FinGraph-MAR: A Graph Retrieval-Augmented Multi-Agent Framework for Intelligent Financial Back-Office Automation and SaaS Revenue Management
DOI:
https://doi.org/10.66238/fsrma133Keywords:
Financial back-office automation, Large language models, SaaS revenue management, Knowledge graphs, Risk-aware decision-making, Robotic process automation, Graph retrieval-augmented generation, Multi-agent systemsAbstract
The increasing complexity of financial back-office operations and SaaS revenue management has created significant challenges for intelligent automation, particularly in transaction processing, billing verification, revenue recognition, compliance monitoring, and operational decision-making. Existing robotic process automation (RPA) systems mainly rely on predefined workflows and struggle to handle dynamic business relationships, heterogeneous financial data, and knowledge-intensive tasks involving customers, contracts, subscriptions, transactions, and regulatory requirements. To address these limitations, this study proposes FinGraph-MAR (Financial Graph-based Multi-Agent Reasoning), a graph retrieval-augmented multi-agent framework for intelligent financial back-office automation and SaaS revenue management. The proposed framework constructs a heterogeneous financial-SaaS knowledge graph to model entities, events, business relationships, and operational rules, and introduces a hybrid graph-semantic retrieval mechanism to provide relationship-aware contextual knowledge for large language model (LLM)-based agents. Furthermore, FinGraph-MAR integrates a multi-agent collaboration architecture consisting of task planning, transaction analysis, revenue management, risk assessment, compliance verification, execution, and audit agents, together with a risk-constrained decision gate for reliable automated operations. Experimental results on four publicly available datasets, including FinQA, TAT-QA, UCI Online Retail, and IBM Telco Customer Churn, demonstrate that FinGraph-MAR achieves 95.6% task understanding accuracy, 94.9% F1-score, and 97.1% workflow completion rate, outperforming rule-based RPA, standalone LLM, conventional RAG, LLM+RPA, and Graph RAG+LLM baselines. Ablation experiments further confirm the contributions of graph retrieval, multi-agent collaboration, risk gating, and audit feedback.
References
[1] Li, Siyu, Rongshuang Sun, and Jinlin Hu. "Adaptive risk control model: A customer credit default detection framework based on random forest and neural network." Journal of Technology Innovation and Engineering 1.3 (2025).
[2] Aguirre, Santiago, and Alejandro Rodriguez. "Automation of a business process using robotic process automation (RPA): A case study." Workshop on engineering applications. Cham: Springer International Publishing, 2017.
[3] Lacity, Mary, Leslie P. Willcocks, and Andrew Craig. "Robotic process automation at Telefonica O2." (2015).
[4] Willcocks, Leslie, Mary Lacity, and Andrew Craig. "Robotic process automation: strategic transformation lever for global business services?." Journal of Information Technology Teaching Cases 7.1 (2017): 17-28.
[5] Lou, Renze, Kai Zhang, and Wenpeng Yin. "Large language model instruction following: A survey of progresses and challenges." Computational Linguistics 50.3 (2024): 1053-1095.
[6] Wei, Jason, et al. "Chain-of-thought prompting elicits reasoning in large language models." Advances in neural information processing systems 35 (2022): 24824-24837.
[7] Wu, Shijie, et al. "Bloomberggpt: A large language model for finance." arXiv preprint arXiv:2303.17564 (2023).
[8] Yang, Hongyang, Xiao-Yang Liu, and Christina Dan Wang. "Fingpt: Open-source financial large language models." arXiv preprint arXiv:2306.06031 (2023).
[9] Chen, Zhiyu, et al. "Finqa: A dataset of numerical reasoning over financial data." Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.
[10] Zhu, Fengbin, et al. "TAT-QA: A question answering benchmark on a hybrid of tabular and textual content in finance." Proceedings of the 59th annual meeting of the Association for Computational Linguistics and the 11th international joint conference on natural language processing (volume 1: long papers). 2021.
