Designing conversational agent user experience at home

Abstract

Since the advance in technology began the spread of the conversational agent (CA), CA has been adopted by various devices such as smartphones and smart speakers. CA is appealing to people by taking a step further from the conventional voice user interface that responds to voice commands to an AI agent that helps users with spoken conversation. Yet, compared to the high adoption rate for CAs, the frequency of use or dependency lags behind. In particular, the sales of smart speakers are exploding, but the continuous use of the CA of smart speakers is low, suggesting a problem in the user experience. This research aims to provide design implications for smart speaker CA experience in a home environment by exploring the problems users experience with the smart speaker CA and by identifying what factors affect the experience. To achieve this, this research includes a 12-week user study of CA use, from vague expectation to acceptance, to depict how the user experiences in each stage change over time. In addition, I suggested that the experience of having a spoken conversation with it, at-home use, and its independent speaker form affect user expectations and perceptions. As a next step, I conducted an observational study of the first-time users' CA use and revealed that they experienced frustrations due to inappropriate feedback form the CA in the initial stage. The mismatch between the mental model and the actual system also continued, implying deterioration of the later stage of use, together with the frustration. In order to solve this, employing refusal strategies used between people, the satisfaction and understanding of each feedback strategy were evaluated with first-time users. Finally, it was demonstrated that selecting the appropriate feedback strategy according to the limitation types of the CA improves the initial experience by reducing the early stage frustration. This dissertation suggests a direction to improve the features of a CA to meet users' expectations based on the user journey. It also suggests implications that will still be effective when the currently immature CA advances with technological development.