family/work stuff and trying to get back now. Week and a half ago I did aiftp expecting it to drop since I was maybe doing 1 ride a week for the previous 2 months since it ran last. But I got a 1% bump. looking at power records in that time I did 20 minutes at about 2/3 of the new FTP. I didn’t even spend 2 minutes at the new suggested FTP. I did 1 threshold 2.0 workout and the rest were endurance 1-2 hours all below a 4.0 level.

If I set my power records for this year to the date of my first aiftp for the year in april to the last one end of july 100% of my records for the year are between the april and a june 2nd aiftp, not one is between june and end of july. In that time every stat went down other than more actual workouts being completed and they were all sub maximal.

Again I don’t know the sauce but mean max power record definitely didn’t occur for me to get a bump.

[burnett2k]

Does that mean you don’t believe in the AI FTP detection then? It does not force you to go all out.

[burnett2k]

Interesting. I was not aware of that. Definitely rare for me to have a full week off the bike.

[JSTootell]

@burnett2k

I do use intervals.icu

I personally wouldn’t use a 6 min effort to estimate FTP.

I’ve seen that bite other people.

[burnett2k]

I agree, that’s just the estimate from the Ramp test. It is another data point I was using to reinforce that the ramp test FTP may be more accurate. But, yes I would be skeptical of that for sure.

Thanks everyone for adding in your 2 cents. Although it seems like I am in denial, I am just playing devil’s advocate and trying to convince myself the Ramp test is accurate, but pretty sure it is not. I will move my FTP back down to avoid burnout. I’ve never failed a workout, so part of me has always thought the AI FTP was a bit low, but that could just be TR doing a good job of giving me the right workout at the right time.

Just frustrated to see my progress stalling out lately.

[Pbase]

Exhaustion. Poor sleep. Heavy legs. Lack of motivation. Dreading the next workout. Grumpiness. Pretty textbook.

[daness40]

I honestly think people get too caught up on an FTP number. That’s why I’ve liked the AI FTP…it’s simple and to the point. Your power threshold is going to vary from day to day by a couple of percentage points anyway. If you’re really curious, just go out and try to do 295w as long as you can. If you make it to like 45-60min that number is probably pretty accurate….or maybe you’ll have an off day and it will be 283w. Or you could just take the AI number and call it good

[mrtopher1980]

@Pbase

Exhaustion. Poor sleep. Heavy legs. Lack of motivation. Dreading the next workout. Grumpiness.

quoting only because this list of complaints mirrors my daughter last night walking into the library with an empty backpack on requesting I take it because of all of the above.. I was going to make a ramp test joke because she insisted we use the ramp to get up, but then weird thought.

The ramp test is actually stairs and the KM test is actually a ramp.