Business Impact

  • ~90%

    Classification Accuracy

  • ~85%

    Extraction Accuracy

  • 6

    Types of Document Classified

Customer Key Facts

  • Location : United States
  • Industry : Financial Services
  • Size : 50 Employees

Problem Context

The customer is a wholesale/correspondent lender. They wanted to classify submission documents and extract required and relevant information for downstream consumption.


  • Extensive time spent by underwriters on a low value, repetitive work
  • High error rate due to manual processing

Document Requirement

Form 1040
IRS W2 Forms
Driving License
Bank Statements


Quantiphi re-engineered the document underwriting workflow from ingestion and classification to the extraction of mortgage applications. We also added features such as human-in-the-loop review and model retraining to the solution for higher accuracy.

  • Ingestion: Simple and robust UI to upload documents
  • Document Classification: Classify documents to determine type and content
  • Extraction: Extract multiple fields, values, and embedded objects from submission documents
  • Human-in-loop: Edit, rename, add and delete documents or extracted values as and when required
  • Downstream Connectivity: Output module in compliant MISMO format
  • Active Learning and Retraining: Based on the HITL actions and processing of newer documents


  • Industry-leading classification and extraction accuracy for underwriting documents
  • Streamlining and standardizing of underwriting workflows

Looking for similar project?

Let's Talk

Get your digital transformation started

Let's Talk