["Journal of Consumer Behaviour, EarlyView. ", "\nABSTRACT\nAlthough artificial intelligence and robot services are widely regarded as the future of the hospitality industry, unmanned smart hotels centred on service robots face significant risks of service failures. Hence, it is worth exploring whether human employees should intervene in the apology process when service failures occur. Based on feeling‐as‐information theory and attribution theory, this study conducts four experiments to explore the interaction effect between service failure severity (high vs. low) and apology agents (robot‐only vs. human‐robot collaboration) on customer forgiveness in unmanned smart hotels, the mediating roles of cognitive fit and perceived empathy, and the moderating role of customers' regulatory focus (promotion vs. prevention). The findings reveal that the robot‐only (vs. human‐robot collaboration) apology leads to higher customer forgiveness in low‐severity service failures. This interaction effect is mediated by cognitive fit. The human‐robot collaboration (vs. robot‐only) apology leads to higher customer forgiveness in high‐severity service failures. This interaction effect is mediated by perceived empathy. Customers' regulatory focus also plays a moderating role in the influence path. For promotion‐focused customers, the interaction effect exists. The robot‐only (vs. human‐robot collaboration) apology leads to higher customer forgiveness in low‐severity (vs. high‐severity) service failures through cognitive fit (perceived empathy). For prevention‐focused customers, the human‐robot collaboration (vs. robot‐only) apology leads to higher customer forgiveness regardless of service failure severity. The findings provide novel insights into the field of service failures in unmanned smart hotels and offer practical guidance for service recovery strategies.\n"]