AI FTP detection, FTP progression with AT, in theory

I just accepted my first AI FTP detection and I got a 3.4% increase following SSB 1 plus one week of SPB (as that is what plan builder gave me). The ramp test was scheduled on the first day of SSB 2.
I am on a mid volume plan.
I am using Adaptive Training (AT).

My question is, using Adaptive Training, and assuming a “normal” progression (e.g. you don’t fail a lot of workouts and you don’t mark all workouts as easy) with your plan and accepting adaptations, is there an “average” expected increase in FTP (via AI FTP detection) through the base phase and the build phase respectively?

My situation is that I have been disappointed in my rather stagnant FTP over the past 2 seasons. Since getting on AT (really only at the start of SSB 1 about 7 weeks ago) I have been able to complete all workouts as prescribed which is great. Prior to AT I would often not be able to complete the workouts as prescribed and would need to get out of Erg mode to take a rest interval (or extend the rest interval …) in order to get to the end of the work out.
I am a masters aged athlete at almost 47 with only about 4 years of endurance training.

I am hoping that AT helps me un-stagnate my FTP this season. I am optimistic it will in that I am no longer “failing” workouts. This season I also moved up to mid volume from low volume in order to seeif that was a limiting factor and it has been manageable so far. If I can get another 3.4% bump I will just get past my previous all time high FTP. As mentioned I am just starting SSB 2 so I am optimistic through this stage and then a build to follow, I can get the additional 3.4% but a bit more hopefully (which is why I am curious on what the “normal” expectation is through the phases, with both AT and AI FTP detection).

I will use my own experience as an n=1 experiment, however, I am curious TR has a “normal” or expected range of FTP increases (via AI FTP detection) through the use of AT and the phases of a season.

  • TR is actively working on what I am guessing may be called “AI FTP Prediction”. It is a future state forecast of where your FTP is expected to be.

  • We know very little about it at this time other than some hints presented by Nate & Amber on the podcast. Presumably, that will cover your overall question, and likely be more “tailored” to each individual in similar ways that the AI FTP Detection is at this time.

I have not reviewed these results, but here are search results that may give you some answers within the forum:

1 Like

Of course there is an average. The problem is the average is only one point in the distribution. It is almost meaningless.

The more meaningful answer is that it depends on things like age, experience, FTP history, etc. Example: After @ambermalika retired and joined TR her FTP dropped quite a bit. She then did one block of training and did a live ramp test with the other podcast hosts. I don’t remember how much her FTP increased, but I do remember them joking that she “broke the ramp test” because she went so long. Point being, Amber completing that block of training and me (or you) completing that block of training are two very different things. As such, the average for her would have been meaningless.

You just got a 3.4% increase. Seems like you should just keep going with AT.

2 Likes

^^^ this.

And an interesting thread from a user:

Interesting read, but not an official position of TR.

@vanbc Congratulations on the jump! I wouldn’t focus on averages since that’s going vary significantly across individuals based on inherent ability, experience with structured training, etc. For example, someone that’s well-trained over the last 10 years probably has less room for growth without making significant changes than someone new to structured training. As long as your improving and enjoying the training, then you should keep up with the work your doing. Just my thoughts.

1 Like

I have been thinking about this…more from the feedback of users that keep posting. Not enough data to assess but it feels like everyone who follows a plan and is doing the workouts is getting a bump in ftp. Self fulfilling? If you get sick or cant do the work then FTP declines. Not sure that I have seen someone indicate they did all the work and didnt get a bump.
Estimation that a certain amount of work will produce an improvement in FTP? There has to be a few factors involved in their data set…age would be one I expect…it does sound like it is more past history from their data set to what happens in the future with some subset of variables. They are just working on how to fit the data…but given all this it would lend its approach to the prediction of your FTP.

The interesting part of this is are they using workouts as a data set…or have they translated this into some function of power to have work produced. This is the interesting part…to me anyways. Need a work produced output to do the outside ride part…in my view.

I dont expect to get he detail but would love to play with an FTP prediction model. Say Jan 1 FTP is xxx. Based on workouts entered into calendar we get output of FTP at various points in time. Would then get to play with how to change what we do…how does it help or hurt the outcome.etc.

It’s honestly really hard to tell from AT if you are getting fitter because of the PL adjustments. You can get almost the exact same workouts with the right combination of PL and FTP setting.

My preferred way of tracking fitness is through MMP curves at durations relevant to what I care about. You should look at it for both raw power and W/KG.

It’s not a good idea to use the FTPD as a main measure of fitness. I didn’t like the ramp for that purpose either, but even that yielded more insight than the FTPD in that category.

To your question on AT - I’ve been rewriting my record books on it and I’ve been cycling for 20+ years and am in my mid-40s. AT isn’t solely responsible for that, but it has certainly helped a lot.

I would be willing to bet they have several hundred features they have at their disposal to put into the model. Not all of them are necessarily used, but the amount of (worthwhile) feature generation you could create with their dataset is insane. I just brainstormed for like 30 minutes (because I am a data scientist and a nerd) and came up with over 200. It is really easy to come up with them when you bring time into the equation. E.g., not only what my volume is, but is it more or less than the month before, and what is the rate of change.

1 Like