Reevaluating personalization in AI-powered service chatbots: A study on identity matching via few-shot learning

Abstract

This study explores the potential of AI-based few-shot learning in creating distinct service chatbot identities (i.e., based on gender and personality). Further, it examines the impact of customer-chatbot identity congruity on perceived enjoyment, usefulness, ease of use, and future chatbot usage intention. A scenario-based online experiment with a 4 (Chatbot identity: extraverted vs. introverted vs. male vs. female) × 2 (Congruity: matching vs. mismatching) between-subjects design with N = 475 participants was conducted. The results confirmed that customers could distinguish between different chatbot identities created via few-shot learning. Contrary to the initial hypothesis, gender-based personalization led to a stronger future chatbot usage intention than personalization based on personality traits. This finding challenges the assumption that an increased depth of personalization is inherently more effective. Customer-chatbot identity congruity did not significantly impact future chatbot usage intention, questioning existing beliefs about the benefits of identity matching. Perceived enjoyment and perceived usefulness mediated the relationship between chatbot identity and future chatbot usage intention, while perceived ease of use did not. High levels of perceived enjoyment and usefulness were strong predictors for the future chatbot usage intention. Thus, while few-shot learning effectively creates distinct chatbot identities, an increased depth of personalization and identity matching do not significantly influence future chatbot usage intentions. Practitioners should prioritize enhancing perceived enjoyment and usefulness in chatbot interactions to encourage future chatbot use.