OCR Implementation with Amazon Textract Ebbed A Cash Management Company’s Reconciliation Efforts by ~65%

The customer is a full-service cash management company, providing a range of services including ATM site selection, currency forecasting, reconciliation, and cash replenishment. The customer helps its clients optimize the cash circulation and minimize downtime of the ATM network. The customer was seeking a solution for automated reconciliations to reduce the manual overhead and delays in the process.

The Challenge

The criticality of the currencies business mandated the customer to look for organizational reconciliation. The reconciliation was required to be performed on hundreds of images from ATM slips of their different bank partners. Their expectation was that the information gathered from the reconciliation is provided in real-time for early detection and rectification of errors.

The activities in reconciliations previously included detailed manual punching and transmitting the electronic image of validated slips to the respective bank partner network. This process was time-consuming and tedious, owing to the manual reconciliations via analyzing the images and identifying the discrepancies.

Optical character recognition (OCR) process was carried out through simple documents that were error-prone and resulted in a flat bag of words. The extraction process was template-based, limited by the accuracy of OCR, and involved development and management overhead.

The Solution

Blazeclan proposed the customer on implementing an automated solution for analyzing and processing the information in images and mitigating challenges in real-time. Blazeclan helped the customer with the implementation of Amazon Textract, an optical character recognition (OCR) proof-of-concept, which directed resulted into bag of selective words, faster processing, thereby saving time and cost. With this POC, the customer was able to parse images including the ATM slips, deposit slips, and scratch cards, and harness relevant attributes from multiple formats. The result of this was presented as APIs for further use.

The Approach Followed:

• Developing a cost-effective, efficient, and scalable OCR solution for the bank partner networks’ ATM slips, scratch cards, and deposit slips.
• Setting up the OCR process for data identification from images with the help of Amazon Textract.
• Performing data engineering on all image formats for extracting important keywords from raw texts received from Amazon Textract.
• Developing an API for locating processed data in real-time.
• Setting up the AWS infrastructure for automating the flow of the OCR solution.

The implementation of the OCR solution by leveraging Amazon Textract delivered the customer with automated data validation activities.

Architecture Diagram

Benefits Achieved by the Customer

• Automated manual data validation helped the customer save an enormous amount of time and effort.
• OCR Pipeline implemented is highly scalable. It provided the customer with the ability to process a huge volume of ATM slips in real-time. Aso, it helped the customer eliminate human errors completely along with enhanced accuracy
• The solution offered the customer with significant cost savings in terms of document processing.
• The solution delivered high accuracy in data validation as compared to manual data validation, which was error-prone.
• The customer achieved high accuracy in their information extraction processes due to implemented error checks and complete elimination of human intervention.

Tech Stack