Coforge delivers ROI with AI by Optimizing and Simplifying Document Processing

A large team used to manually process mortgage and loan documents. The processing included manual document extraction, verification, and logging into the Client’s bank application. The tasks were repetitive in nature, and there was an obvious potential to optimize the operating costs, reduce errors, increase processing velocity, and bring in operational efficiencies.

Challenges in Current Process

  • There were 35 product lines and 5000+ document types across packages. Package size varied between 35 – 100 pages each. As an example, one product line had 1200+ standard document types and 400+ non-standard document types. Overall 3M+ pages were processed every month.
  • All the related documents are scanned into one large tiff file of 150+ pages for each mortgage loan. The structure of these documents is varied:
    • Obtaining high scan quality documents was challenging
    • Handwritten content like signatures, initials, and printed information made the task at hand complex.
  • There were defined document types, and the “n” number of document type(s) could be present in a package. Two broad categories of documents in a package – Standard and Non-Standard existed which led to operational inefficiencies.
  • Operators had to perform steps for each package received, including signature identification, classification, splitting of documents, and extraction of data elements.


  • An automated intelligent solution to read packages, extract important data elements, and split the packages with high accuracy.
  • Coforge proprietary accelerator “Document AI” was leveraged to address these challenges for the client. To assist in addressing client challenges, signature extraction modules, NLP-based classification, and extraction, logo identification modules were employed.
  • For each package following activities were implemented:
    • Identify signature, bank logo, and form number.
    • Classify, extract and split forms into pre-defined groups.
  • The output was provided as an API, integrated with the downstream application.


  • Automated error handling and improvement observed in overall output quality.
  • Augmented efficiency in output with intelligent automation processing millions of applications per month.
  • Optimized resource allocation, with an overall reduction in FTE.