Automate Procurement Workflows with AI Agents using Amazon Bedrock AgentCore

Procurement teams increasing struggle with managing vendor selection and Request for Quotation (RFQ) processes effectively. Buyers spend significant time navigating multiple Enterprise Resource Planning systems and external data sources. They must consolidate supplier information, validate compliance requirements, and compare costs across vendors to identify the optimal option. This fragmentation causes delays, manual errors, and missed opportunities to optimize costs, delivery outcomes, and supplier diversity. Organizations need to make procurement faster, more accurate, and policy compliant without adding complexity to their workflows.

In this blog, we will discuss how Amazon Bedrock AgentCore and Strands Agents address these challenges by developing the Smart Procurement Assistant, a custom-built, agentic AI-powered chat assistant that automates procurement workflows to reduce cycle times, enhance real-time compliance validation, and enable strategic focus on supplier management.

Intelligent Orchestration with Amazon Bedrock AgentCore and Strands Agents

Addressing procurement challenges requires intelligent orchestration across diverse data sources and complex business rules. Amazon Bedrock AgentCore Runtime, combined with the Strands Agents SDK, provides a foundation for building sophisticated agentic systems that can deploy and manage AI agents with specialized tool capabilities. These agents handle complex, multi-step procurement tasks autonomously by leveraging focused tools for distinct functions including compliance validation, supplier recommendation, financial analysis, and RFQ management. The system uses intelligent reasoning to select and execute the appropriate tools based on user requests.

Amazon Bedrock AgentCore provides a fully managed infrastructure that eliminates technical complexities such as session handling and state management, significantly accelerating development while improving both task success rates and response accuracy.

Security and privacy are important for enterprises embracing generative AI, especially when handling proprietary procurement data. Customers rightly worry about whether their sensitive data or prompts might be used to train foundation models or be exposed publicly. AWS is designed to address these critical concerns through a secure hosting environment with comprehensive protection measures, including data encryption in transit and at rest and compliance with security standards.

Amazon Bedrock does not use customer data and prompts to train or improve its foundation models, designed to maintain data confidentiality. This secure framework allows organizations to confidently adopt generative AI for procurement modernization while mitigating risks for their intellectual property or vendor relationships.

Solution Overview

The Smart Procurement Assistant (SPA) uses advanced conversational AI and an intelligent tool-augmented agent architecture to improve procurement workflows by integrating SAP enterprise data with external vendor compliance ratings.

This solution architecture applies to any enterprise-grade procurement application across industries and platforms including Oracle E-Business Suite, Oracle JD Edwards, Oracle PeopleSoft, Oracle Fusion Cloud ERP, Salesforce, Snowflake and other packaged applications. While we illustrate SPA with SAP S/4HANA running on Amazon EC2, the architectural patterns and benefits apply broadly across hybrid and multi-cloud procurement environments.

Users interact naturally through a React-based conversational chat interface to access comprehensive supplier, material, financial, and compliance information. SPA enables procurement professionals to quickly identify the best vendors and directly create Requests for Quotation (RFQs) within the same interface, eliminating fragmented system navigation and accelerating sourcing cycles.

The Strands Agents SDK provides the intelligent orchestration framework that powers SPA’s conversational capabilities. It implements a unified AI agent that handles intelligent tool selection based on user intent, manages conversation flow, and delivers real-time streaming responses token-by-token over WebSocket connections for immediate user feedback.

The solution’s intelligence is driven by an agentic design equipped with eight specialized tools that work together through intelligent orchestration via the Strands Agents SDK and Model Context Protocol (MCP):

