I honestly think a lot of their 3px, 4px, or 5px decisions are actually an exercise of finding local maxima for their current design.
If they're looking for interfaces that are truly better, they would do far more testing on innovative/educated directions and iterate each of those to see how good they can get. Instead you just make one version as business-helpful as possible, which may be much weaker than your true potential.
I like the idea of data, but there's only so much progress you can make with the same designs in the same mold, tested pixel by pixel.
If they're looking for interfaces that are truly better, they would do far more testing on innovative/educated directions and iterate each of those to see how good they can get. Instead you just make one version as business-helpful as possible, which may be much weaker than your true potential.
I like the idea of data, but there's only so much progress you can make with the same designs in the same mold, tested pixel by pixel.