Effectively, but not literally.

With perfect information I might expect 3% savings from optimal changes. As perfect information is NP-hard, I expect the the reality to be that attempting to cut the waste will result in mistaking useful things for waste, and cutting those useless-seeming-useful-things will cost at least as much as is saved from cutting the actually-useless-things.

This reply ran away from me a bit, for which I apologise. I don’t see how to improve it though.