About the Customer

The customer is a global financial technology company that provides services to businesses in the financial services sector, including banks, credit unions, thrifts, securities brokers, mortgage, insurance, leasing, and financing firms, as well as retailers.

Business Challenge

The customer needed help expediting the data extraction and validation process through automation for one of its daily operations, which required manual labor and was using resources that could be reassigned.

Our client would issue cheques to its merchants, which, due to various reasons, could return them. These cheques were then scanned and uploaded to the remittance manager application by the customers’ data center team and classified as Corporate and Draft cheques.

When returning the cheques, merchants would attach the letter/document which explained the return reason either digitally or handwritten & scanned. These documents had to be validated, and the corresponding case updated in the CRM platform.

Brillio’s Solution

Brillio assessed their business requirement and explored the features of the Enterprise Intelligent Document processing tool and Opensource OCR Platform/Algorithms to attain the required accuracy and precision on data extraction.

The solution identified includes a combination of Robotic Process Automation (RPA) technology and AI/ML & Deep Learning models for cheque data extraction.

The solution facilitates the extraction of data from handwritten or printed checks and mail and uses RPA bots to process the data on BU applications.

The implementation steps followed by Brillio to achieve the desired outcome:

•  Data analysis:

• • Scan financial documents from cloud sources and categorize them according to their type: handwritten or digital.

• • Identify metadata patterns and variations.

• • Localize document metadata pattern complexity (key entities on the document).

• Heuristic mapping:

• • Financial documents contain bank cheques and letters of rejection. The mapping of these bank cheques and subsequent multiple/single/handwritten letters was achieved by comparing metadata extracted on the cheques and letters and then, based on the data, building subsequent mapping.

• Pre-processing:

• • Enhance documents using image processing technology by smoothening images, doing morphological transformations, and leveraging noise removal techniques.

• Entity Detection: 

• • Built object detection models to identify key cheque entities.

• • Increased model performance through loop model evaluation and fine-tuning.

• • Leveraged the detection model to extract the text on ROI regions, using Google tesseract OCR solution.

• • Applied image processing techniques to enhance ROI regions for better OCR extraction.

Business Impact

Following Brillio’s implementation, the client was able to:

• Seamlessly identify and classify cheques for different merchants.
• Quickly extract metadata for checks processing in CRM based on the rejection reason provided.
• Create records of each transaction for audit purposes.
• Automate the processing of 5000+ checks/ month.

The results led to:

• 60-80% accuracy in cheque identification, classification, and processing.
• ~65% reduction in the effort required for check processing.
• The implementation of a scalable solution for other BU that process merchant cheques.