Human-Like Social Robots in Retail: Investigating Perceived Similarity and Physiological Responses in AI-Driven Recommendations

Carina Wiedenhöft, Alexander Piazza, Carolin Kaiser, Anna Kaindl

Keywords: Human–robot interaction, social robots, personalization, perceived similarity, trust, comfort, intention to use, electrodermal activity, retail services

TL;DR: A study found that personalizing a retail social robot's speech to match customers offered no significant benefit over a standard "good sales assistant" approach, suggesting default well-designed behavior is sufficient.

Abstract: AI-powered social robots are increasingly discussed as a way to improve the in-store customer experience, but the effects of similarity-based personalization remain unclear. This study examines whether language-based similarity features in a retail recommendation dialogue improve user comfort, trust, and intention to use. In a controlled laboratory experiment, participants interacted with the social robot Furhat either in a basic “good sales assistant” or in an adaptive, similarity-based variant (speech rate, volume, pitch and similarity statements). Questionnaire data and continuous electrodermal activity (EDA) were combined to capture both self-reported and physiological responses. Using robust mean comparisons (Yuen tests), structural equation modeling (SEM) and EDA analyses, no significant differences between conditions in terms of comfort, trust or intention to use were found. However, SEM revealed stable internal mechanisms: rapport strongly predicted comfort, usefulness strongly predicted trust and comfort predicted intention to use. These results suggest that general similarity matching in early social robot interactions in retail is not reliably beneficial. In retail sales consulting, it seems sensible to leave robot behavior unadjusted and to base it on a generally good sales consultant. Improvement through personalization is not generally advantageous but it depends heavily on the context.

Submission Number: 5