How to Use the New AI FTP Prediction (Without Letting It Drive You Crazy)

After spending some time with the new AI FTP Prediction, reading the forum threads, and doing some experimenting myself, I wanted to share how I think this tool is actually useful — and how it can become a distraction if you let it.

This is not official TrainerRoad guidance, just one user’s take after using at it for a while.

What AI FTP Prediction is

AI FTP Prediction is best thought of as: A 4-week forecast of how your current training setup is predicted to impact your AI FTP metric.

It’s not:

  • A grade

  • A promise

  • A live measure of how fit you are today

It’s a model that says: “If you train like this, here’s where your TR AiFTP trend likely ends up over the next ~4 weeks.”

What it’s actually useful for

Where I’ve found it genuinely helpful is planning and tweaking, not monitoring.

Specifically, it’s good for answering questions like:

  • What happens if I change my Training Approach (Conservative / Balanced / Demanding / Aggressive)?

  • What if I add a weekly group ride?

  • What if I extend my long ride?

  • What if I drop intensity because life is busy?

You can make those changes and see how the projection responds over the next few weeks.

Once you’ve done that initial tuning, the tool has mostly done its job.

How I use it (step by step)

  1. Look at the prediction when you get a new detection (once every 28 days)

  2. Make intentional adjustments if needed:

    • Training Approach

    • How outside rides fit in

    • Duration vs intensity

  3. See if the projected trend roughly matches my intent
    (maintain, small build, or push)

  4. Stop looking at it

After that, I mostly ignore it until the next AI FTP update ~28 days later.

Why watching it day to day is counterproductive

The prediction will move around:

  • Especially early in the cycle

  • Especially if workouts get moved, skipped, or stacked

  • Especially if you’re adding/removing intensity

That doesn’t mean anything is “happening” to your fitness.

In fact, I’ve noticed:

  • Changes tend to happen earlier in the prediction window

  • In the final week or so, the number usually stabilizes and doesn’t move much anyway

Obsessing over daily fluctuations is like checking a weather forecast every hour and getting stressed when rain chances go from 30% to 35%

A note on distraction (and turning it off)

If you find yourself:

  • Reacting emotionally to small changes

  • Questioning every workout

  • Feeling pressure to “fix” the number

That’s a sign the tool is distracting you more than helping. If that’s the case, turn the prediction off and just train. You can do that permanently or toggle it on/off every 28 days each time you get a new detection. The workouts don’t stop working if you don’t look at the forecast. And the system is still doing its thing in the background.

Set your training up in a way that makes sense, then let the plan run. Consistency still matters way more than any prediction line.

Hope this helps someone avoid a little unnecessary stress.

57 Likes

I am struggling to find out why the predicted value keeps decreasing in some situations.

Take today. I am on a “Mountain Bike - Increasing FTP” plan in the Build phase. Today is neither a yellow nor a red day. I do my prescribed 1-hour endurance workout and rate it Easy. Just after finishing my workout, my future workout plan gets recomputed and my predicted FTP (in 22 days) decreases a few watts. WHY? In just a few days, my predicted 5% gain decreased to 4.2% gain. If it keeps doing this, there will be no FTP increase at all come the AIFTP detection day.

8 Likes

This is exactly why I recommend turning off the prediction once you have set your 28 days. Do you need to know the reason there is a 35% chance of rain tomorrow now instead of 30%. The model is a prediction and it will tighten the results with 1) new data and 2) the passage of time. Watching it day to day will drive you nuts. Would you have asked the same questions if the predicted gain was 5.8% instead of 5%.

Also, why do you assume it will go down every day? You are basing that decision on one data point. A lot of the “noise” will smooth out if you look at 28 day progression over the course of a year….and you won’t stress out.

Given that the “secret sauce” isn’t likely to be exposed by the TR team (just like asking why is Coke so much better than Pepsi…is the secret formula and we won’t ever know exactly what that is :wink: ), I don’t think it is worth obsessing about that. Obsess about consistency, recovery, and fueling. And then it absolutely worth spending a little time once a month to fiddle and see what takes your prediction up or down after a recent detection. I just don’t think its worth the mind space and stress to watch it after every workout or every day.

Just my $0.02. Hope that helps!

