How it started, how it's going aka predictions vs reality

Training
227

So wheels started coming off after a hardgroup ride, race and high z2 / tempo solo ride inside the last month.

Started last month with 293, the predicted FTP was 307 and ended up with 287. Mostly because of the 3 unsrructred rides mentioned above.

Weather good so cancelled my account. The system works really well if training indoors and not tempted by fun stuff :joy:.

Had a tough cat 2 race over 100km, didnt go especially well due to feeling burnt out week before race and no intensity.

So I am going back old school and will race club league thur and open races on sunday, everything else easy on feel for a total of 8ish hours weekly. No AI, TSS, CTL, TIZ, FTP, analysis etc.

228

I’ll be surprised if I can stick to the plan for the next 4 weeks, but here we are for another update.

  • Detected AI FTP April 5: 244
  • Prediction then for next update May 3: 257
  • AI FTP Detection yesterday, May 10: 248
  • Prediction for 28 days on Jun 7: 268

I traveled for about a week at the beginning of April right after the detection so didn’t really get any good workouts in until a week into the next block. Missed more scheduled workouts due to life than I hit after that, swapped some for outdoor rides and some group rides, and then updated my training plan to accommodate a charity ride I’m going to do in August. Basically, I didn’t stick to the plan and the drop from prediction to detection shows that.

I’ll be making more changes to the plan to ride more outside and increase volume as daylight and work schedule allows. So right now that prediction of 268 is based on indoor workouts that I know I won’t be doing. Priorities for me are going to be some more group rides rather than workouts … just don’t know how those are going to fit into the calendar at the moment. So I’m not really expecting that prediction to hold.

eddie 229

Hey y’all! :waving_hand:

I just wanted to follow up on this one and a few other threads to let you know that the team just tracked down and fixed a bug that was causing AI FTP Prediction volatility. In some cases, an upcoming workout in your plan could swap to the next best workout (instead of the first best). If your workouts and AI FTP shifted without anything changing on your calendar, you were affected by this bug. The fix has already taken effect so you don’t need to do anything on your end for your workouts to be set correctly. :bug:

As always, stay in touch and let me know if you run into issues moving forward, but we think this is going to play a major role in eliminating FTP Prediction volatility moving forward. :partying_face:

11 Likes
230
  • Original prediction 4/13: 247 —> 261
  • Result 5/11: 247 —> 244

Not exactly sure what happened. The biggest shifts occurred when I rated a sweet spot workout “very hard” when TR was split between “hard” and “very hard” (went from 259 —> 251); and then again when I did the same for a VO2 workout (246 —> 243). I basically watched my predicted FTP gradually decrease for a month.

My next prediction is 253.

231

Two weeks into the block, I had a bad first week missing the Threshold workout, and swapped in my first group ride of the year for Threshold in the second week.

So my ftp is predicted to fall again. I’ve tried a dozen different variations over the last two days to get back to gains bye no dice.

Prediction started at +10W, now -5W, no outdoor workouts.

I’ve used TR recommended options since January, so I’m just going with what I want to do (1x hard, 1x endurance, 1x hard group ride) and turning off Prediction.

232

I am in week 3 now for my first TR AI FTP expierence. Prediction is a little optimistic 249w->266w, ain’t gonna happen in 28 days ;-). But I don’t mind. I try to follow the plan as good as possible. I think issue could be is that he uses old history maybe? I train now on TT bike as I want too improve my power on that bike.
One problem is that I am on a short holiday for 5 days this week and next week I have a recovery week in TR plan. So not sure how I can ‘override’ my 3 endurance rides from next week with some proper workouts like treshold/sweetspot. My plan did not adapt when I put 5 days off in my calendar? Should I use Trainnow next week?

233

Fourth round of random number generation, I mean AIFTP for me…

14th Apr prediction, 392 W. Good training and compliance with AIFTP increasing to 395 w and being fairly steady until last Thursday when I got ill and wasn’t able to train for 4 days. Removing workouts dropped it to 383 watts but rescheduling in some threshold for yesterday (to fit with shifts) bumped it back 390 w until I actually did the workout rates it HARD (as expected) and it then dropped the Aiftp for today to 384 w. Wondered if it might bounce over night but so pressed the detect my ftp button this morning

12th May AIFTP 385 watts with a new prediction for 4 weeks of 408 watts ( I assume I have to sell my soul to the devil and start hitting the EPO to reach that). My prediction is it’ll plateau…

At least I’ve progressed this month but so little faith in any of this now…

234

You can build the week in the same template as your calendar and populate with an AI workout from the desired zone.

