Citation:

Himnish A, Mayank Singh, Ujjol Chakraborty, Dr. Vasudha Vashisht, 2025. "Beyond Static Retrieval: Evaluating Agentic RAG for Sales Intelligence Applications", ESP Journal of Engineering & Technology Advancements  5(4): 47-57.

Abstract:

The quick growth of enterprise data has made Retrieval-Augmented Generation (RAG) a critical approach for enabling precise and context-rich responses in domain-specific applications. In the sales intelligence domain, where decision- making relies on both structured and unstructured data, conventional RAG approaches—such as vector-based retrieval, graph-augmented retrieval, and hybrid frameworks— usually lack balancing factual accuracy, contextual reasoning, and adaptability to diverse query types. This paper introduces and evaluates an Agentic RAG for Sales Intelligence (ARSI) framework, which combines agent-driven orchestration with tool-enabled retrieval strategies to dynamically decide when to query structured sales databases, consult vector embeddings of unstructured data, or incorporate external knowledge sources. Unlike traditional RAG pipelines that rely on static retrieval, Agentic RAG employs a reasoning-acting loop (ReAct-style architecture) to tailor retrieval strategies to the query intent, thereby improving accuracy for quantitative fact-based questions and providing more accurate information for explanatory or causal queries. We make comparison of ARSI with vector-based, graph-based, and hybrid RAG approaches using sales intelligence scenarios such as quarterly revenue reporting, root-cause analysis of sales fluctuations, and cross- market. Our findings show that while vector and graph methods are better in semantic matching and relational reasoning respectively, the agentic approach achieves superior performance in adaptability, interpretability, and integration of multi-source evidence. This work places Agentic RAG as a robust methodology for next-generation conversational business intelligence systems in the sales industry.

References:

[1] A. Srivastava, S. Agarwal, and M. Bhattacharya, “Conversational Business Intelligence: From Static Dashboards to Intelligent Assistants,” IEEE Access, vol. 10, pp. 122351–122369, 2022.

[2] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge- Intensive NLP Tasks,” Advances in Neural Information Processing Systems (NeurIPS), 2020.

[3] Y. Chen, M. Fang, and Z. Lin, “Graph-RAG: Enhancing Retrieval- Augmented Generation with Knowledge Graphs,” arXiv preprint arXiv:2310.07514, 2023.

[4] S. Khattab, J. Saad-Falcon, and C. Potts, “Hybrid Retrieval-Augmented Generation: Leveraging Heterogeneous Knowledge Sources for Factual Consistency,” Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL), 2024.

[5] S. Yao, J. Zhao, D. Yu, I. Shafran, T. Shi, P. Liu, C. Narayanaswami, and R. R. Salakhutdinov, “ReAct: Synergizing Reasoning and Acting in Language Models,” arXiv preprint arXiv:2210.03629, 2022.

[6] J. Xu, Y. Zhang, K. Wang, and H. Lin, “Agentic Retrieval-Augmented Generation: Tool-Integrated Reasoning for Dynamic Knowledge Access,” arXiv preprint arXiv:2405.07872, 2024.

[7] H. Luo, Z. Liu, and M. Sun, “AgentRAG: Agent-Enhanced Retrieval- Augmented Generation for Adaptive Knowledge Integration in Biomedical QA,” arXiv preprint arXiv:2403.16212, 2024.

[8] A. Pundir, M. Nguyen, and J. Yang, “LegalBench-RAG: A Retrieval- Augmented Generation Benchmark for Legal Reasoning,” Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL), 2023.

[9] V. Karpukhin et al., “Dense passage retrieval for open-domain question answering,” Proc. 2020 Conf. on Empirical Methods in Natural Language Processing (EMNLP), pp. 6769–6781, 2020.

[10] Y. Sun, Z. Yu, and W. X. Zhao, “Survey of retrieval-augmented large language models,” arXiv preprint arXiv:2312.10997, 2023.

[11] Y. Luo, X. Lin, and W. Wang, “Natural language interface for relational databases: A survey,” IEEE Trans. Knowl. Data Eng., vol. 32, no. 1, pp. 124–142, Jan. 2020.

[12] R. C. Basik et al., “Conversational business intelligence: An integrated approach for querying enterprise data,” Proc. 2021 IEEE Int. Conf. on Big Data (Big Data), pp. 1232–1241, 2021.

[13] N. F. Liu, et al., "Lost in the Middle: How Language Models Use Long Contexts," arXiv preprint arXiv:2307.03172, 2023.

[14] S. Yao et al., “ReAct: Synergizing reasoning and acting in language models,” arXiv preprint arXiv:2210.03629, 2022.

[15] L. Schick et al., “Toolformer: Language models can teach themselves to use tools,” arXiv preprint arXiv:2302.04761, 2023.

[16] A. Setlur, B. Lee, and M. Tory, “Conversational interfaces for data analytics: Opportunities and challenges,” IEEE Computer Graphics and Applications, vol. 40, no. 4, pp. 99–109, 2020.

[17] B V, P., Kumar, A., & N, A. K. (2024). "HybridRAG: A Powerful Strategy for Combining BM25 and Vector Search." arXiv preprint arXiv:2408.04948.

[18] Cormack, G. V., Clarke, C. L., & Buettcher, S. (2009). "Reciprocal Rank Fusion Outperforms Condorcet and Individual Rank Learning Methods." Proceedings of the 18th ACM conference on Information and knowledge management.

[19] Pan, L., et al. (2024). "Unifying Large Language Models and Knowledge Graphs: A Roadmap." arXiv preprint arXiv:2306.08302.

[20] Nogueira, R., et al. (2019). "Passage Re-ranking with BERT." arXiv preprint arXiv:1901.04085.

Keywords:

Retrieval-Augmented Generation (RAG), Agentic RAG, Sales Intelligence, Conversational Business Intelligence, Vector-based Retrieval, Graph-based Retrieval, Hybrid RAG, ReAct Architecture, Enterprise Data, Knowledge Retrieval.