I only agree with that statement if you're drawing from the set of all possible problems a priori. For any individual domain I think it's likely you can bound your analytic. This ties into the no free lunch theorem.