235

I am really curious how this will play out if outdoor rides are now “fixed” in the model.

I finished a block and started another FTP gain block. I was hoping that predicted FTP will be a very slight but realistic increase (say 1%) and would sustain as long as I complete all the workouts with predicted ratings. But predicted js a +4.2% bump after 22 workouts in 28 days. So it still seems overly optimistic. That’s probably possible for someone who is coming back to cycling after a long absence.

236

Another cycle for me. Prediction for 12th may was 242 and the actual detection was 234. Followed my low volume plan. Workouts are spot on! Next prediction is 245ish. I am confident that 234w is my current ftp. Wish I had more time to train, but from now on daddy first athlete second. Can’t wait for kiddo to start kinder garden after 1 year so I can bump up my volume. I am satisfied. Hitting new dad power records. I’m 20-25% off where I was before fatherhood /thats not even important for me/. Plan to sign up for my first 25k trail run in September if I got my ITB in shape. Brutal route in the local mountains.

237

When this bug is completely fixed, your predictions will be accurate, just like they are with indoor plans right now.

It will help a lot when athletes are building plans with outside workouts. :+1:

238

AIFTP Update - 213 May 19, 2026 (Down from 249, very bad month)
AIFTP Prediction - 216 June 14, 2026
I’m going to let TR do it’s thing on this one but I don’t think I’ve lost that much due to missing some workouts, so the next prediction may be interesting.

239

Here’s mine.
Preparing for a 150km cyclo 30 may with a B goal at 6 april (125km cyclo);

Started at FTP 226 FTP in january after a lacklustre winter (almost no cycling december), was down from ~260/270 in september.

  • First prediction I dont even remember, I think it was 240? Ended at 236 FTP
  • Second prediction was 253 FTP. But I had a messy february and missed some workouts. Ended at 238 FTP
  • Then I had a really lacklustre march and again didn’t make the prediction of 250 FTP and ended at 234 FTP. All is totally logical so far as I simply skipped too many workouts or replaced them with simpler rides
  • Remainder of March and April was very good, including the 120km B goal. FTP up to 254. Which is nearly exactly what it predicted at the beginning. I replaced maybe 2 hard rides with solo rides that’s it.
  • Next prediction was 267 FTP for today. I skipped maybe a few rides, i think 2 endurance and 1 treshold ride. And also added two solo rides in that period. I ended up with 266 FTP which again is nearly right on the money.

What I learned from the new AI FTP:
-Take balanced, as it seems to lead to the most accurate predictions for me
-Do not skip or replace too many rides manually with other stuff.
-No compliance gets punished hard and rightfully so. It also seems accurate because if I ignored it and set different FTP I’m pretty sure I would have ended up with too difficult workouts. Now I feel ready for them.

Rate RPE properly (!!). A treshold ride you barely finish is not “moderate” and an easy endurance right should not be “hard”.

240

5/17: detected 241- a couple workouts that were messed up due to the afore mentioned bug, several races, a couple group rides subbed in to build the group skills needed for races

new prediction: 265

241

Good point about the algorithm. If it’s overestimating fatigue from other activities, it will drop your levels. Keeping the data minimal might stop it from over-correcting.

242

Turned it back on to do detection and planning, sure enough I lost 8W. 250W to 242W, now if I execute I’m predicted 256W.

243

Not sure how I am just seeing this thread, but this is great.

My predicted a week ago looked decent. Still lower than Garmin and Intervals’ estimation. Then the detected yesterday only went up two watts.

Now my new predicted is a staggering 32 watt increase. Which I don’t think is entirely too out of the question and where I hope to be by Breck Epic.

Reading @nebur86 's thoughts on it: I only follow the plan by doing the two hard workouts. I chose low volume masters as the recommendation of the TR staff who say, choose low volume so you can add endurance easily etc.

I move my hard workouts around based on weather. I’m not doing threshold, ss, vo2 on the trainer. TR has me doing about fours a week, I do about 10 hours. I should adjust the available hours to line up better since I’m commuting now to work and riding as much as I can. An example of me just ‘winging it’ is Saturday. I had a 45 min threshold workout. I did it, ended the workout on my garmin and then did 2 hours of z2 after for a total 3 hour ride.

What I haven’t done this year is to just go all out on my 20-30 min climb just to see what the numbers look like and how close they align. I had a 5 hour gravel race last week and my IF was .91. I did ride well, the race was about 5K’ in lower elevation, I took it pretty conservative at the start, and did finish third. Just based off on feel, I’d say my FTP is probably 20 watts higher than what it currently is set to.

