The Business Need

A UK-based financial services provider faced difficulties automating NAR1 form data extraction, which was dependent on a manual process that was both inefficient and error-prone. This inefficiency in data extraction led to decision-making delays, hindered scalability, and posed challenges in maintaining regulatory compliance. The need for an automated, reliable solution became crucial to streamline workflows and ensure accurate, compliant operations.

Challenges

The client struggled with data extraction due to unstructured data, intricate layouts, and scalability concerns. With growing document volumes, maintaining regulatory compliance and ensuring consistent accuracy across extracted data became increasingly complex. Legacy extraction processes and disparate systems caused delays and bottlenecks, further hindering efficient data processing.

• Complex formats, including checkboxes and tables, disrupted seamless data extraction.
• Extracting 150+ attributes created significant difficulty.
• High document volume limited scalability options and smoother integrations.
• Unstructured data sources hindered efficient information retrieval, requiring quality checks.

XDAS approach

To address the client’s NAR1 form data extraction needs, the XDAS team developed a robust, structured workflow to ensure precise data processing, minimize manual effort, and enhance overall efficiency.

Seamless data extraction

We build a workflow intending to transform static NAR1 forms into machine-readable documents using a searchable PDF extractor. This process made the data accessible for automated processing. To handle unstructured data effectively, we deployed a PDF chunker to segment large documents into manageable sections, facilitating targeted attributes. By leveraging a filter bot, irrelevant or null data was discarded, allowing the client’s teams to focus on actionable data.

GenAI-powered data structuring

The filtered data was processed using an advanced key-value pair GPT model, which structured the extracted information into organized fields. This GenAI-powered approach ensured the accurate identification and mapping of essential attributes, minimizing errors and reducing the need for manual intervention. The structured data was seamlessly integrated into the client’s system for further analysis and reporting.

Human-in-the-loop for quality control

We embedded the human-agent mechanism into the NAR1 form extraction process, allowing expert reviewers to validate captured data against precise regulatory benchmarks. This integration ensured that intricate data points, such as transaction codes and compliance indicators, met strict accuracy standards. The HITL process minimized errors, mitigated risks, and provided clients with confidence critical for industry compliance.

Customizable reporting

Integrating a report generator bot into the XDAS solution enabled stakeholders to access insights in various formats tailored to their specific requirements. This flexibility simplified decision-making and ensured the outputs aligned with the client’s business objectives.

Results

High-coverage attributes

Extracted over 150 attributes from complex layouts of NAR1 forms.

Achieved 98% accuracy

Achieved high precision in data extraction with faster processing, with HITL facilitating the final validation.

Enhanced scalability

Assisting teams across the organization to integrate seamlessly with legacy systems and scale larger data volumes effortlessly.

Improved operational efficiency

A fully automated workflow reduced processing time by 60% and eliminated inconsistencies.