AIEndurance … thoughts?

Eric Schlange (ZwiftInsider) posted a comment on a recent Strava ride that he’s using AIEndurance. Took a quick look at the site and am interested in other’s thoughts on it.

Hard pass. Just using science and AI means nothing. In sports science there’s gobs of contradictory results. There’s also many things confounding factors outside of the data collected. So let’s take a look at the details.

I found this page, then fell off my chair laughing.

They trained AI on a small sample (126 runners, cyclists and triathletes) historical training data and looked at what TID they had that had better performance gains. Then categorized each athlete as better at a certain TID type and simulated all their training like that. All of their conclusions are based on simulations, not actual training according to their plans. Also, their simulations of these athletes continuing to do what they have done almost all predicted the athletes losing performance, despite not changing their training. So their hypothetical simulated gains are all measured compared to a negative baseline! They also are only looking at 8 weeks from now. There were no longer simulations.

All that this is doing is looking at your past training, finding what happened to correlate with your best performance gains and having you do that for the rest of your life.

Many issues here.
0. If you never try a different TID approach, then this system will assume it doesn’t work for you. No mention of what TIDs the “studied” athletes used.

  1. Correlation is not causation. Many things off the bike affect performance improvement. Sleep, nutrition, life stress, etc.
  2. The time frame of 8 weeks is ridiculous. A proper training plan includes several such blocks.
  3. Newb gains or gains returning from time off can be massive and severely skew results.
  4. This is a way, way too small sample size to train AI on. AI needs thousands of data points to train properly.
  5. The only science involved here seems to be using power/pace zones. They’re changing intensity distribution but keeping hours fixed. However, your body can’t handle equal volume in all training zones.

In all, this is just using the buzz of AI to sell a product with quite mediocre science behind it. You’re better off not giving them a cent and just following a generic training plan.


Very thorough reply - thanks. I recently got accepted into the AT Beta so I’m not at all interested in switching. Found it interesting to see another AI-inspired program and fully expect to see more.