How Rivian Accelerated Finance Operations with AI Agents on Amazon Bedrock

Co-authored by Arunkumar Malli (Rivian)

Rivian optimized its month-end financial close processes to better support rapid growth by deploying intelligent ERP AI agents built on Amazon Bedrock AgentCore, eliminating over 15 days of manual work per cycle. If your finance teams are struggling with manual ERP workflows, lengthy close cycles, or scaling challenges as your business grows, Rivian’s journey offers a proven blueprint for business process transformation.

In this blog, we share how Rivian’s SAP finance operations scaled beyond limitations, describe the technical architecture using Amazon Bedrock AgentCore with leading LLM models, and share measurable results and lessons learned your organization can apply.

The Challenge: Scaling Finance for a New Product Launch

Rivian, headquartered in Irvine, California, is a leader in electric vehicle innovation, manufacturing its R2 SUV, R1T pickup, R1S SUV, and Electric Delivery Van (EDV) in Normal, Illinois. With a global team of approximately 15,000 employees, the company is managing exponential operational growth driven by strategic initiatives, including the R2 vehicle line and a joint venture with Volkswagen to advance electric vehicle software.

As Rivian accelerated operations to support manufacturing growth, finance systems, while functional and robust, faced the opportunity to scale even further. Rivian sought to automate the manual elements of its monthly and quarterly closing documents, resource forecasting, and supplier onboarding. By optimizing these existing SAP ERP operations, Rivian aimed to accelerate performance analysis and enable more immediate incorporation of insights into its planning cycles. To support the significant growth driven by anticipated vehicle demand, Rivian proactively adopted GenAI to further unlock productivity and streamline our processes.

Why GenAI Offered a Solution

Traditional automation approaches such as hardcoded scripts couldn’t solve Rivian’s complex finance workflows because the business logic was too dynamic and context-dependent. Purchase order accruals required interpreting detailed Standard Operating Procedures (SOPs) with conditional rules based on dollar thresholds, PO characteristics, delivery timelines, and GAAP requirements. Encoding these procedures into traditional code could require thousands of lines of if-then-else logic that would become brittle and require developer intervention every time business rules changed.

Generative AI agents offered a fundamentally different approach: the ability to read and interpret SOPs in natural language, make context-aware decisions based on each unique situation, and adapt to changing procedures without reprogramming. This orchestration capability—to understand complex instructions and dynamically determine the right workflow path—made GenAI the technology Rivian needed to scale their finance operations.

Amazon Bedrock AgentCore AI agents can automate and modernize manual processes.

The Rise of Rivian’s Agentic Strategy

Rivian reached out to AWS with several workflows as candidates for improvement. One example was purchase order accruals for custom tooling purchases—a portfolio of significant annual value across a large volume of active purchase orders. This labor-intensive process involved extracting data from ERP Systems building complex spreadsheet models, reaching out to purchase order owners via email to confirm delivery dates, and calculating time-proportional accruals to support GAAP compliance.

The business opportunity was multifaceted. Custom tooling for automotive manufacturing, such as stamping dies and injection molds, involves development periods spanning 12–24 months, but invoices often don’t arrive until 18+ months after the purchase order is created. Under GAAP, expenses must be accrued gradually over the tool development period rather than recognized all at once when invoices arrive. This meant finance teams needed to track hundreds of purchase orders simultaneously, validate delivery schedules, calculate proportional accruals, and update SAP—all while maintaining complete audit trails for external auditors.

This complexity was present not just in one workload, but across operational domains including finance, supply chain, and resource planning. An accelerated proof-of-concept was successfully completed in five weeks. Based on the promising results, Rivian fast-tracked the solution to production and began preparing a repeatable foundation for extending to other teams with minimal duplication.

The Strategic Turn to Generative AI

Rivian finance and technology teams collaborated with AWS specialists to develop the SAP Purchase Order Accrual Automation System built on Amazon Bedrock AgentCore. The system was designed around several core principles:

SOP-Driven Intelligence

Rather than hardcoding business logic into application code, the team documented all accrual processing procedures in detailed SOPs and stored them in Amazon Bedrock knowledge bases. The AI agent uses these SOPs as its “instruction manual,” retrieving relevant procedures based on the specific characteristics of each purchase order. This approach delivers business agility (finance managers update procedures by editing documents, not requesting code changes), audit transparency, domain expertise capture, and consistent regulatory compliance.

Materiality-Based Workflows

The system implements sophisticated decision logic based on materiality thresholds—a concept fundamental to financial accounting where different levels of scrutiny apply based on dollar amounts. The AI agent automatically routes purchase orders through different validation workflows based on dollar value, mirroring how experienced finance analysts prioritize their efforts.

Autonomous Agent Architecture

At the heart of the system is a Strands Agent built on Amazon Bedrock AgentCore, powered by a leading LLM model. The agent operates continuously, following a sophisticated orchestration pattern: detecting exceptions via SAP OData events, assembling context from DynamoDB and knowledge bases, making intelligent routing decisions, executing tools via a Model Context Protocol (MCP) server to interact with SAP S/4 HANA and send emails, calculating time-proportional accruals using straight-line methodology, creating parked journal entries for finance review, and managing approval workflows. Rather than following a rigid path, the agent adapts its approach based on each purchase order’s specific circumstances.

