The human factors aspects of designing and introducing automated services using automatic speech recognition (ASR) over the public switched telephone networks are examined. Current applications, the problems encountered, and the lessons learned are reviewed. In addition, the authors are currently engaged in a project to introduce automation using ASR, and they discuss the five-stage approach that they are using, which is based on published accounts of previous field trials as well as current techniques for user-interface design. They also describe more advanced capabilities that will be required for both near- and far-future applications, identifying areas where research is needed and discussing the difficult problems that must be overcome. In particular, data on the vocabulary requirements of human conversations are presented.>