Well at least if you pull numbers out of your arse and then make a decision based explicitly on the assumption that they are valid, the decision is open to rational challenge by showing that the numbers are wrong when more evidence comes in. And who knows, the real numbers may be close enough to vindicate the decision. If you just pull decisions out of your arse without reference to how they relate to evidence (even hypothetically), you are denying any method of improvement other than random trial and error. And when the real numbers become available, you still don't know anything about how good the original decision was.

That's a good point. Plugging gut assumptions into models to make sure that the assumptions line up with each other generally produces better results for me. Beyond it just feeling better, it gives me things I can go away and test that I'd never have got otherwise. Like if I think something's 75% likely to happen in X period and I think that something else is more likely to happen than that - do I think that the second thing is 80% likely to happen? And does that line up with information that I already have? Numbers force you to think proportionally. They network your assumptions together until you can start picking out bits of data that you have that are testable. Intuitions aren't magic, of course, but they're rarely completely baseless.