Ellby accelerates invoice automation using GenAI on Amazon Bedrock

As customer expectations rose for faster onboarding and greater efficiency, Ellby identified an opportunity to improve how it delivered accounts payable automation. Many customers still relied on template-based optical character recognition (OCR) systems. These systems required significant effort to configure and maintain, especially as vendor formats changed frequently.

Ellby aimed to create a more scalable experience by reducing the time spent onboarding new customers and updating invoice templates. For each customer, the team could take 24 hours or more setting up profiles for high-volume vendors, and still had to manually validate 40 percent of invoices. On top of this, more than 300 hours each month was required to maintain the system.

To improve automation outcomes, Ellby turned to Amazon Bedrock. The fully managed service provides access to leading foundation models and tools to build and scale generative AI applications.

Before selecting AWS, Ellby tested multiple cloud-based platforms to evaluate real-world performance. The results showed that AWS delivered higher invoice automation rates and greater field-level accuracy.

Using Amazon Bedrock, the team embedded generative AI into EllbyExtract to interpret diverse invoice formats with intelligent prompts. This eliminated the need for static templates tied to individual vendor layouts. The shift increased invoice automation from under 60 percent to over 94 percent, and reduced customer onboarding time by over 55 percent.

The AI-powered workflow significantly reduced time spent on manual template configuration and maintenance, freeing Ellby's team to focus on higher-value projects. Sensitive financial data is processed within AWS infrastructure hosted in Australia, supporting the company's data residency requirements.

Ellby selected Claude by Anthropic in Amazon Bedrock after evaluating multiple foundation models for speed, accuracy, and cost-effectiveness. Specifically, the team deployed Claude Sonnet and Haiku in EllbyExtract. By designing targeted prompts, EllbyExtract can obtain and validate invoice data across varied layouts. This removes the need to rebuild configurations when vendor formats change.

To maintain accuracy, EllbyExtract uses their proprietary rules-based engine that evaluates responses by Claude on Amazon Bedrock. When data confidence is low, EllbyExtract flags invoices for human review. As a result, the platform can adapt dynamically to changing document structures, while reducing errors and the need for manual intervention. Amazon Bedrock's fully managed, pay-as-you-go model also supports cost-effective operations without increasing internal overhead.

"Using generative AI on Amazon Bedrock, we've made our automation smarter and easier to scale. It's helping us move faster and build a foundation for broader innovation that benefits our customers," said Steven Chenery, Chief Executive Officer, Ellby.

Building on the success of its invoice automation, Ellby aims to extend its use of Amazon Bedrock to extract detailed line item data from invoices. The team is also exploring vision-based AI to better process complex formats, including handwritten documents. These enhancements will deliver greater automation depth and accuracy for customers.

Beyond accounts payable, the digital solutions company is developing generative AI-powered solutions for other document-intensive workflows. These include expense claims processing, contract summarisation, and clinical coding for healthcare. By extending its generative AI strategy on Amazon Bedrock, Ellby aims to streamline operations across industries, reduce manual work, and help customers shift their focus to more strategic tasks.