To arrive at the optimal solution, we followed a design thinking-inspired approach, structured around the following key stages:

Discovery Workshop: A 2-week deep-dive exercise with key stakeholders from business, operations, and technology teams. The goal was to map out the current process and technical architecture, identify and prioritise pain points, define success metrics, explore solution alternatives, and evaluate their feasibility and desirability.

Proof of Concept (PoC): Over the following 2 weeks, we conducted a hands-on PoC to experiment with various libraries and AWS services for real-time document classification and custom entity recognition. Using mock datasets that mirrored real-world complexity uncovered during discovery, we validated the technical feasibility and selected the most suitable AWS components for the solution.

User Research & Testing: Through a series of collaborative workshops with nominated records managers and analysts, we:

• Examined their current workflows
• Co-designed the ideal processes and user tasks
• Prototyped and tested interfaces with real users
  This ensured that the final solution was intuitive, task-friendly, and aligned with user expectations.

Cloud Infrastructure Design: Worked closely with the client's Cloud Engineering team to co-design the AWS infrastructure in line with enterprise architecture standards. The design was reviewed and endorsed by relevant technical stakeholders to ensure smooth implementation.

Security & Compliance Controls: Conducted focused workshops with internal security and compliance teams to define the necessary controls for meeting APRA requirements related to the storage and handling of sensitive customer data (e.g., health, financial, and identity information). This exercise also enabled access to real, production-grade data to train our AI models for accurate document classification and entity extraction.

‍

The Fabric team designed and implemented a custom-built intelligent document processing (IDP) solution tailored to the client’s requirements. Key components of the solution include:

• Real-time email ingestion from multiple mailboxes, automated using Microsoft Power Automate.
• Training data generation through labelling jobs powered by AWS SageMaker Ground Truth.
• AI-based document classification, using models built on AWS Textract and Comprehend, trained on a curated dataset to handle diverse types of incoming digital correspondence—ranging from unstructured emails to customer-submitted forms and supporting documents (e.g., IDs, bills).
• Custom entity extraction, using a fine-tuned AWS Comprehend model to identify critical data points such as policy numbers, customer reference IDs, and advisor details—essential for processing customer requests.
• End-to-end orchestration logic, built with AWS Lambda and DynamoDB, to manage data ingestion, enrichment, status tracking, audit logging, and error handling.
  Lightweight user interface for records managers to monitor processing workflows and manually intervene when needed—enabling continuous AI model improvement via human feedback loops.
• Real-time system integration with the central document repository and downstream operational systems to ensure timely and seamless handling of customer requests.
• Infrastructure as Code, leveraging Terraform to deploy and manage environments within the client’s AWS VPC across development, testing, and support stages.‍
• MLOps capabilities, implemented to track model performance, ensure version control, and maintain compliance with governance protocols.

How We Deliver
We adopted agile principles and an iterative development approach to build the platform—ensuring continuous feedback, incremental improvements, and scalable AI model performance. Our delivery methodology was structured into the following phases:

• Inception Phase: A two-week collaborative engagement grounded in design thinking. During this phase, our team partnered with client business stakeholders to deeply understand operational workflows and desired outcomes. We then worked with client technology teams to co-design a solution that aligns with both current needs and long-term strategic goals.
• Agile Delivery: Using an iterative delivery model, we:
  
  • Built AI models, business logic, and integration components in focused sprints
  • Conducted regular collaborative design sessions to refine requirements
  • Delivered incremental features for early feedback from users and business stakeholders, this approach ensured continuous alignment with user needs and the solution’s evolving goals.
• Pilot Launch & Phased Rollout: We initially deployed the solution on a limited dataset, processing incoming correspondence from a single mailbox. This allowed us to monitor model performance in a controlled environment, gather insights, and refine the AI models. After successful validation, we moved to a full-scale rollout, completely replacing the legacy solution.
• Continuous Support & Improvement: Post-launch, we transitioned the platform to the client’s internal application support team. This included:
  
  • Comprehensive user training
  • Clear documentation and streamlined production support processes
  • Ongoing monitoring to detect performance drifts and initiate AI model updates when needed

‍

The solution delivered measurable improvements across key operational and customer experience metrics. Early results highlighted significant business value:

• 70% of enquiries are now processed end-to-end without manual intervention, with 90% document classification accuracy.
• Manual handling of misclassified or unclassified documents dropped from days to minutes.
• Operational costs—including third-party services, internal team effort, and infrastructure—were reduced by 3x.
• Same-day response rates improved significantly, leading to a noticeable uplift in customer satisfaction.‍
• Compliance risks were mitigated through full control over sensitive personal data, now processed entirely within the company’s secure cloud infrastructure.

Curious how AI can create real impact for your business? Let’s explore the possibilities together.