Transform Invoice Processing and Drive Innovation with Amazon Bedrock: A Holcim Success Story

Organizations across industries face the challenge of processing a substantial volume of invoices while maintaining accuracy and compliance. Holcim, a global leader in innovative and sustainable building solutions, tackled this challenge by developing an innovative solution using Amazon Bedrock that transformed their invoice processing operations.

 The Challenge

Holcim’s operations in one European country managed a complex invoice processing workflow through a single email inbox that received approximately 2,000 monthly invoices along with numerous vendor communications. A dedicated full-time staff member manually processed this constant stream of emails, performing critical tasks including:

• Sorting incoming emails between actual invoices and other vendor communications
• Classifying invoices by company code (legal entity)
• Routing vendor communications to appropriate service desk teams
• Reviewing invoice details to ensure all required information is included
• Managing multi-topic emails that require splitting documents or combining related materials
• Identifying and removing non-relevant communications
• Handling documents in different languages

This labor-intensive process not only consumed significant staff resources but also created potential bottlenecks in processing time and introduced human error risks. With thousands of emails to process monthly, the manual nature of the work made it challenging to maintain consistent processing times and accuracy levels.

The Solution: Omnivore

Holcim developed “Omnivore”, a sophisticated invoice processing solution leveraging generative AI. When an email is received, Omnivore analyzes the content extracting key information regardless of format, layout, or language. Using Anthropic Claude’s natural language processing capabilities, the solution automatically distinguishes between invoice emails and general vendor communications, sorts invoices by company code, and intelligently routes communications to appropriate service desk teams. The system handles different languages and identifies when emails contain more than one topic and needs to be separated or combined.

Anthropic Claude is a large language model (LLM) trained on vast amounts of data, including business documents, enabling it to understand context and patterns without requiring specific training for invoice processing. Unlike traditional Optical Character Recognition, or template-based solutions, Claude leverages its foundation model capabilities to identify invoice elements through natural language understanding, making it adaptable to any invoice format. This generative AI approach allows the system to handle new suppliers or invoice layouts without requiring additional training or template creation.

This AI-powered processing empowers automatic classification and routing of invoices, significantly reducing the need for manual review of 10% of cases.

Technical Architecture

The solution leverages a serverless architecture powered by AWS, with Amazon Bedrock at its core. Amazon Bedrock is a fully managed service that provides access to your choice of high-performing foundation models through a single API, making it easier to build generative AI applications. In Holcim’s case, it allows integration with Anthropic Claude for invoice processing.

The orchestration is made with two AWS Lambda functions. The first handles email polling and configuration management. The second Lambda function drives the core processing pipeline, preparing the invoice content for AI analysis.

Configuration tables maintain processing rules and track performance statistics, enabling continuous optimization of the workflow. The entire system operates as an event-driven, serverless architecture, automatically scaling with invoice volume while maintaining consistent processing accuracy.

Development Approach and Implementation

Delivering the solution required cross-functional collaboration. While the Holcim IT Infrastructure & Operations team handled architecture decisions and the development partner led technical implementation, Holcim business units provided crucial input from their daily invoice processing experience. A dedicated product owner maintained strategic direction throughout the project.

The entire project completed in four months using an agile methodology. Teams met weekly for sprint planning, ensuring continuous progress through regular collaboration between IT, development, and business teams. As development progressed, the team began testing by manually forwarding emails to simulate real workflow conditions, allowing them to validate processing accuracy and system responses.

The solution then entered production through a careful deployment strategy. The system initially ran live under human supervision, operating in parallel with existing processes to confirm reliability. The team continues to improve the solution based on user feedback, implementing minor feature enhancements as needed.

This structured approach allowed a smooth transition from development to production while maintaining processing accuracy and business continuity.

Why AWS?

Holcim strategically chose to build its solution on AWS because of:

1. Model Flexibility Through Amazon Bedrock

Amazon Bedrock provides a crucial advantage by offering a model-agnostic approach to AI implementation. While the solution currently utilizes Anthropic Claude, Bedrock’s unified API allows Holcim to easily switch between different foundation models or use different models as needed, allowing solution flexibility as AI technology evolves. This flexibility allows Holcim to leverage the best-performing model for their specific use case without architectural changes.

2. Enterprise-Grade Security and Privacy

The Holcim global security team thoroughly vetted and approved Amazon Bedrock for enterprise use. All processing occurs within the organization’s Virtual Private Cloud (VPC), ensuring data remains within a controlled environment. AWS Identity and Access Management (IAM) roles provide granular control over access rights, simplifying security management while maintaining strict security standards.

3. In-House AWS Expertise

Holcim’s IT team already had experience in AWS development, allowing them to leverage their existing skills and experience. This meant accelerated development cycles and a productive use of internal resources, without the need for extensive additional training or external consultants.

4. Serverless Architecture Benefits

The solution’s serverless architecture, built on AWS Lambda, provides multiple advantages:

• Pay-as-you-go pricing that aligns costs directly with actual usage
• Automatic scaling to handle variable invoice volumes
• Minimal operational overhead
• High availability and reliability without manual intervention

Benefits and Results

The Omnivore implementation transformed Holcim’s invoice processing operations. Data from April 2025 demonstrates the solution’s impact:

• Average number of emails processed per month: 3,906
• Automated processing: 95% of volume
• Manual processing: ~300 emails (remaining 10%)
• Processing cost: reduction in communication processing costs

“Omnivore has been a game-changer for us, helping remove 90% of the manual work in invoice processing while also improving timeliness and quality,” said Niklas Gruber, Digital Transformation Manager at Holcim. “With low processing costs per email and the ability to handle nearly 4,000 emails monthly, we’ve achieved both efficiency and cost-effectiveness. What’s most impressive is that our end users didn’t even notice the switch from manual to automated processing – a clear sign that the solution is working as intended.”

Future Scale and Impact

Following its successful deployment in one European country, Holcim plans to expand Omnivore to four additional countries, leveraging the serverless architecture’s ability to scale automatically and seamlessly. The implementation success has sparked additional innovation opportunities within the organization. The team identified further process automation opportunities, particularly in their SAP-based workflows, streamlining touchpoints between teams using similar LLM technology.

The project demonstrates that LLM-based solutions are implemented cost-effectively and rapidly. The solution’s impact extends beyond operational efficiency, supporting compliance and risk mitigation through automated processes that help maintain regulatory requirements and industry standards. The reduction in manual processing delivers labor and cost savings, as well as allows staff to focus on strategic initiatives.

Conclusion

Holcim’s success with Omnivore demonstrates how organizations leverage AWS technology to solve complex operational challenges. The combination of Amazon Bedrock advanced AI capabilities with AWS’s broad and deep set of cloud capabilities enable Holcim to achieve a level of automation and productivity that sets a new standard for invoice processing in the industry.

As organizations continue to seek ways to optimize their operations, Holcim’s implementation serves as a compelling example of how innovative cloud-based solutions drive significant business value. The solution not only addresses immediate operational needs but also provides a foundation for continued digital transformation and success.

Ready to explore how generative AI transforms business operations? Learn more on how to innovate faster with generative AI on AWS and start building your own AI-powered solutions today.