Agentic Invoice Processing at Volkswagen Financial Services

Executive Summary

Volkswagen Financial Services (VWFS) handles a continuous flow of supplier invoices connected to its claims operations. One category is considerably more complicated than the rest: each invoice has to be matched against purchase orders and validated against several price lists, external agreements, and service records before it can be approved. Handled manually, a single invoice in this category took a claim specialist up to 20 minutes.

VWFS and Awiant have partnered to build an AI Center of Excellence (CoE) and add the engineering capacity to deliver agentic AI solutions. That work follows a Lighthouse approach: take a concrete, business-critical problem and prove out AI on it all the way from idea into production, rather than leaving a proof of concept on the shelf. Invoice processing was chosen as the first Lighthouse case, and the first version has been in production since April 2026.

Handling time for these invoices went from up to 20 minutes to about five minutes each, based on the customer's own estimates. For VWFS, the gain is capacity rather than headcount: the same specialists can process a substantially larger volume of these invoices and put more of their time into the cases that need human judgment, letting the operation grow with rising volumes instead of scaling the team to match. The case has also become the starting point for giving the agents more autonomy and automating more of the same area, step by step.

About the Customer

Volkswagen Financial Services provides financing, leasing, insurance, and fleet services for the Volkswagen Group's brands. Its claims operations process high volumes of supplier invoices, where accuracy and traceability are as important as speed.

Challenge

The invoice review process was manual, fragmented, and time-consuming. For the most complicated invoice category, a claim specialist had to:

• Match each invoice to the correct purchase order
• Validate line items against multiple price lists
• Run a technical validation in a separate internal system
• Perform additional price list checks and adjust supporting documentation
• Approve the invoice or return it to the supplier

Every step meant switching between systems and documents. The process was slow, hard to standardize, and carried a high risk of errors — and it tied up specialist time that was needed elsewhere.

Solution

Awiant built an agentic AI solution that brings the whole review into one interface, on top of the claims management system VWFS already uses. At its core is an agent loop: the agent analyzes the matched invoice and purchase order, decides which validation tool to use next, executes it, and reasons over the result before taking its next step.

The agents work against the same sources a specialist would use:

• Customer price lists
• Reseller price lists
• External agreements
• Service information
• Customer history

When the review is complete, the agent presents a recommendation to accept or reject the invoice, with its reasoning. The claim specialist stays in control of the decision: invoices are approved by a person, and only invoices meeting predefined criteria will move toward autonomous handling as the solution matures.

The solution runs on Anthropic's Claude models on Amazon Bedrock and was built end to end with Claude Code.

How It Works

1. Match — An incoming invoice is matched against its purchase order.
2. Analyze — The agent analyzes the matched pair and identifies what needs validating.
3. Select and execute tools — The agent chooses among its validation tools (price lists, agreements, service information, customer history) and runs them, reasoning iteratively over each result.
4. Recommend — The agent consolidates its findings into an accept/reject recommendation with supporting evidence.
5. Decide — The claim specialist reviews the recommendation and approves the invoice or returns it.

Working as One Team

The technology was only part of what made this work. VWFS and Awiant ran the project as one team, with close dialogue and short decision paths, working through problems as they came up. Because agentic AI is a fast-moving field, much of the effort went into getting the organization and VWFS's existing systems in sync — as much as into building the agents themselves.

"Agentic AI is a new field and it's moving fast, so we solved the challenges together — without ego, and making decisions as we went. The hard part isn't the new technology; it's getting the organization and our existing systems in sync."

Overseeing, adjusting, and teaching the agents is itself a new way of working for the team — one that frees up specialist time for the tasks that need human judgment.

Result

• Handling time from up to 20 minutes to about 5 minutes per invoice (customer estimate) for the most complicated invoice category
• Room to handle more — the same team can process a substantially larger volume of complex invoices, letting the operation absorb rising volumes without adding headcount
• Manual process steps replaced by agent-executed validations, with the specialist making the final decision
• One interface instead of a fragmented, multi-system workflow
• Lower risk of errors through consistent, source-backed validation
• A platform for what's next — with the agent platform and underlying data in place, VWFS can step by step give the agents more autonomy and automate further processes in the same area, with specialists overseeing the orchestration and focusing on data quality

"Monitoring, adjusting, and teaching the agents new things is a completely new way of working — it's more enjoyable, it frees up time for other work, and we're all looking forward to what comes next."

Technologies Used

The solution runs on Anthropic's Claude models on Amazon Bedrock and was built end to end with Claude Code.

Foundation models

Anthropic Claude models, served via Amazon Bedrock

Agent runtime

Amazon Bedrock AgentCore

Agent pattern

Agentic tool use — an agent loop with iterative reasoning

Development

Built end to end with Claude Code

Integration

The customer's existing claims management system