Services provided
- Artificial Intelligence
- Cloud Engineering
- Serverless Architecture
- Intelligent Document Processing
Platforms used
- AWS (Amazon Bedrock, Amazon Textract,
AWS Lambda, Amazon SQS, Amazon S3,
Amazon DynamoDB, Amazon SES, Amazon CloudWatch, AWS CDK)
Engagement length
- 12 weeks
Other stats
- ~8,300 purchase orders processed
per month - ~35,000 serverless workflow events
processed monthly - 5 AWS Lambda functions powering the processing pipeline
- High-confidence purchase orders processed
in under 1 minute - Automated extraction, validation, duplicate detection, and confidence-based routing
Background on the customer
Our customer is one of Australia’s largest trade, plumbing, and building supplies organisations, supporting thousands of business customers across a nationwide branch network.
A significant proportion of B2B customers submit purchase orders directly from their own procurement systems via email. As order volumes continued to grow, the organisation needed a more scalable and reliable approach to processing purchase orders while maintaining customer experience and operational efficiency.
Challenge
Purchase orders arrived in a wide range of customer-specific formats, layouts, and document structures. Processing these documents required significant manual intervention and relied heavily on template-based extraction approaches that became increasingly difficult to maintain as customer volumes grew.
The existing process created several challenges:
- High levels of manual handling and exception management
- Significant effort maintaining customer-specific document templates
- Delays in order processing and validation
- Limited auditability across document processing workflows
- Difficulty scaling operations as customer and order volumes increased
- Inconsistent handling of non-standard purchase order formats
- Operational dependency on specialist knowledge and manual review
The organisation needed a solution capable of automatically extracting, validating, and routing purchase orders while preserving existing customer ordering behaviour.
Stack Highlights
The Approach
Codex designed and delivered a serverless AI-powered document intelligence platform on AWS to automate purchase-order ingestion, extraction, validation, and workflow routing.
Our approach included:
1. Designing an event-driven architecture using Amazon SES, S3, SQS, Lambda, and DynamoDB
2. Automating email ingestion and PDF document processing workflows
3. Leveraging Amazon Textract for OCR and document digitisation
4. Using Amazon Bedrock and large language models to perform schema-guided purchase-order extraction
5. Implementing confidence scoring and per-field validation to improve extraction quality
6. Building duplicate detection and validation capabilities before downstream order processing
7. Creating confidence-based routing to enable touchless processing for high-confidence orders while directing exceptions for human review
8. Implementing audit logging, lifecycle tracking, and operational monitoring throughout the workflow
9. Delivering the platform through Infrastructure as Code using AWS CDK
This approach enabled intelligent processing of highly variable purchase-order formats without relying on brittle template-based extraction techniques.
Technical Outputs
Codex delivered a production-grade intelligent document processing platform capable of automating purchase-order ingestion at scale.
Key outputs:
The resulting platform provides a scalable foundation for intelligent document processing while reducing operational effort and improving processing consistency.
QA Platform Highlights
AI-Powered Purchase Order Extraction
Combines Amazon Textract and Amazon Bedrock to extract and validate structured purchase-order information from highly variable customer documents.
Confidence-Based Automation
Purchase orders are automatically routed based on extraction confidence, enabling touchless processing for high-confidence documents while directing exceptions for review.
Event-Driven Serverless Architecture
A fully serverless design built on Lambda, SQS, and S3 scales automatically with document volumes while minimising infrastructure management overhead.
Complete Auditability and Traceability
Lifecycle tracking, structured logging, confidence scoring, and audit artefacts provide visibility across every stage of document processing.
Built for Document Variability
Designed to support diverse purchase-order layouts, terminology, pricing structures, delivery requirements, and customer-specific document formats without requiring ongoing template maintenance.
Business and Commercial Outcomes
The intelligent document processing platform delivered measurable improvements in operational efficiency, scalability, and customer experience.
- Automated extraction, validation, and routing of purchase orders from customer emails
- Reduced manual handling and template-maintenance effort across order-processing teams
- Enabled high-confidence purchase orders to be processed in under one minute from receipt to downstream validation
- Improved auditability through lifecycle tracking, confidence scoring, and structured workflow state management
- Increased scalability by supporting growing purchase-order volumes without proportional operational growth
- Improved operational visibility through monitoring, alerting, and workflow analytics
- Preserved customers’ existing ordering behaviour, avoiding process change requirements for trade customers
- Established a reusable AI-powered document processing foundation for future business workflows
The platform now enables the organisation to process purchase orders more efficiently, improve operational scalability, and deliver a faster, more consistent ordering experience for customers.
Talk to Us
We would love the opportunity to connect and understand more about the problems you are trying to solve.
Adrian Cambpell
Associate Partner, AI
Martin Campbell
Managing Partner
Get in touch to coordinate a meeting with one of our technical experts.
Australia: +61 7 3132 3002.
