Neither closed mathematical models for optimal subset selection, nor the opposite approach of intuitive decisions made by managers are applicable for large scale problems in which 700 out of 1200 items are to be selected. The approach suggested in this paper is an interactive man-machine system that includes: models for grading projects; a model for self learning of the system which utilizes the information produced during the decision process so as to update and improve project grading, which is in effect the system's recommendations for decisions; a model for examining and comparing actual policy with stated policy; a set of decision aids; and an algorithm for the dialogue with the Decision Maker.