loading

Business Impact

  • ~80%

    extraction pipeline accuracy across three sets of documents

  • <3

    minutes to process, extract and compare each set of documents

  • Scalable to accommodate larger volumes of data and include other document structures

Client Key Facts

  • Location : North America
  • Industry : Commercial Real Estate Finance

Problem Context

Data, text, and figures typically contained in loan agreements,  commitments, and system upload templates must match precisely. The comparison of these lengthy, complex commercial mortgage documents is typically manual and can take over an hour per loan to complete. The objective of this exercise was to determine if the comparison of values in these documents could be automated given the length, file size, and variation in loan agreements.

Challenges

  • Difficulty in extracting data using traditional methods due to complex data contained in loan agreement documents 
  • Low model performance due to less volume of data
 Challenges

Tools & Methods

Google Cloud Storage
Google Compute Engine
Google Kubernetes Engine
Google Cloud Registry
Google Cloud SQL
Google Workbench

Solution

Quantiphi developed an end-to-end machine learning (ML) pipeline (entity extraction and field comparison process) which helped the Greystone team automate the comparison between different sets of documents. It showed the matches and exceptions so that the discrepancies within the documents can be effectively resolved.

The web application was built and hosted using Google Kubernetes Engine, which allowed the user to upload the required documents for processing and download the comparison results in an xlsx format. The entire lifecycle of the documents involved uploading to Google Cloud Services, pre-processing, extracting data using Cloud Vision OCR, post-processing on the output JSON, running through a comparison and discrepancy check, storing into a SQL database and viewing on the UI in a spreadsheet format.

Results

We implemented the engagement roadmap to:

  • Deliver an end-to-end ML pipeline that extracts entities from three types of documents and runs a comparison and discrepancy check with satisfactory accuracy.
  • Deploy the solution with a Ul that allows users to upload documents, run comparisons, view and download results on a single screen.

“Greystone is continually seeking ways to improve processes and our relationships with Agency partners including Fannie Mae and Freddie Mac. In this case, finding a faster, more efficient process to compare critical loan data using Document AI has improved our ability to exceed Agency timeline expectations.”

Leslie Dominy, Sr. Vice President, Greystone

“Greystone is always looking to find innovative and forward-thinking technology solutions and partners. As a partner, Quantiphi went above and beyond to deliver this solution – both in project execution and technical expertise. I will not hesitate to work with them again if the opportunity presents itself.”

Niraj Patel, CIO, Greystone

Disclaimer : Greystone is a private national commercial real estate finance company with an established reputation as a leader in multifamily and healthcare finance, having ranked as a top FHA, Fannie Mae, and Freddie Mac lender in these sectors. Loans are offered through Greystone Servicing Company LLC, Greystone Funding Company LLC, and/or other Greystone affiliates. For more information, visit www.greystone.com.

Looking for similar project?

Let's Talk

Get your digital transformation started

Let's Talk