Another relevant research result answers this question: When waterfall requirements analysis is attempted, how many of the prematurely early specified features are actually useful in the final software product? In a study [Johnson02] of thousands of projects, the results are quite revealing—45% of such features were never used, and an additional 19% were “rarely” used. See Figure 5.1. Almost 65% of the waterfall-specified features were of little or no value!
Craig Larman, Applying UML and Patterns: An Introduction to Object-Oriented Analysis and Design and Iterative Development (pp. 45), October 30, 2004, Prentice Hall