Enterprise Resource Planning (ERP) systems are a critical operational backbone for business-to-business (B2B) supply chain operations in regulated industries like defense, pharmaceuticals, and healthcare. Although ERP systems collect vast amounts of data related to transactions and operations, many aspects of purchasing, logistics and procurement decisions such as assessing the risk of suppliers, approving purchase orders and validating compliance are still done manually. Manual processes result in inefficiencies and increase the risk of operational errors. Large Language Models (LLM) allow users to interact with enterprise data using natural language; however, LLM have seen little use in regulated environments due to concerns regarding hallucinations, policy violations, and lack of auditability. A retrieval augmented ERP Assistant was designed to enhance decision-making for B2B supply chain operations in mission-critical environments. The proposed architecture consists of an ERP system or CRM platform providing contextual retrieval of data, along with an external data source containing regulatory and sanctions information, to produce responses that enforce enterprise policies, approval restrictions and regulatory requirements through a Compliance-Aware Generation Layer. A benchmark dataset of procurement and fulfillment question answering tasks was created to measure the effectiveness of the assistant and its performance was compared to those of experienced procurement professionals within a simulated enterprise workflow. Results demonstrated that retrieval-augmented, compliance-aware assistants can significantly enhance decision efficiency, consistency and trust while maintaining the necessary auditability and governance required for regulated B2B supply chains.