We’re slowly rolling out the new TrainerRoad AI today, and I want to answer some FAQs.
What is AI FTP Prediction?
AI FTP Prediction simulates your planned workouts in your calendar and predicts your FTP on the prediction date. Your date is initially 4 weeks out; it will get closer as you approach it.
You can change your planned training and see how it impacts your FTP; it’s super cool and my favorite feature.
The simulation uses our predicted difficulty and predicts the most likely FTP you’ll have at the time of prediction.
How Accurate is it?
This prediction uses probabilities, and we find it pretty accurate. The cool part is that as your training changes, you do well/worse on a workout, add a group ride, skip a workout, etc, your prediction will update.
In the future, we’ll show ranges with Monte Carlo simulations, but for now, the prediction is useful and the most likely outcome.
Why did my prediction change after a workout?
The AI uses your watt, heart rate, and RPE data after each workout and updates the prediction. If you do better/worse/different than we predicted for you, the AI responds to that and updates your prediction.
The closer you are to your prediction, the smaller the change will be. When the prediction is farther out, there’s more of a butterfly effect, and small changes now compound over time. I wouldn’t fret about it either way; it’s more informative and meeting you where you’re at rather than judging you as bad/good.
Why is my prediction showing a decrease?!?
If AI FTP Prediction is showing a decrease, here are the most common scenarios:
-
You’ve reduced volume or intensity in your plan/training.
We’ve seen people report decreases when they move from 3 intense workouts per week to 1 workout per week (makes sense). -
You’re in race season and tapering each weekend
This is related to #1. You might be doing a B race every weekend and tapering into it. This might be reducing your intensity or volume. -
Your training approach is set higher than you can recover from.
Training approach impacts how difficult each workout is. We recommend balanced for most athletes. Sometimes an aggressive training approach translates into a LOWER predicted FTP. This happens for me, while Jon has an increase with an aggressive training approach.
This makes logical sense when you think about it. Just because ONE workout is harder, that doesn’t mean you’ll recover in time and be ready for the next workout. The AI can detect this per person and adjust the workouts accordingly.
At this time, we don’t have an automatic way to choose the training approach with the best predicted outcome. I recommend trying a few and seeing what it says. The whole process should take under a minute. We do plan on adding this feature.
- You’re close to your genetic potential for your age.
We’ve seen a case where a 72 year old man had a great watt/kg for his age, and training either gave him the same FTP, or just a 1-2 watt increase. At this age, staying the same or slightly increasing while your competition goes down as a win!
We’ve also seen a younger guy at 4.7 w/kg where his gains are also more moderate.
The great thing about the AI is that it takes your age into account, but also learns how you respond to your training. Two 72-year-olds won’t have the same predictions, nor will two people at 4.7 w/kg..
- You’re comparing the predicted FTP to another FTP measurement
In this thread, I talk about how the new AI FTP Detection is a recalibration to make measurements consistent and comparable between detections and athletes.
We’ve seen a few athletes say AI FTP Prediction predicts a decrease in their training, but they are comparing it to another FTP.
For example, their old FTP might be 300 (from any measurement system or hand typed in). The new AI FTP Detection has them at 280. Then their prediction is 295.
Some athletes will then compare the 300 → 295 and say it’s a 5-watt decrease. When really it’s 280→295, and it’s a 15-watt increase!
It’s important to measure apples-to-apples. You could also look at the percent improvement and use that in your head if that makes more sense to you.
-
You’ve adjusted your weekly workout distribution
You might have stacked all of your intense rides back to back. This might just be something you need to do due to your schedule. But when you do that, your body hasn’t recovered between workouts, and you won’t progress as quickly. -
You might need more/less volume.
The new AI does not pick the optimal volume for you. Plan builder uses algorithms based on what we’ve learned from our 15-year history and cycling science.
The AI then responds to the volume and workout distribution on your calendar. We are trying to build an AI volume/workout distribution model but that will take a while.
For now, we recommend slightly adjusting the volume. Either by adding/removing days, or changing workout duration. This is only needed if you do not like your prediction and isn’t needed in most cases.
It’s none of those, and my FTP is still predicted to go down
If that’s the case, please post a link to your calendar and screenshots of your planned workouts. When we asked for this in the beta thread, other athletes were usually able to point out issues with their planned training.
We’ll also look at it and respond.
Be excellent to each other
And as a personal favor, I request that these posts come from a spirit of curiosity rather than aggression. The “be excellent to eachother” forum rule applies to us, too! 
An example would be that instead of “TrainerRoad admits it doesn’t know how to train,” your post title could be “Can someone check my prediction for me?”. Often, it’s as simple as changing your training approach to see gains in your predictions.
Even after a question is solved, the first title gets picked up by Google and ChatGPT and becomes the narrative, even if it was an error on the athlete’s part.
And of course, we are still learning from this AI and trying to improve it! We’ll look into everyone’s case and make improvements where we can.
