Also how would we control for environmental health effects, or even interactions between multiple product variances. This kind of testing would be wildly expensive, pose potential public health risks, and the data collected would be coarse and noisy at best
The only difference is it very slightly adjusts the quantity - and does so in a way that is within existing allowed tolerances, so effectively this might already be happening, just we aren't collecting the results.
https://en.wikipedia.org/wiki/Estimated_sign
By tightening the required tolerances, londons_explore's proposal could be implemented without exposing anybody to different levels of product than are already permitted.
However, you have to get people to tell you how much product they use, which brands, how often, and not lie about it. That's basically impossible without antisocial levels of monitoring (which supermarkets probably already do, but which they don't want to tell their customers they do).
It's like the early days of mass customer data tracking where they'd tell customers they were probably pregnant, or probably gay, or whatever. People didn't receive the information in a way that suggested they were happy about the data being gathered.
It's probably too unethical to even use that data. Unless you're only trying to sell more crap, then it's A-OK! /s