9 Likes

I am not obsessing, but as a scientist (who knows a bit or two about ML and AI :smiley:) just want to know why :slightly_smiling_face:. What is it that I did/did not do to have the prediction go this way? I had a small decrease or two when I did a slightly different workout (and did some ice skating and downhill skiing), which I completely understand. I moved a threshold workout from Saturday to Sunday (which was a rest day), which did not seem to affect the prediction. I suspected this would give me a yellow day on Monday (I rated the threshold workout Very hard), in which case I would downgrade or skip today’s endurance ride. But it didn’t. I did the ride and got a lower predicted FTP, which seems puzzling. I’m curious what happens after tomorrow’s Sweet Spot :thinking:

1 Like

I’m definitely missing something. Yesterday, after completing the prescribed Sweet Spot workout, the prediction changed upwards to 4.5%. However, today (planned rest day with no activities), the prediction went back to 4.2%. There was absolutely no input from my side between completing the workout and now. Would like to know what is going on. Is the ML model being changed continuously?

1 Like

Awesome post - thanks for spending the time to pull that together :ok_hand:

2 Likes

I understand your take, but mine is different. I was in the beta and I’ve been using it. I understand what the AI and FTP Prediction are supposed to be doing. But I do look regularly at my predicted FTP now. Why? It’s motivational. “If I stay consistent and hit all these planned workouts, I’ll achieve this.” Nothing I didn’t already know in the past, but having that number to look at is nice, easy validation.

4 Likes

I think part of it is being influenced by work in the past dropping off or being weighted less highly. I noticed my prediction dropping during a rest week now going back up slightly once I’m back in a non rest week.

I think it would be fair to expect TR to explain how the past is influencing the prediction. Is it using the previous FTP value as an anchor, is it only looking at work between the two points? Or is it doing something else.

Make it a little less black box.

2 Likes

That does sound like a strange shift. That’s a case to see if @SeanHurley @Caro.Gomez-Villafane @Nate_Pearson can explain. Or email support?

Totally see the value from your perspective. If it is motivational and works, keep doing it. I’ve just seen a lot of people posting about relatively minor fluctuations and are really worked up about it. My post was to help users who are getting distracted, not seeing the forest for the trees.

I’m a HUGE fan of this new release and was also a beta user. I love the flexibility it offers. There is definitely a new UI to get used to, but overall I think it is a major step forward.

Final thought: I hope people try the system for a full month and let it learn and improve the training. Seems like some people are trashing it after one workout or one adaptation.

7 Likes

Thank you for this! I’ve been using TR for maybe 10 years and the change is stressing me out mostly because I’m old and I liked the way it always worked - I had a system :slight_smile: Can I still get auto FTP detection like we’ve had for some time if I turn off the new AI feature? I’m not sure how to turn the new AI off just looking in settings.

2 Likes

Yes if you turn off the prediction you will still get the detection every 28 days

Go to your calendar, find your future prediction, click on it, find the edit/pen in the lower left corner, and deselect AiFTP prediction. Then you won’t have to deal with this!

you can do it on the app too

1 Like

Thank you! Looking in app now for future prediction. Maybe I should get on a computer :slight_smile: just back in town so need to get on the trainer today!

Thanks so much for pointing this out, great to be able to do away with this.

Ok, now it’s getting ridiculous. Another successfully completed SS workout, expected to be hard, rated to be hard. Feeling fine. After completing the workout, my prediction just went from 4.2% to 2.6%.

One possible theory of what is going on: My AIFTP was calculated incorrectly (too high) and, even though I’m following the plan and nailing the workouts, TR AI realizes I’m not as fit as it expected and downgrades the expectations based on my heart rate during the scheduled workouts. Which, in turn, means the workouts are not corresponding to the expected training stimulus (e.g., the planned sweet spot veers in practice more towards threshold).

There are, of course, other possibilities: I may be getting ill soon, the prediction is broken, the scheduled workouts do not provide enough training stimulus for any increase to occur…

To sum up - it does not inspire much confidence in TR software, if in just 9 days after AI FTP detection, and following their plan, the predicted gains go from 5% to just 2.6% :confused:

4 Likes

The system is just fine tuning! I’m on the other end in that my prediction hasn’t not changed at all. Can’t explain why there are variances. In the end it is a prediction. You never had one before. You just trained and periodically tested or used Ai detection. Just turn it off and regain your serenity.

Also, would you be as upset with the system if it started at 2.6% increase and was now showing 5.2% increase.

3 Likes

No, I would not :grinning_face: . I would also like to point out I’m not upset :slightly_smiling_face: - just wondering about the quality and usefulness of both AI FTP detection and these predictions. I’m just a curious person who really likes to know how things work :smiling_face_with_sunglasses:

3 Likes

Exactly.

It would be super cool if there was a bit more guidance on what things are contributing to it going down and which things are contributing to it going up. I made a change to mine and added a hour of training and the prediction went from 323 → 333. Then I tweaked it some more and it went to 329 and I was unable to get the 333 back…

3 Likes

Can’t help but feel that your are rather brushing past the part where your FTP increases 2.6% in 28 days :thinking:

4 Likes