The MCP-based modularity means individual components—email provider, database, or SAP connector—can be swapped without changing core agent logic, ensuring the solution remains flexible as Rivian’s technology stack evolves.

Human-in-the-Loop Design

Despite its autonomous capabilities, the system maintains critical human oversight at key decision points. Finance managers retain final approval authority before any journal entries are posted to SAP. This satisfies both internal control requirements and external audit expectations while delivering massive time savings. It also enables continuous improvement—when managers review the agent’s work, they identify edge cases where procedures need refinement, creating a feedback loop that enhances accuracy over time.

Technical Architecture

Rivian’s system leverages a comprehensive suite of AWS services integrated with their SAP S/4 HANA system running in SAP HANA Enterprise Cloud (HEC):

• Amazon Bedrock AgentCore — Intelligent reasoning engine providing natural language understanding and complex decision-making capabilities for interpreting SOPs and orchestrating multi-agentic retrievals
• Amazon Bedrock Knowledge Bases — Stores SOPs and SAP API documentation, using Retrieval-Augmented Generation (RAG) to provide contextual decision support
• AWS Lambda — Handles scheduled tasks including polling SAP for exceptions and processing approval emails; scales automatically without infrastructure overhead
• Amazon DynamoDB — State management and complete audit trail with timestamps satisfying SOX compliance
• Model Context Protocol (MCP) Server — Custom tools for SAP REST API connectors, email via Amazon SES, and case management
• Amazon SES — Sends personalized emails and processes automated response parsing
• Identity & Security — AgentCore Identity for MCP servers, AWS IAM and Cognito for role-based access, Secrets Manager for credentials, CloudWatch for monitoring, and AWS CDK for infrastructure-as-code

The architecture follows AWS Well-Architected Framework principles with particular emphasis on security and operational excellence.

Business Results and Impact

The transformation has delivered measurable benefits across multiple dimensions:

Time Savings: The agent manages a high volume of purchase orders of significant total value annually, resulting in a substantial reduction of manual effort each year for Rivian’s accounting team. Finance analysts who previously spent a large portion of their time on accrual calculations now focus on higher-value activities: financial analysis, variance investigation, and strategic decision support.

Scalability for R2 Launch: With the R2 vehicle line now successfully launched, Rivian is experiencing the anticipated significant increase in custom tooling purchase orders. The AI-powered system can handle this exponential growth, directly supporting the company’s path to profitability.

Accuracy and Compliance: Consistent SOP application and complete audit trails in DynamoDB have enhanced accuracy and strengthened internal controls. External auditors have praised the transparency of the AI-driven process compared to previous spreadsheet based calculations.

Accelerated Financial Close: Compressed month-end timelines enable faster reporting to investors and management—particularly valuable for a public company operating in a fast-moving industry.

Cost Optimization: Serverless architecture and Amazon Bedrock AgentCore’s usage-based pricing align operational costs with actual workload.

The success of this implementation has established Amazon Bedrock AgentCore as a strategic platform for intelligent automation. The Rivian finance team is already planning additional use cases, each following the same SOP-driven, human-in-the-loop pattern.

Yogesh Yadav – Senior Director, Global Finance Systems, Rivian: “The partnership with AWS and the implementation of AI-powered automation for our purchase order accruals represents a fundamental shift in how we approach finance operations. We’ve not only eliminated weeks of manual work each month, but we’ve created a scalable foundation that will support our growth as we continue to scale R2 and expand production. This is the kind of productivity enhancement that directly impacts our bottom line.”

Conclusion

Rivian’s journey to AI-powered finance automation yielded key insights for other organizations considering similar transformations: start with well-defined processes for SOP-driven design, maintain human oversight, leverage consistent patterns to avoid “reinventing the wheel,” and partner with experts as needed.

The success extends beyond time savings—it represents a fundamental shift in how enterprise software can be built and maintained. By storing business logic in natural language SOPs rather than code, Rivian has empowered business users to own and evolve their processes while leveraging modern AI capabilities. Rivian’s partnership with AWS and adoption of Amazon Bedrock AgentCore demonstrates what’s possible when you apply intelligent automation to critical finance operations. The AI agents don’t just solve today’s problems—they scale as your business grows, delivering the productivity enhancements essential for sustainable growth in competitive markets.

About the authors

Sagar Tallapragada
specializes in guiding enterprises through digital transformation at AWS, automating business processes and modernizing existing systems. He drives the development of tools and solutions that help customers migrate, modernize, and automate with AI. Outside of work, he balances his passion for technology with cherished moments spent with family, a deep love for Formula 1, and a lifelong love for Sachin Tendulkar’s cricketing legacy.

Ramkrishna Borhade
is a Senior Solutions Architect at Amazon Web Services with 19+ years of SAP expertise, specializing in cloud migrations, S/4HANA conversions, and applying Generative AI to modernize legacy SAP landscapes. He guides enterprises through strategic cloud adoption, helping customers automate SAP business processes with Agentic AI. Outside of work, he enjoys spending time with family, watching the Premier League and supporting Manchester United.

Matt Reed
is a Senior Solutions Architect in automotive and manufacturing at AWS working directly with Rivian. Matt helps customers design and implement production-ready solutions across AWS services. He’s an avid skier and mountain biker in the Pacific Northwest, getting out doors as much as possible with his family and pets.o