Iām having a similar issue. I let my subscription lapse a few days in August and my FTP was messed up. Customer Service manually fixed it. But now I had another auto AI FTP and it dropped from 282 in August to 255 on Tuesday. Well tonight during my nearly 2 our ride I had a 60 minute high of 256w average and NP of 282?
I suspect this is due to unscheduled outdoor rides and no scheduled trainerroad indoor rides and the AI program is favoring TR workouts over the unscheduled workouts.
Hey @PavlovsDogzCyclist 
Apologies for the late response here.
The team looked at your AI FTP Detections, and this is not a bug. They confirmed that your latest AI FTP Detection of 255W is correct. This is a sensible detection since you havenāt done a lot of structured training recently.
We understand that it might seem like a big drop in FTP, but the overarching goal is to ensure that youāre getting workouts at the right intensity, so that you can be consistent in completing them. This will help your training in the long term.
Thank you for the reply.
I have valued my TR subscription and the AI FTP detection.
I have listened and watch just about every TR podcast & YouTube video since 2019, and have heard Jonathan and Nate preach many times how the TR AI program looks back at āALL YOUR" workouts to determine the right workout, every time.
Yes, I havenāt been doing a lot of unstructured training, I live in Denver and have been enjoying riding with my friends in the terrific weather, while I can! This is where AI canāt quantify the enjoyment of cycling. If AI canāt review my efforts based on HR data, power and RPE, maybe I need a different platform. Even Strava estimated my FTP at 270, which seems to be more in line with my recent efforts including an NP last week of 276 with a 60 minute average of 256w, and that wasnāt a 100% effort. I think I rated the RPE as Hard (it was hard +.5, I debated on rating it Very Hard, but I felt good enough to have a 1 minute sprint and still felt strong on the way home after the ride).
TR has all of my ride data, and hundreds of millions of records from other riders, but it seems that the program heavily weights the TR programmed workouts. I can see why, that is a lot easier to measure. But I have been setting segment PRs on my normal rides, including 1 minute, 20 minutes, 30 minutes, an a 1 hour efforts just last week that was higher that the AI assigned FTP. Thatās nonsensical.
Yes, I know that I can get out there and perform an FTP test, although that is somewhat difficult outside here as we have rolling terrain in my area and being in the city, its harder to find a road that I can just go flat out on for 20 minutes or longer without a traffic light, stop sign or sharp turn or downhill that messes with the effort (spinning out). And yes I have a trainer, but my Elite Direto is having issues with holding higher watts (the power fades over 200w, so it would be very difficult to hold 300w). But this is why I have opted to pay for and been loyal to TR for many years, to help establish this important number without performing an FTP test.
Here is my concern I have suspected this the past couple years. If the FTP is set too low and because I havenāt been doing structured rides combined with my Progression Levels that are lower than they should be, then I start training at a lower starting place than I should. It takes many weeks for the Progressional Levels to increase and a month for the FTP adjustment. Therefore Iāll be starting out behind schedule.
Here is a serious question that I have looked in the TR forum for, but I havenāt seen an answer to: When I am training with friends, should I add a TR outdoor workout that would have a similar overall effort and just do the group training ride? I assumed that TR was looking at my ride data regardless, and I am just taking a hit on the Progression Levels by doing unstructured rides, but that seems not to be the case as it also affected my FTP detection?
For now, I think I will override the FTP this time, 255 is too low.
Please let me know your opinion on if I should add a TR workout and just override that on my group training rides (I was doing that for part of this summer, but I started losing rides when my Hammerhead just turned off, so I stopped doing that).
J. Brian Anderson
Aka PavlovsDogzCyclist
You are free to use the value of 255 or 270 - I think there is enough slack with the progression level system to get you productive workouts whichever one you choose.
But TR thinks 255 is the optimum value to get you the best training so Iād use that. You will probably find that your AI FTP detections will increase fairly quickly while you train.
I wouldnāt get too hung up on what your ārealā FTP is - sounds like your fairly sure itās 270 
That being said, that trainer of yours doesnāt sound like it will be doing you any favours, could be affecting the validity of the measurements.
Keep the 270, do Marion on a trainer and see how it goes. Personally I would just go with the AIFTP, especially if all your riding has been outdoors
This 
Keep the 270 and do Marion, it should feel hard but is a good workout for FTP verification
Bit easy that.
At least Gray.
@PavlovsDogzCyclist of course We value you too!
AI FTP Detection does factor in both inside and outside power data. TrainerRoad is able to see the culmination of all of your efforts, including outdoor unstructured rides. So totally ok to enjoy riding with friends in this terrific weather!
That said, the most effective way to raise FTP is through structured interval training designed to target specific energy systems. While unstructured riding does bring fitness, it may not be the most efficient way to bump those FTP numbers.
On the other hand, you have Progression Levels, which are designed as an internal TR benchmark to recommend TR workouts. It is only natural for them to decrease if you are not completing structured TR workouts, but these adjust fairly quickly once you start completing workouts again. Progression Levels do not currently take into account unstructured riding, but this does not mean that if they go down, your AI FTP Detection does too. These can exist a bit independently of each other.
Using your recent PRs as a proxy for whether your FTP should be going up or down, can seem totally reasonable at face value. But the thing with these specific types of efforts (It sounds like a race-type effort or something more variable you are describing) is that they typically arenāt the best for getting an FTP estimate. Activities with larger discrepancies in normalized vs average power over some period of time (say 60mins), can make it tougher to glean a reliable FTP estimate from that activity alone.
Most other platforms use all-out efforts (commonly called capacitive efforts) to estimate your FTP using an FTP estimation model. This means that they rely on you completing maximal efforts within your training data.
For example, if you had a very hard 20-minute climb during a ride (ridden at ~ 105% or greater of current FTP), they might use that effort to estimate a higher FTP. Our model does not rely on capacitive efforts. Instead, our model predicts your FTP based on your training history and your personal biometrics, along with many other features. Capacitive efforts and FTP tests rely on one performance. AI FTP Detection accounts for much more with a broad range of data points.
In short, AI FTP Detection is designed to give you a training benchmark (FTP) that represents the full scope of your demonstrated abilities, so you get the most productive training possible.
As @UWDawgs mentioned, you can always perform the workout Marion to test out your FTP at 270. Thatās a 4.2 Threshold workout at 3x12 @ 98%. If you can do it, then youāll give TrainerRoad more data to lock you in. If you have problems in that workout, weād recommend moving forward with the original AI FTP detection.
Lastly, to answer your question about whether you should find a TR Workout that is similar to match your unstructured ride, the answer is typically no. We donāt recommend matching a TR Workout to a ride file unless youāve followed the workoutās structure because youāre essentially giving the system bad data (telling it you did something you didnāt, haha). This can negatively impact Progression Levels.
As I mentioned above, the system takes into account all rides with power data, whether theyāre indoors or outside, structured or unstructured. This can include group rides, races, and even rides on other indoor training apps synced to TrainerRoad.
Hope this addresses your concerns!
Thanks, Caro. I did Collingworth yesterday. On Zwift. With an FTP of 166. The first interval set a target of 183-190; the second of just over FTP; the third of 187-209; the fourth of 160 - 197. Which seems unusual for a Threshold 4.5 workout with an FTP of 166?
Collingsworth is (should be) 4x 7mins @ 105% FTP
So for an FTP of 166w each interval should be 174w 