That's only true if there's a fixed set of work to get done. But that's rarely the case. Often, management has N different things they could have done, and enough people and time to do M of them, for M < N. Which ones should they do? Well, whatever maximizes profits. So they (management) estimate income from each thing that could be done, and ask engineering (hopefully) to estimate how much it will cost to implement (or how long it will take, which equates to cost). Then they make a (hopefully) more informed decision than they otherwise could have made.

Look, there's lots of ways this gets done badly. I get that. But the idea itself is not nonsense.