Can someone point me in the right direction on how to understand forecast AI and get it setup correctly? I tried using it once this past year but it said my fitness would actually go down for my target event. I just tried to set it up again for a race in March I want to target and it’s giving me projected numbers that are little changed from my current fitness and are well below numbers I was putting out in 2024. When I increase the target power numbers even a little it say they are not attainable, even though they are still numbers I can achieve with 100% certainty.
Pay attention to the TL and low/high/peak table deltas before running a forecast.
See this post: How Are You Using Xert’s Forecast AI Training Plans? - Training - Xert Community Forum
If you are not seeing what you like try different forecasts using Goal versus Event versus Race (with Readiness option). More than one road can lead to Rome.
You can choose Achievability between “will likely” versus “may not” or “will likely not” and review the forecast results before deciding which plan you prefer.
“will likely” is going to be a conservative prediction. I’ve hit or exceeded predictions prior to target date during my XFAI testing, but I’d rather do that than assume a loftier goal is attainable. YMMV
Tap the keys below the forecast chart to isolate whatever data set you’re interested in.
Here’s an example Goal forecast for an increase in 4 minute power over 9 weeks.
You can also hover over HIT matchsticks to get an idea what you’re in for during the plan.
how useful is the “AI” if I have to keep massaging the thing to get it to give me some sort of useful plan? Should it not be able to at least look at my past 1-2 years of data and see that I’ve easily surpassed the max power numbers It is predicting I can achieve by the target event? My training load is currently 121, I’ve been as high as 140 in the past 2 years. My TP is currently only 88% of what I’ve seen the past 2 years, my max power is down 200 watts due to the off season and my HIE is down 8. Maybe I’ll just stick to the old system until this AI becomes usable in some way.
1 Like
There is something amiss with the setup guide. I believe the issue is related to Readiness and Max Hours versus Acheivability. I will be reporting this issue once I can replicate the steps consistently.
IME Forecast AI works fine – and should accomplish what you want – by following these steps.
- If signature is stale ride one of the fitness tests in Slope mode to confirm your current signature. Even if no BT you’ll know if your sig is in the right ballpark.
- Select Change Program or Modify Plan and choose Goal.
- Enter Target Date and the Focus Type/Specificity from your Event or Race AI entry. For example, 22 March, Mixed Rouleur.
- Start increasing Focus Power watts while watching the TL and Signature tables.
Keep going until you see the TP and TL values you expect based on past performances. - Note the Acheivability text as it changes between green (“will likely”) to yellow (“may not”) to red (“likely not”).
- Slide Max hours/week to what you reasonably expect you can reach by your Target Date.
- Run Forecast regardless whether Acheivability indicates likely or not. Bypass any warning.
- Review the results. If Final Sig, Focus, and TL look good to you Save the forecast.
Using the steps above I can generate successful forecasts that meet or exceed numbers I have reached in the past (at associated TL) as long at Target Date provides a sufficient plan duration.
I can even get a green checkmark when the pre-run warning text is red (“likely not”).
When I inspect details on the forecast chart everything looks good for a plan to follow.
So the Acheivablity text warning can be false and that’s what I’ll be reporting as a bug.
I don’t think XFAI should automatically create a forecast that matches fitness levels in the past, but I understand why you think it should. Others are expecting the same behavior so it’s something Xert needs to address regardless which XFAI setup option you select.
1 Like
Thanks for the response,. I’ve just now had the time to sit down and work on this again. My signiture might be slightly off, TP is probably 10-15 watts higher than what my current signature shows since I haven’t done a real breakthrough effort in a month. I’ll do one on my next high intensity day where I’m fully recovered, as soon as possible.
When I run the forecast the numbers are very off. For example, I’ve had my TP as high as 325 the last 2 years and that was with just doing base training. I’ve had my HIE up to nearly 29 and my PP just under 1300 watts. The forecast says I can only reach 293 TP, HIE of 24.1 and a peak power of 1090. My HIE and PP numbers are just from this past summer, MY TP was in the 320 range at the time. I’m around 165 lbs with race weight being around 145 lbs.
Question for you, I guess it doesn’t really matter what it expects my numbers to be. Since my numbers will adjust as I get fitter and do breakthroughs, All the forecast does is prescribe me enough work in the zones I need to improve. Correct me if I’m wrong.
If I’m correct, then all good and I’ll just move forward with this. I’m not sure how it differs from the old non AI training plan however. I’ll try to spend some time doing reading and getting caught up on the changes.
I want to like Forecast AI, but I can never get it to work. I just asked it to project a training plan over a three-month period to raise my climbing power by 10W, and it required 13h per week to do so. Then I asked it to raise my climbing power by just 1W, and it also required 13h per week. Makes no sense to me.