Anyway this is a great thread and I should look at it more. Good information and insight.

edit: Even just doing the two hard workouts and rest mostly z2, my durability and time I can just ride at SS racing is great. So, regardless, I feel the TR stuff is working.

244

Here is my data and my comments after 4 months of training.

image
image555×196 4.56 KB

Average predicted gain per month is +20w or ~8%, achieved gain (ignoring April) is +12w or ~5%

Note; April I didn’t see a gain due to 2 missed sessions / travel / family commitments. I’m fine with this.

For context, I am 40 years old, 173cm, 69kg (with ~3kg to lose) on a Cross Country Marathon plan set to Balanced as flows

image
image1027×294 15.7 KB

I do the three intervals sessions inside in ERG with around a 90% hit rate of completing them at the prescribed intensity and predicted difficulty.

The other 10% I have either marked as very hard, or I turned down the last interval intensity by 5% or so, to try and hit a Hard RPE, rather than very hard to avoid cooking the legs.

Ive missed around 4 of the interval workouts over this time period, but I have generally swapped in racing.

I have not been good at doing the Sunday endurance session as there has been Cyclocross / night racing (its winter here in NZ) and adding in a long Sunday ride is too much.

The workout difficulty has been spot on, I am really happy with the progress and the workout selection. I also find the predicted gain super motivating.

For context, the previous two years on TR I gained +20w over a 10 month training season, I am already on a +40w gain after 4.5 months and now at 4.1w/kg. Thanks to the new TR AI!

An achievable gain for me seems to be around +12w or ~5% per month. The AI seems to think +20w +8% is achievable each month, but I don’t think so, and neither does the raw outcome data.

Are any of you ~40 year olds getting +20w month after month?? What about the younger cohort?

I would like to see the FTP prediction model be more conservative (not the workout selection). Give me an achievable predicted gain of 5%, then push the prediction up if I am doing well. The current model always ends in a lower number but still generally a solid gain. +10w a month is great and super motivating, I don’t need nor can achieve +20w.

The workout difficulty and recovery is fine, I don’t want the model to select easier workouts (not the first week anyhow, the model is adjusting the following weeks), or do workouts less often. It would just be amazing if the prediction was more realistic.

I just got the new 305W prediction on Sunday for the June training outcome. Another +20w! this is super aspirational. To achieve this I would need to quit work and ignore my family, eat, sleep and recover perfectly.

Note, 305w would put me at 4.4w/kg. I could easily loose 3kg which would put me over 4.6w/kg at 40 years old which is mad. This would be a dream for my key race in October, I don’t need this now, nor can I achieve it.

It would be awesome if I could tell the program, lets not plan week 2 and 3 workouts for a +19w gain to 305W, lets plan these sessions around a more realistic +10w gain, if the first week goes well, lets up the intensity of week 2 and 3 and see if we can get to say +15w.

The model does not seem to learn from prior months training outcomes when predicting the next months FTP gain. It would be amazing if if could. A human coach could see and do this.

@eddie Id be keen to hear your thoughts on my training plan settings / progress / AI Predictions etc.
My interval workouts are set to inside so I shouldn’t be seeing the outside workout Bug. I am happy for you to share on the Forum.

I’m hesitant to lower the setting from balanced as the workouts selection is spot on. I don’t want to game the system to get the BETA AI Prediction to be accurate at the expense of good training.

eddie 245

It looks like you’re pretty fit right now. You’ve set quite a few all-time PRs over the past six weeks. Nice work! :flexed_biceps:

It seems like what you’re doing is bringing good results, and if you’re enjoying the process, I’m not sure I’d change much.

If you’re curious about the predicted gains and why you might not be realizing what was initially predicted, I’d say that you might be influenced by that outside bug to some extent, since you do ride outside often and have recently done outside workouts. Your Sunday rides are scheduled to be outside, too, so that is likely playing a role.

Aside from that, though, it’s really rare for you to hit all three of those hard workouts in a week. Oftentimes, you hit two and then get in a fair bit of unstructured riding. The same goes for that dynamic Sunday endurance ride – I haven’t seen you knock out a long weekend ride in quite a while.

If you’re really looking to try to squeeze all of the gains you can out of yourself, then I’d definitely recommend following your plan as closely as possible. If I were in your shoes, though, and enjoying my time on the bike, I wouldn’t change anything. :raising_hands:

3 Likes
246

Predicted: 288w

Actual: 284w

Came close to prediction and actual aligning but a heatwave hit the UK and I melted during the final workouts.

Super stoked, fairly sure I’m now 8-10 weeks ahead of 2025 fitness for the same hours.