Oldcastle Boosts Efficiency with Amazon Bedrock for Document Processing

Oldcastle APG, a global architectural products manufacturer, faced challenges processing 100,000-300,000 proof of delivery (POD) documents monthly. Their existing OCR system achieved only 30-40% accuracy, leading to significant manual effort and inefficiencies. To address this, they partnered with AWS to implement a solution leveraging Amazon Bedrock and Amazon Textract. The new system uses Amazon SES to receive ship tickets, Amazon S3 for storage, and an event-driven architecture for processing. Amazon Textract analyzes the PDFs, handling rotated pages and variable formatting, and extracts key data. A Lambda function processes the output, resolving rotation issues and generating markdown. This markdown is then fed into Amazon Bedrock, which extracts key values. Finally, the results are stored in Amazon RDS for PostgreSQL. This solution dramatically improved accuracy, reducing manual processing from 4-5 hours daily per dispatcher to near zero. Oldcastle achieved significant productivity gains, improved data reliability, and real-time visibility into PODs. The cost per page is less than $0.04. The success led Oldcastle to explore expanding the application to other document types, showcasing the scalability and cost-effectiveness of the AI-powered solution. This case study highlights how Amazon Bedrock, combined with other AWS services, can transform document processing workflows, significantly improving efficiency and accuracy.

(Source: https://aws.amazon.com/blogs/machine-learning/oldcastle-accelerates-document-processing-with-amazon-bedrock/)