Reducing Decision Latency in Hotel Operations: An Agentic AI Co-Pilot Approach using ReAct Framework

Abstract

Decision-making latency and high cognitive load often hamper hotel operations due to the passive design of modern Property Management Systems (PMS). Although the adoption of artificial intelligence (AI) in the hospitality sector is increasing, its implementation is still focused on the guest-facing services. Conversely, in the back office, most inefficiencies remain unresolved. This paper addresses this gap by developing an agentic AI co-pilot designed to assist front office staff as a proactive operational partner using natural language interaction. Using the ReAct (Reasoning and Acting) framework, static and administrative workflows are reorganized into an autonomous execution loop. The system is designed by integrating a Large Language Model (LLM) with a Server-Side Rendering (SSR) environment. Comparative analysis evaluated efficiency, accuracy, and robustness. Results show that the AI agent performs 6.34 times faster in generating performance reports. For the daily reservation workflow, there is also a 2.32-fold increase in speed compared to interactions through passive GUI navigation. Regarding reliability, the standard LLM had a hallucination rate of 76.7%, while the proposed agentic AI co-pilot demonstrated significantly higher reliability and achieved 90% resilience against robustness testing. Although the agent encountered minor failures, it achieved an average success rate of 93.3% across 30 test scenarios. Overall, the AI co-pilot with the ReAct framework has shown significant efficacy in addressing hotel operational inefficiencies.