Methods for selecting promising expert system applications

Two well-known approaches to selecting promising expert system applications, the checklist approach and the generic task approach, are compared and evaluated. Improvements to and criticisms of the two methods are discussed. A new method, the cognitive-technology fit approach, that attempts to overcome some of the deficiencies found in the other two methods is introduced. The cognitive-technology fit method consists of four steps: decomposing the task into cognitive primitive subtasks; matching the subtasks to currently feasible artificial intelligence (AI) techniques; filtering out subtasks with AI techniques in which the designer is not knowledgeable; and prioritizing the remaining subtasks according to user management needs. The three methods are demonstrated by applying them to a typical application, the Taxpayer Service Assistant