E2E Process Automation Leveraging Generative AI and IDP-Based Automation Agent: A Case Study on Corporate Expense Processing

Abstract

This paper presents a case study of end-to-end (E2E) automation of corporate financial expense processing by combining generative AI (GenAI) and intelligent document processing (IDP) technologies with automation agents and shows the automation of intelligent tasks in a modern digital transformation environment. Although conventional RPA is effective in automating repetitive, rule-based, and simple tasks, it has limitations in handling unstructured data, responding to exceptions, and making complex decisions. In this study, we designed and implemented a four-step integration process, including automatic recognition of proofs such as receipts through OCR/IDP, item classification based on policy database, intelligent judgment support for exceptional situations through GenAI (LLMs), and human final decision and system learning (human-in-the-loop) through automation agents. As a result of the application to Company S, a large Korean company, quantitative effects such as reducing the processing time of branch receipt expenses by more than 80%, reducing error rates, and improving compliance rates were confirmed, as well as qualitative effects such as improving work accuracy and consistency, increasing employee satisfaction, and supporting data-based decision-making. In addition, the system learns from human judgment and continuously improves its ability to automatically handle exceptions, creating a virtuous cycle. This study empirically demonstrates that the organic combination of GenAI, IDP, and an automation agent overcomes the limitations of existing automation and is effective in realizing E2E automation of complex corporate tasks. In addition, it suggests the possibility of expansion to various business areas such as accounting, human resources, and purchasing in the future, as well as the development direction of AI-based hyperautomation.

Received: 30 May 2025 | Revised: 30 July 2025 | Accepted: 22 September 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Author Contribution Statement

Cheonsu Jeong: Conceptualization, Supervision, Writing – original draft. Seongmin Sim: Software, Writing – review & editing. Hyoyoung Cho: Software, Writing – review & editing. Sungsu Kim: Software, Writing – review & editing. Byounggwan Shin: Software, Writing – review & editing.