1. Schema Lookup Tool – Retrieves metadata from Amazon Bedrock Knowledge Base about database structures, providing intelligent access to data catalog information and enabling dynamic query generation across SAP and compliance databases
2. Custom Query Tool – Runs custom SQL queries against procurement data via Amazon Athena, allowing flexible data exploration through natural language by translating user requests into optimized SQL queries
3. Financial Analysis Tool – Pulls cost analysis and spending patterns from supplier invoices, querying purchase orders and financial data to assess cost and spending metrics across vendors
4. Quality Metrics Tool – Examines delivery reliability and material quality scores, evaluating material attributes and supplier delivery performance to provide reliability insights
5. Compliance Validation Tool – Verifies regulatory certifications including REACH (Registration, Evaluation, Authorization, and Restriction of Chemicals), RoHS (Restriction of Hazardous Substances), CMRT (Conflict Minerals Reporting Template), and RBA (Responsible Business Alliance), enriching vendor profiles with external compliance certifications
6. Validate RFQ Data Tool – Parses and confirms RFQ parameters from conversational input to ensure data accuracy before submission
7. RFQ Creation Tool – An MCP client which invokes SAP API to create RFQs in SAP systems through OData integration, automating the generation of RFQs directly in SAP based on user inputs and streamlining the purchasing process
8. Data Visualization Tool (Execute Python) – Runs visualization code in an isolated sandbox for generating charts, executing Python code in a sandboxed environment to create charts and graphs for supplier comparison and trend analysis

Figure 1: Smart Procurement Assistant Architecture

Managed Runtime (Amazon Bedrock AgentCore)

Amazon Bedrock AgentCore Runtime provides the production-grade infrastructure for hosting and scaling the Strands-powered agent:

• Persistent Memory: Stores conversation history per user session, enabling context-aware interactions across multiple exchanges
• Containerized Deployment: Runs agent workloads on Amazon ECS with AWS Fargate for automatic scaling and high availability
• Code Interpreter: Provides sandboxed Python execution environment for secure data visualization
• AgentCore Gateway: Provides MCP-based bridge for SAP OData API integration through AWS Lambda

Data Foundation

AWS Glue Data Catalog serves as the central metadata repository, enabling Amazon Athena to query structured procurement data stored in Amazon Simple Storage Service (Amazon S3). Pre-configured AWS Glue views deliver ready-to-use analytics for financial trends, quality benchmarks, and compliance status. Amazon OpenSearch Service powers document search functionality across procurement records.

Enterprise System Connectivity

SAP integration occurs through Open Data Protocol, with AWS Secrets Manager securely managing authentication credentials. The architecture’s flexibility supports connections to additional ERP platforms like Oracle and Workday.

This design delivers conversational access to supplier analytics, automated compliance verification, intelligent sourcing recommendations, and direct RFQ creation—combining Strands’ intelligent orchestration with Amazon Bedrock AgentCore managed infrastructure.

Security Architecture

The solution implements defense-in-depth security. AWS Web Application Firewall (WAF) filters malicious web traffic at the edge, while Amazon Cognito provides authentication via JWT tokens and enforces role-based access controls. Amazon API Gateway manages secure WebSocket connections. Data encryption protects information both in transit and at rest, with comprehensive audit logging tracking all system interactions.

Sample Code Repository

For the sample code and demonstration on the solution discussed in this post, refer to the accompanying GitHub repository.

Business Outcomes

Implementing the Smart Procurement Assistant delivers business value across multiple dimensions. The solution accelerates decision-making by reducing procurement cycle times through automated supplier evaluation. This automation translates directly into cost-savings through streamlined operations, data-driven supplier selection, and enhanced spend visibility.

Compliance improves as every supplier recommendation is automatically validated against organizational policies and regulatory requirements. The natural language interface eliminates the need for procurement professionals to navigate multiple domain-specific ERP systems, boosting user productivity.

This comprehensive approach transforms procurement from a time-consuming administrative function into a strategic advantage.

Conclusion

By combining Amazon Bedrock, Amazon Bedrock AgentCore Runtime, Strands Agents SDK, Knowledge Bases, and other AWS services, the Smart Procurement Assistant (SPA) transforms traditionally manual, time-consuming workflows into automated and intelligent processes. This AI solution accelerates RFQ creation, vendor analysis, compliance verification, and real-time analytics, enabling procurement teams to redirect effort from repetitive tasks toward strategic sourcing decisions. SPA’s agentic AI approach modernizes procurement operations and fast-tracks intelligent automation.

To get started, contact your AWS representative and refer to the following resources for additional guidance:

• Amazon Bedrock AgentCore Documentation
• Model Context Protocol Documentation
• Strands Agent Documentation