AI-Enabled Modeling and Weighted-Utility Decision Support for Consumer Complaint Resolution Processes

Abstract

Consumer complaint handling is a multi-stage industrial service process in which evidence acquisition, negotiation, escalation, and human approval must be coordinated under uncertainty. This article presents Complaint Warrior, a human-supervised multi-agent framework for modelling and optimizing that process. The method represents each dispute as a persistent state containing events, evidence, constraints, inferred participant states, process indicators, and admissible actions. Specialized agents perform intake, narrative reconstruction, negotiation planning, financial-dispute preparation, public-communication support, and small-claims preparation, while a supervisory agent ranks actions through a configurable single-objective weighted utility function. A prototype evaluation used 120 realistic dispute scenarios and 48 volunteer users across manual resolution, large-language-model drafting, and the full framework. Complaint Warrior achieved 95.2% evidence coverage, reduced mean preparation time from 32.8 to 9.4 min, increased the observed settlement rate from 40.0% to 82.5%, and reduced mean resolution time from 24.7 to 11.8 days. Users reported lower workload and high trust, explainability, and willingness to reuse the system. Inferential analyses showed significant improvements over both comparison conditions for six continuous outcomes; settlement rates and reported cluster-adjusted intervals are presented without model-based pairwise significance results. These exploratory results indicate that persistent state modelling and coordinated AI decision support can improve the throughput and consistency of service-recovery operations. The study also identifies validation, governance, privacy, and jurisdiction-specific compliance requirements for industrial deployment.