VIVID: Tool-Augmented Agents for Enhanced Tourism Management

Paper: VIVID: Bringing Tool-Augmented Agents into Enterprises

Authors: Yichu Fang, Ziqiang Zhang, Qiyang Shi, Zilong Wang, Lin Shen, Zhikang Ou, Jing Ma, Jiayuan Chen, Lei Sun, Dai Cheng, Yi Qiao, Yu He, Wei Zhang, Xiaoyu Shen

Executive Summary

This research introduces VIVID, a system designed to enhance the operational efficiency of tourism management through the use of tool-augmented large language model agents. These agents connect user requests expressed in natural language to various enterprise services. While they show promising capabilities, their effectiveness hinges on reliable execution, organizational oversight, and quick system response.

The study involved deploying VIVID in a real-world enterprise environment, analyzing its performance based on an audited dataset of 283 queries. The Qwen3-8B configuration of VIVID achieved a strict end-to-end success rate of 59.06% across various decision-making stages. This indicates that while the system can autonomously handle requests, there are limitations to its capabilities that need to be acknowledged.

One significant improvement noted in the research is the implementation of stage-wise KV-cache eviction, which led to a 34.4% reduction in mean latency. This means that the system became faster and more responsive when processing requests. Additionally, targeted optimization efforts focused on the five slowest business tools resulted in a dramatic reduction in their mean latency, between 81.1% and 92.3%. Such enhancements could lead to a more efficient workflow for organizations utilizing VIVID.

In a three-month deployment period, VIVID served over 1,500 personnel and significantly improved operational efficiency. The average processing time for each order dropped from approximately ten minutes to under two minutes. Moreover, the final-task error rate remained impressively low at below 1% after incorporating human oversight in the decision-making process.

These findings highlight that the viability of enterprise agents like VIVID is not just about the individual tools' accuracy but also about the overall system design and response dynamics. For businesses considering the integration of AI-driven systems, these insights suggest that careful attention to operational structures and control mechanisms is essential for achieving optimal results. The research underlines the potential for improved efficiency in enterprise operations through the thoughtful deployment of AI technologies, paving the way for future applications in various sectors.

Academic Abstract

Tool-augmented large language model agents connect natural-language requests to enterprise services, yet their operational viability depends on reliable execution, organizational control, and system responsiveness. We present VIVID, a deployed tourism-management agent system designed to study these requirements in a real enterprise environment. On an audited dataset of 283 queries, its Qwen3-8B configuration achieves 59.06% strict end-to-end success across the evaluated decision stages, indicating useful but bounded autonomous capability. Stage-wise KV-cache eviction reduces mean latency by 34.4%, while targeted optimization reduces the mean latency of the five slowest measured business tools by 81.1--92.3%. In a three-month deployment serving more than 1,500 personnel, VIVID reduced average per-order processing time from approximately ten minutes to under two minutes, with a final-task error rate below 1% after human-in-the-loop control. These results show that enterprise agent viability requires system-level consideration beyond isolated tool-calling accuracy.