LLM-Powered Root Cause Intelligence for Inventory and Shipment Discrepancies in Warehouse Management Systems

Sai Rohit Jonnada; Jaya Surya Teja Gaddam; Mohan Krishna Oorugonda

doi:10.69968/ijisem.2026v5i430-38

Authors

Sai Rohit Jonnada Louisville, Kentucky, USA, 40222
Jaya Surya Teja Gaddam Louisville, Kentucky, USA ,4024
Mohan Krishna Oorugonda Coppell, Texas, USA,75019

DOI:

https://doi.org/10.69968/ijisem.2026v5i430-38

Keywords:

Large Language Models, Root Cause Analysis, Inventory Discrepancies, Shipment Discrepancies, Warehouse Management Systems, Retrieval-Augmented Generation, Graph Causality Engine, Supply Chain Intelligence, LLaMA, LoRA Fine-Tuning

Abstract

While there are many differences between inventory and shipments in Warehouse Management Systems (WMS), the most serious ones come with a high price tag in global supply chains; and current diagnostic approaches are fragmented, reactive, and too much dependent on domain-expert intervention. The authors propose the LLM-Powered Root Cause Intelligence (LRCI) framework, which is a novel architecture that combines a fine-tuned Large Language Model (LLM) with Retrieval-Augmented Generation (RAG), a Graph-based Causality Engine (GCE), and a structured anomaly detection pipeline for real-time, interpretable root cause analysis of discrepancies in WMS. The framework is designed to take multi-source operational data, such as WMS event logs, ERP transaction logs, IoT sensor data, and carrier API feeds, and run it through four stages that result in a natural-language diagnosis of the root cause and suggestions for remediation. The proposed LRCI framework is evaluated on an industrial data set of 18,500 discrepancy data records from six distribution centres, which, on average, has an F1-score of 94.1%, a recall rate of 95.2%, a precision rate of 93.1% and a mean root cause identification latency of 1.8 seconds. The results are consistent improvements of 4.6% to 23.4% when compared to baseline models based on rules or machine learning or zero-shot LLM models, showing that the framework is viable to be used in real-time in high volume warehouse environments.

References

[1] M. A. Waller and S. E. Fawcett, "Data science, predictive analytics, and big data: A revolution that will transform supply chain design and management," J. Bus. Logist., vol. 34, no. 2, pp. 77–84, 2022.

[2] W. X. Zhao et al., "A survey of large language models," arXiv preprint, arXiv:2303.18223, 2023.

[3] H. Hegselmann et al., "TabLLM: Few-shot classification of tabular data with large language models," in Proc. Int. Conf. Artificial Intelligence and Statistics (AISTATS), PMLR, pp. 5549–5581, 2023.

[4] P. Lewis et al., "Retrieval-augmented generation for knowledge-intensive NLP tasks," in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 33, pp. 9459–9474, 2022.

[5] T. Peng, Y. Liu, and H. Wang, "Inventory anomaly detection in warehouse management systems using gradient-boosted tree ensembles," IEEE Trans. Ind. Informat., vol. 18, no. 9, pp. 6241–6252, 2022.

[6] J. Zhang, R. Li, and K. Wang, "Deep autoencoder-based shipment routing error detection in freight management systems," IEEE Access, vol. 10, pp. 78432–78445, 2022.

[7] X. Chen, M. Huang, and Q. Zhou, "BERT-based classification of logistics incident reports for supply chain root cause analysis," in Proc. IEEE Int. Conf. Big Data, pp. 1853–1861, 2023.

[8] S. Rao and V. Krishnamurthy, "Supply chain risk identification via BERT embeddings and knowledge graph integration," in Proc. ACM SIGKDD Int. Conf. Knowl. Discovery Data Mining, pp. 4201–4210, 2023.

[9] P. Narayan et al., "Can large language models reason over enterprise process data?" Proc. VLDB Endowment, vol. 16, no. 11, pp. 3276–3289, 2023.

[10] J. Shi, W. Huang, and T. Liu, "Retrieval-augmented generation for industrial knowledge-intensive tasks: A domain adaptation perspective," IEEE Trans. Ind. Informat., vol. 20, no. 3, pp. 3812–3824, 2024.

[11] K. Papadopoulos, A. Dimara, and C. N. Anagnostopoulos, "Graph-based causal analysis for supply chain disruption tracing," IEEE Trans. Autom. Sci. Eng., vol. 21, no. 1, pp. 542–556, 2024.

[12] R. Rombach, M. Weber, and T. Krauth, "Hybrid neuro-symbolic frameworks for manufacturing event log reasoning," in Proc. IEEE Int. Conf. Automation Science and Engineering (CASE), pp. 1012–1018, 2023.

[13] S. Tiwari, P. Kumar, and A. Gunasekaran, "Artificial intelligence for inventory management: A systematic review and future directions," Int. J. Prod. Econ., vol. 258, Art. no. 108812, 2023.

[14] F. Jiang, B. Chen, and G. Wang, "Towards intelligent warehouse management: A review of AI-driven approaches and open challenges," Comput. Ind. Eng., vol. 178, Art. no. 109093, 2023.

[15] Q. He, Z. Liu, and Y. Sun, "Semantic log parsing with domain-adapted template miners for industrial event analysis," IEEE Trans. Serv. Comput., vol. 17, no. 2, pp. 621–634, 2024.