JR East tests AI at ticket counters, reveals hurdles to full rollout

JR East conducted a proof-of-concept trial of generative AI at its Midori no Madoguchi (ticket counter) service to evaluate whether the technology could assist in proposing optimal travel routes to customers. The trial was designed to test whether AI could handle the complex task of route recommendation, which typically involves factors such as cost, travel time, number of transfers, and passenger preferences. During the trial, the AI system generated route suggestions and presented them to customers. In some cases, the AI's recommendations proved impressive and earned positive feedback — one observer reportedly remarked that the AI's output was so good that it made them feel they could not compete with it. However, the trial also revealed significant limitations. The AI's suggestions did not consistently match the quality and reliability of routes proposed by experienced human staff at the counter. The system struggled with edge cases, contextual reasoning, and the subtle judgment calls that seasoned employees routinely make. These findings expose a fundamental barrier to full-scale deployment: while generative AI can generate plausible route recommendations quickly, it lacks the refined domain expertise, error-checking capabilities, and adaptive learning that human ticket-counter employees possess. JR East's experience suggests that moving from a limited proof-of-concept to operational deployment at scale will require substantial additional work — including more sophisticated model training, integration with real-time operational data, extensive validation protocols, and ongoing human oversight — before the company can confidently replace or significantly augment human staff in this customer-facing role.