When AI Decides: A Field Experiment on Agentic Returnless Refunds

Status: Working Paper
Coauthors: Lu Fang, Zhe Yuan, Kaifu Zhang

Abstract

E-commerce returns impose substantial hassle costs on consumers and heavy financial burdens on firms. Advances in agentic AI now allow autonomous systems to evaluate multimodal evidence and make consistent customer service decisions, enabling returnless partial refunds that compensate customers without product returns. We collaborate with a leading global e-commerce platform to study this innovation through a randomized field experiment involving nearly one million customers. Treatment group customers received AI-generated returnless partial refund options, while the control group followed the traditional process requiring return shipment for full refund. We develop an economic framework that delineates drivers of the platform’s payoff and identifies the conditions under which AI-enabled refund policies create and maximize value. Overall, the agentic returnless partial refund policy increases refund payouts but also reduces processing costs and enhances long-term customer value, yielding a positive net effect on profitability. Customers exposed to the AI policy make more subsequent purchases but also exhibit signs of strategic behavior. The study provides one of the first large-scale causal evaluations of agentic AI in customer service, demonstrating its potential to transform return management with autonomous, scalable, and economically rational refund policies.

Key Words: AI agents, agentic AI, customer service, return policy, returnless refunds, field experiment