AI-enabled complaint handling and customer recommendation: A quasi-experimental difference-in-differences (DiD) perspective

As retailers increasingly deploy AI-based systems to manage customer complaints, fundamental questions arise regarding how such technologies shape customer evaluations, the psychological mechanisms through which they operate, and the boundary conditions under which they are most effective. This study investigates the impact of an AI-enabled complaint-handling system introduced by a large UK omnichannel retailer that automates triage, routing, and structured process updates, focusing on customers' likelihood-to-recommend (LR; 0–10) and unit-month Net Promoter Score (NPS), perceived provider competence, and the moderating role of complaint affective intensity. Using a sequential mixed-methods design, we first conduct semi-structured interviews with customers and frontline managers to explore competence cues, emotional expectations, and perceptions of AI-human role allocation. Building on these insights, we analyse customer-level survey data and administrative records through a quasi-experimental difference-in-differences (DiD) framework exploiting the retailer's staggered AI rollout. Quantitative results show that AI adoption increases LR by 0.38 points, equivalent to a 5.1% improvement over the pre-treatment mean. At the unit-month level, AI raises NPS by 4.1 points, a 22% relative increase. Perceived provider competence increases by 0.28 standard deviations post-AI adoption and mediates 45% of the total effect on LR, confirming competence as a central psychological mechanism. However, AI's impact is moderated by affective intensity: gains in LR are 37% smaller for high-affect complaints compared with low-affect complaints, indicating that AI-driven informational support is less effective when customers express intense negative emotion. Event-study analyses and robustness checks confirm the validity of the causal estimates. The findings provide practical guidance for retailers designing AI-enabled complaint-handling systems by highlighting when automation can enhance customer advocacy and when timely escalation to human agents is essential, particularly for emotionally intense complaints. These findings advance service recovery theory by identifying the conditions under which AI enhances customer evaluations and offer guidance for designing hybrid AI–human systems that balance efficiency with emotional responsiveness.