AI updates and long period off the bike FTP detection issues

TrainerRoad Software
1

Background:

I have been cycling for more than 30 years off and on. I have a good feel for what a good progression rate should feel like for all types of workouts and the fine line between doing too much where you set yourself up for failure and not doing enough to maximize potential adaptations.

Current Situations:

I have been off the bike for about 9 months due to some general life circumstances. During this time I had next to no activity. I recently got back on the bike, accepting the suggested FTP. All was well and the suggested initial FTP was maybe just a bit lower than what my natural baseline is after an extended period off the bike. Once again, I have many data points over the years as to what my body goes back to for it’s default and have a very good idea what the baseline FTP should be. The first workout back was a simple 1 hour moderate endurance ride, followed by the next day doing a 9.1 level 1 hour sweet spot workout. The sweet spot workout felt like it was somewhere between moderate and hard, and I ranked it as moderate. From these two ride data points, it suggested a 40%+ increase in FTP. This is absolutely unreasonable and would only set me up to failure if I tried to continue at this number. I went ahead and accepted the suggestion, and not only was the FTP increase over-estimated but it set my sweet spot level to 9.1. Based on not only experience but a ramp test to confirm my real number should have been closer to a 12% adjustment upwards from the initial underestimated value.

Moving Forward:

I contact support with all of this information and have been given generic answers that the AI is basing it’s data based on the one sweet spot workout that was rated as moderate with the initial FTP it set. This is coming from somebody I’m sure doesn’t have a background in data science, mathematics and/or other engineering backgrounds. The “AI”, could have looked at my history and observed other period where I was off the bike for extended period of time and seen the natural progression level that occurred with ramp test and the prior AI models, which both were far more reasonable in setting me up for success, but it doesn’t appear to use that data at all.

I’m going to have to consider alternative training options outside of TrainerRoad at this point if I can’t get the software to recognize realistic updates/suggestions.

2

It sounds like that issue with getting level 9+ workouts after a big FTP increase is definitely a bug. :bug:

Can I dig into this a bit more and let you know what I find with your account?

3

Yes please dig into this more.

I question not only the level not resetting as a bug but the overall algorithm not using enough data to suggest such a big jump. The data that it’s using is most likely based on optimal conditions without a long break off the bike. In my opinion the algorithm could be improved by checking gaps in training history and not providing an FTP update without checking a person’s full history to see how they progress after time off. The algorithm should also have some basic sanity checks and know that a drastic jump, say more than 20% after a single event doesn’t make since. If the sanity check detects an increase greater than some defined upper limit ignore the results, maybe log conditions for dev team to review and see what updates are required, if any.

In the past, when coming back from a long break, either using ramp test or the prior FTP detection model I was able to gain fitness at the correct level as to not carry over so much load that it sets me up to abandon workouts. If my FTP changed and the first ride afterwards shows that my muscle tension/ in the legs and heart rate are lower than expected I would simply adjust the workout percentage up to be in the correct rang for the targeted workout. If the new AI algorithm will not be updated, would you consider respecting FTP test and setting the levels back to say 3-5? This way if a user takes a test or manually enters an FTP they would override the AI and then allow the progression levels to move up over time before suggesting a new FTP.

Note, I’m not speaking about any of this from a general population’s point of view. I have a comp science degree so can knowledgeably speak to what I’ve observed in the software I’m using.