Human-Centered AI for Real-Time Self-Checkout Assistance: An Event-Driven Architecture with Human-in-the-Loop Decision Support for Enhanced Customer Experience and Shrink Reduction

Purpose: Self-checkout (SCO) systems face persistent operational challenges including delayed assistance, inconsistent triage, and false-positive rates of 18 to 25 percent, contributing to industry shrink losses exceeding 112 billion dollars annually. Stores with high SCO utilization experience shrinkage 75 to 147 percent above industry averages. This paper proposes a human-centered artificial intelligence (AI) architecture grounded in explainable AI (XAI) and human-in-the-loop (HITL) principles, treating AI as decision support rather than autonomous enforcement. Methodology: The event-driven system prioritizes assistance events using multi-factor scoring based on wait time, anomaly likelihood, and associate workload, while preserving human discretion and providing transparent explanations for all recommendations. The framework is analyzed through a theoretical lens using parameters calibrated to publicly available retail industry benchmarks. Findings: Theoretical analysis suggests the framework has potential for significant reduction in assistance wait time, decreased false alert rates, and improved detection accuracy while maintaining associate decision authority. Unique contribution to theory, practice and policy: This analysis demonstrates how responsible AI design can enhance operational efficiency without compromising human autonomy or customer experience, offering a replicable blueprint for human-centered AI deployment in consumer-facing retail environments.