Challenges

A leading mortgage and finance brokerage platform in Australia leverages AI-driven automation to streamline workflows and enhance lead generation. Its technology automates key tasks such as document collection, messaging, appointment scheduling, and real-time lead responses.

To help mortgage brokers operate efficiently, increase lead engagement, and deliver exceptional customer experiences, they sought to implement a GenAI-based solution for end-to-end process automation. To do so, they needed to overcome the following challenges:

• Document Variability and Complexity: The project faced significant document variability. It involved six distinct document types, each with unique structures, layouts, and information fields. Document length varied drastically, especially with Contracts of Sale, which ranged from 2 to 400-500 pages. The system also had to handle variations in format and layout within the same document type, such as driver's licenses from different states. Furthermore, the system needed to process documents of varying quality, including scanned documents, photographs, and potentially damaged documents.
• Data Extraction Challenges: Data extraction presented its own set of challenges. Different fields required diverse extraction techniques, ranging from simple date extraction to complex financial data interpretation. Identifying and extracting relevant information from lengthy documents like Contracts of Sale, which could span hundreds of pages, presented a significant challenge.
• System Performance and Scalability: The solution needed to handle a large volume of documents efficiently, requiring a robust and scalable architecture. Specifically, it had to efficiently process and extract relevant information from potentially hundreds of pages in Contracts of Sale without significantly impacting overall system performance or processing times.
• Accuracy, Security, and Compliance: Given the sensitive nature of the information (personal and financial data), the system needed to maintain extremely high accuracy levels across all document types and fields. Handling sensitive personal and financial information required strict adherence to data protection regulations and the implementation of robust security measures.
• Integration and Workflow: The extracted data needed to be seamlessly integrated into the client's existing systems and workflows. This integration was crucial for ensuring that the extracted information could be readily used within the client's established processes, minimizing disruption and maximizing efficiency.

Partnering with Searce, they leveraged AWS GenAI to automate data extraction from key documents like driver's licenses, payslips, contracts of sale, notices of assessment, council rates, and credit score reports. The AI model was designed to handle complex document formats, varying layouts, and challenges like poor image quality or handwritten data. The goal was to reduce manual data entry, minimize errors, and accelerate processing times.

Searce Solution

Our solution employed a sophisticated, multi-faceted approach to address the diverse challenges presented by the document extraction project:

• Document Classification: Implemented a two-step classification process combining keyword-based and LLM-based approaches, leveraging AWS Rekognition's detectText API for initial text extraction and keyword matching, followed by a fine-tuned LLM for more precise classification, particularly in identifying payslips.
• Custom Textract Adapters: Developed and trained custom Amazon Textract adapters for driver's licenses and notices of assessment, using a dataset of 138 driver's licenses from all eight Australian states and territories, along with 150 notice of assessment documents, to ensure high accuracy in tax-related information extraction.
• LLM-Based Extraction: Utilized the Anthropic Claude-3 LLM model to extract key information from payslips, contracts of sale, council rates, and credit reports, implementing specialized prompts and data models for each document type to ensure accurate field extraction.
• Hybrid Approach for Complex Documents: Implemented a multi-step process for credit reports, involving PDF-to-image conversion, Textract OCR, and LLM analysis to handle the complexity of these documents.
• AWS Integration: Utilized the Anthropic Claude-3 LLM model to extract information from payslips, contracts of sale, council rates, and credit reports, implementing specialized prompts and data models for each document type to ensure accurate field extraction.
• Robust Error Handling and Validation: Incorporated comprehensive error checking and validation processes to ensure data integrity and handle edge cases.
• Scalable Architecture: Designed the solution using serverless architecture (AWS Lambda) to ensure scalability and cost-effectiveness.
• API Development: Created a set of APIs for document upload, S3 URL generation, and data extraction, allowing for easy integration with existing systems.
• Security Measures: Implemented secure file handling using signed S3 URLs and ensured all data processing occurred within secure AWS environments.

Business Impact

The implementation of our advanced document extraction solution delivered significant benefits to the client:

• Improved Efficiency: Reduced document processing time by up to 80% through automated extraction of key information from various document types, minimizing manual data entry and enabling faster decision-making and improved customer service.
• Enhanced Accuracy: Achieved over 95% accuracy in data extraction across all document types, minimizing errors associated with manual data entry.
• Cost Savings: Reduced the need for temporary staff during peak periods.
• Scalability: Enhanced the solution's ability to handle large volumes of documents and processed 3x more documents without increasing staff.
• Improved Customer Experience: We sped up processing times, helping in quicker loan approvals and credit assessments, enhancing customer satisfaction.
• Data-Driven Insights: Enabled better analytics & decision-making processes through structured extraction of data from various documents.
• Regulatory Compliance: Improved accuracy and consistency in data extraction helped in better compliance with financial regulations and reporting requirements.
• Competitive Advantage: Delivered a significant competitive edge by facilitating faster & more accurate document processing.
• Flexibility and Future-Proofing: Created a future-ready, modular solution allowing for easy addition of new document types or extraction fields as business needs and LLMs evolve.

Our innovative document extraction solution not only solved the immediate challenges faced by the client but also positioned them for future growth and efficiency. By leveraging cutting-edge AWS technologies and a thoughtful, hybrid approach, we delivered a system that significantly improved operations, enhanced customer experience, and provided a strong foundation for ongoing digital transformation efforts.