It is widely recognized that successful businesses usually fall into set routines and become limited by their past. To remain successful, they need to discover new opportunities and niches. Niches are surprising rules that contradict the set routines; they capture significant, representative client sectors that deserve new, more profitable treatments; they are not merely strong-rule and exception pairs. In this paper we study the efficient mining of set routines and niches. We also introduce a semantic approach to select a set of representative patterns, and present an efficient incremental algorithm to implement the approach.