Hospitals worldwide are increasingly adopting patient-facing AI technologies to improve operational efficiency and patient experience. Yet, rigorous causal evidence on their real-world effectiveness remains limited. We conducted a large-scale randomized field experiment with a major public hospital system in China to evaluate an AI navigation system designed to guide patients through complex outpatient workflows. Over a three-month period, the system was deployed to nearly one million patients, spanning 1.7 million outpatient visits. Contrary to expectations, we find no significant reduction in total visit duration. Further analyses reveal that these null effects are driven by two factors: extremely low patient adoption (3.2%) of the AI navigation system and the system’s inability to address core operational bottlenecks, such as limited physician capacity and rigid scheduling, as reflected by unchanged stage-to-stage time intervals within the outpatient process. These findings underscore how behavioral frictions and institutional constraints can undermine the intended benefit of patient-facing AI technologies, highlighting the need for integrated design strategies that align AI technologies with patient behavior, clinical workflows, and organizational processes.