One of the hardest things in the granting game is dealing with the reviews. There's always some conflicting signal and some jerk who didn't read the section on the exact thing they claim is wrong with the proposal. While the natural reaction is to blow off the reviews that appear to be blind to half your proposal, after I sit with the reviews for a bit I am always amazed at what they tell me about what I wrote.
There's the obvious things like lack of clarity. If a couple reviewers all bring up the same point that you thought you addressed, there's a good chance you either didn't make the case you thought you were making or that you need to clarify what you wrote to better guide the reviewer to the intended conclusion. This kind of stuff plagued me early on and it took a while to craft the right way to push an idea that might be controversial.
Another point which has taken me longer to appreciate is the "unreasonable amount of demanded data" issue. Ever gotten a review back where a reviewer or two start asking for the world on a plate? I've had reviewers demand things that would require 3x the typical NSF budget and the invention of new technology. On first read I thought, "What an asshole, that's not even possible!", and while it is still not possible to meet those demands, after a while I began to realize that the demands were so high because that was the level of data the reviewers thought I would need to answer the questions I was proposing. Now who's being unreasonable?
I'm not at all saying that every reviewer has completely valid points all of the time, and often there is a middle ground between what you think needs to be done and what the reviewers are going on about. But I think the proportion of reviews that have something important to be heard is very high. The issue is tuning your hearing to the right frequency.