I’d be interested to know how you see this working in broad strokes (no need to give away trade secrets).
Here’s my take on some difficulties that you might face:
Take the WKO4 solution which plots the Power Duration curve based on the rider’s Mean Maximal Power curve and then takes a turn-point on that line to be indicative of FTP. The model is fairly robust as long as you have enough max efforts.
For that type of model one of the problems I can see is that if you’re just training indoors (as I imagine a high percentage of folk do during the winter), or only have power data for indoor rides (do you know what percentages of users own a power meter?) then the PD curve is self-defining over the course of a plan. There would be nothing that you could tell about the riders abilities after the completion of the plan that you couldn’t have known before they started, assuming everything was completed as prescribed.
Without outdoor rides or tests there would be no max efforts since the intervals are not tests of MMP: My own ability to crush an 8 minute test far outweighs my ability to perform 4 x 8 with 3 minutes rest.
There is also the discrepancy in power between indoor and outdoor rides: it wouldn’t be appropriate to base indoor workouts on a model driven from maximal efforts done outside.
Obviously this is only one way of modelling FTP - there may well be others out there but I’m pretty sure they all require maximal efforts (tests) at different durations.
It’s an interesting topic and was actually the one that I had in mind when I first started the topic. Having said that, it is actually only the determination of the model that requires Big Data, the calculation of FTP only needs a single user’s data set.