[11] Asai, Akari, et al. "Self-rag: Learning to retrieve, generate, and critique through self-reflection." International conference on learning representations. Vol. 2024. 2024.
[12] Yan, Shi-Qi, et al. "Corrective retrieval augmented generation." arXiv preprint arXiv:2401.15884 (2024).
[13] Niu, Cheng, et al. "Ragtruth: A hallucination corpus for developing trustworthy retrieval-augmented language models." Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.
[14] Edge, Darren, et al. "From local to global: A graph rag approach to query-focused summarization." arXiv preprint arXiv:2404.16130 (2024).
[15] Yao, Shunyu, et al. "React: Synergizing reasoning and acting in language models." arXiv preprint arXiv:2210.03629 (2022).
[16] Schick, Timo, et al. "Toolformer: Language models can teach themselves to use tools." Advances in neural information processing systems 36 (2023): 68539-68551.
[17] Wu, Qingyun, et al. "Autogen: Enabling next-gen llm applications via multi-agent conversation." arXiv preprint arXiv:2308.08155 (2023).
[18] Guo, Taicheng, et al. "Large language model based multi-agents: A survey of progress and challenges." arXiv preprint arXiv:2402.01680 (2024).
[19] Ning, Zhongping, Haijiang Zeng, and Zhen Tian. "Research on data-driven energy efficiency optimisation algorithm for air compressors." Third International Conference on Advanced Materials and Equipment Manufacturing (AMEM 2024). Vol. 13691. SPIE, 2025.
[20] Wang, Tangtang, Kaijie Zhang, and Kuangcong Liu. "A knowledge graph and deep learning-based semantic recommendation database system for advertisement retrieval and personalization." arXiv preprint arXiv:2601.00833 (2025).
[21] Xu, Zitong, Yuqing Wu, and Yue Zhao. "TRIZ-RAGNER: A Retrieval-Augmented Large Language Model for TRIZ-Aware Named Entity Recognition in Patent-Based Contradiction Mining." arXiv preprint arXiv:2602.23656 (2026).
[22] Ling, Yunting, and Wenxuan Liu. "SR-DTMA: A Digital Twin-Driven LLM Multi-Agent Framework for Systemic Risk Simulation and Coordinated Decision-Making in Supply Chains." Journal of Computer, Signal, and System Research 3.1 (2026): 123-132.
[23] Gao, Ziyang, Yuanliang Qu, and Yi Han. "Cross-Lingual Sponsored Search via Dual-Encoder and Graph Neural Networks for Context-Aware Query Translation in Advertising Platforms." arXiv preprint arXiv:2510.22957 (2025).
[24] Zhang, Bingli, et al. "Consensus Matrix: A Role-Specialized Multi-Agent Framework for Structured Collaborative Decision-Making in Agentic Visual Media Workflows." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2026.
[25] Zhou, Junjie, et al. "Cgtgait: Collaborative graph and transformer for gait emotion recognition." 2025 IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2025.
[26] Zheng, Zhuo, Jinlin Hu, and Shini Yang. "Artificial intelligence-driven dynamic logistics risk modeling and data risk inspection mechanism research." Journal of Technology Innovation and Engineering 1.3 (2025).
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Yang Shi, Leshen Sui, Xingyu Liu
This work is licensed under a Creative Commons Attribution 4.0 International License.
License Terms (Recommended – CC BY 4.0)
All articles published in Fundamental Scientific Reports in Multidisciplinary Areas are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
This license permits unrestricted use, distribution, and reproduction in any medium, provided the original author(s) and source are properly cited. Users are allowed to copy, redistribute, remix, transform, and build upon the material for any purpose, including commercial use, as long as appropriate credit is given.
Authors retain copyright of their work while granting the journal the right of first publication. The journal provides immediate open access to its content on the principle that making research freely available supports a greater global exchange of knowledge.
For more information about this license, visit:
https://creativecommons.org/licenses/by/4.0/
