AI FTP Detection or Raising FTP Plan…one, or both, can’t be working as intended. I’ve been using Trainer Road for several years. Love it and it has helped me become a better cyclist. However, I’ve gone through two consecutive 8 week ftp plans for a total ftp “gain” of -7!! Either the AI ftp detection algorithm is bogus or the plans don’t work. Both can’t claim to be accurate and have this occur. I’ve followed these plans to a T, only slightly decreasing volume the week of Christmas for travel and one brief three period of a head cold. I’ve done the surveys, wore a heart rate monitor for most interval works, completed ALL of the hard intervals and only substituted some easy recovery rides for Zwift. What gives? I’m stronger than ever. I’ve weight trained, I feel good after the workouts. What am I doing wrong? I’ve got some big goals for May and June and I’m considering just cancelling TR and putting that money towards a coach instead. I like TR, I want it to work for me but help me make sense of the ftp build plans or AI detection not in alignment when used exactly as intended.
What plans did you follow over the period? If you stuck to the plan it does indeed seem odd for FTP detection to drop you unless it was less than you were doing before?
Your post does not give enough information to analyse possible causes.
Make your TR profile public and post link to your calendar.
Or ask TR support to take a look and give you some feedback.
Nope, actually an increase in volume.
I followed the mid volume custom ftp build program.
I think the concern is valid. If I had properly completed 16 weeks of FTP plans and achieved -7 FTP, I would also be thinking something was wrong somewhere too.
Maybe an easy way to find out if it is the AI, do an FTP test - the type you did prior to using AIFTP.
The concern’s valid, I’d be a bit annoyed as well if my FTP went down over a proper training block (assuming I was not like, already peaking before I started), but there’s not really enough info here to accurately diagnose - could be any number of things.
E.g., could be the training in the plan has been mostly indoors and prior to that they were riding mainly outdoors - well most riders can do better numbers/lower HR outdoors (better cooling etc), but TR can’t see the difference, so TR would just see slightly worse performance.
TR support should be able to give a better answer in any case because they’ll actually be able to see your training history and they know what the algorithm is doing (neither of which are visible to us here).
But anyway: if you’re feeling stronger and setting PRs etc outside then to some extent is the number that a piece of software is giving you really that important? At the end of the day what matters is has your performance actually improved? All FTP numbers are just estimates and will vary day to day, in the end it is just a number of a screen in an app, not real life… Without knowing your starting FTP, 7 watts might only be a couple of %.
Thanks. That is my plan, interested to see what an ftp test (or maybe two different tests) show in regards to my current ftp.
Thanks!
I find with TR AI detection if you miss a couple of blocks like you have, it won’t budge.
You said you followed the plan to a T, but you haven’t you reduced volume at Christmas and a period with a head cold. I’ve found once I have nailed a complete month following the plan to an actual T I get results.
Hey, @mheist, welcome to the forum! 
I’ve taken a look at your recent training history, and I’ve noticed a few things.
First, I’m wondering what your FTP assessment timeline looks like. Your most recent FTP update was on April 3rd of last year, and that figure remains current on your account. Are you tracking your FTP somewhere else? Also, does that mean that your last AI FTP Detection gave you an estimate of 7 watts less than your current FTP shown in TR?
Next, I noticed that you’ve recently come off of a long break with no training, and while you did start riding a bit late last year, your structured training really became consistent at the start of 2025.
It looks like things started off really well, but then you ran into some illness in early February. The following week looked pretty good again, but since then, things have been a bit off. You haven’t been following a plan, and your training structure has dropped off. You’ve been doing a lot of riding in Zwift, and I wonder if that’s had an effect on the more important structured work.
Generally speaking, I’d be curious to see where you’re getting your recent FTP numbers from and how regularly you assess your fitness. Also, given the short duration of time since you’ve been training consistently, the illness, and the recent stagnation in structured training I’d say that it is possible that you haven’t had any major breakthroughs just yet, but it’s really hard to make that assumption without knowing what your FTP actually was at the start of the year. All I know is what it was last April. 
Let me know your thoughts. I’m sure we can find a good path for you moving forward.
Best of luck with your training!
I have been surprised a few times now by the difference between the opening post in a thread like this one and the contribution by a TR staffer that has looked at the training history of the opening poster.
In this case, I stick with my previous comment “I think the concern is valid. If I had properly completed 16 weeks of FTP plans and achieved -7 FTP, I would also be thinking something was wrong somewhere too.”
But, based on Eddie’s contribution, maybe “properly completed 16 weeks of FTP plans” should have been in bold italics
By contrast, I take such responses from TR to be a certainty when threads like this are started. There have certainly been occasions where AI and FTP detection have been off, but overwhelmingly, the case is almost always that the athlete has either not been as diligent as they thought, or they internally downplay the significance of time off or lack of plan adherence.
(Not a criticism of those posters…I think it is very easy to convince yourself you are doing the work)
I was thinking the same thing recently about AI detection and my FTP. My solution was to do a ramp test and push through. Interestingly, since I’ve only been doing sweet spot training lately (my conclusion anyway), my ramp test result was lower than my AI-detected FTP. It had been years since my last ramp test, and I had forgotten how deep you have to push—it’s a skill in itself. In the end, I decided to alternate between AI and ramp tests for a balanced approach. (I kept the higher AI FTP number 
)
Have you actually completed ramp tests?
My ramp tests can vary by 15 watts if I take them when I am run down after a business trip. Yes I have been known to do ramp tests. A few days later and come back with the gains expected in the second test.
There was an interesting GCN video recently. One of the presenters did a ramp test every day for seven days…
Small nit, but “The Grade” is not a ramp test…it is basically a 20 min FTP. You ride up the climb as hard as you can and Zwift uses you AP to determine your FTP.
Speaking of this, ftp can be all over the place. I’ve taken to testing ftp toward the end of my 3 week blocks in order to test my fatigue durability. Usually do the kolie protocol but last time went out and just blasted 6x10. Once you get a a good sense of what ftp feels like it’s pretty easy to feel it out on any given day. Ramp tests are kind of a waste in my opinion unless you just need something short. With the longer form test, you get a solid threshold session in as well as an ftp estimate and your tte. Do long form, or just trust the AI and see if you can complete a 4x10 (Lamarck).
Ramp tests are not accurate enough to be relied on for FTP.
Really depends on where you are on the curve. For me, they are nearly perfect. For others, not so much.
People do put too much stake in them. People really should learn how to feel their own FTP, or shouldn’t take long.
Mine are accurate and consistent when sleep is normal. I have followed up on 20 minute tests with ramp tests within a few days and received same results.
The one thing that really throws an FTP test for me is being run down on no sleep/jet lag. Legs will fail before HR even gets much above RP. Tends to occur in days HRV is trashed.
Small skews I’ve observed. Throwing a hail Mary standing sprint for final minute can over state by 3-5w if you are a majorly anaerobic rider (I am) Result would be workouts that may be difficult to complete in 90-115% intervals. Stay seated…
I’ve observed riders that are terrible sprinters with a domestique profile get lower ramp results than their 20 min test. Those guys of course hate them because they get a lower number 
For most everyone they are repeatable and reliable. I like to do them with some fatigue in my legs, because that’s how I will normally be training anyway. I’ll save my 20 minute threshold effort for a race effort, thank you very much.