There's a tremendous popular fallacy which holds that significant research can be carried out by trying things. Actually it is easy to show that in general no significant problem can be solved empirically, except for accidents so rare as to be statistically unimportant. One of my jests is to say that we work empirically — we use bull's eye empiricism. We try everything, but we try the right thing first!
Edwin Land - The Journal of Imaging Science and Technology, Vol. 37, No. 3 (1992), p. 537
The notion that the architecture will emerge, that requirements appear as we go along spending other peoples money, and the customer knows the right solution when they see it, is just that "notional."
Without some agreed upon description of what Capabilities are to be provided by the system, you'll just be spending time and money to find out you don't know what DONE looks like.
