Client Success Story

Intelligent Document Processing (IDP) and Workflow Automation for a Commercial NBFC

20-30%

Reduced IT
Costs

35%

Higher
Throughput

60%

Reduced Manual
Effort

90%

Reduced Data
Entry Error Rate

Service

  • Business Process Automation
  • Computer Vision

Technology

  • AWS
  • OCR
  • TensorFlow Models
The client

A Commercial NBFC

Our client is a mid-sized commercial finance firm operating across the North American middle market. They specialize in business credit and asset-backed loans, managing hundreds of new loan applications monthly.

THEIR CHALLENGE

Manual Underwriting Stalling Disbursal Velocity & Processing Accuracy

The client faced stiff competition from other agile FinTech firms as they were still relying on legacy systems and manual processes for due diligence (verifying complex documents like financial statements, tax returns, etc). They were struggling with:

  • High Labor Cost: Their dedicated team of financial analysts spent an average of 20-30 hours per client file simply aggregating, verifying, and manually entering data from unstructured documents.
  • Processing Delay: The lengthy internal process resulted in a Time-to-Yes (pre-approval decision) averaging 7 to 10 business days (much more than what modern companies offer).
  • Compliance Risk: Manual data entry across hundreds of monthly applications sustained an unacceptable 7% error rate in critical financial data used for risk scoring.
THE REQUIREMENT

Digital Workflow Engineering

To overcome manual workflow inefficiencies, the client was looking for a workflow automation and document processing solution that would securely handle all client data while adhering to privacy regulations (CCPA, SOC 2, etc). This solution should:

  • Integrate with their existing legacy systems
  • Provide centralized, secure access to all digitized loan-related documents
  • Automate internal document processing and loan underwriting
  • Free up resources from low-value data entry to high-value tasks like exception handling and client relationship management
OUR SOLUTION

Intelligent Document Processing (IDP) and Workflow Automation Platform

We engineered a secure, cloud-native IDP and automation platform hosted on AWS. The platform is designed with a microservices architecture to isolate tasks, a DynamoDB/Amazon RDS database for a single source of truth, and a custom-trained IDP module with a dynamic workflow engine to automate the entire due diligence process.

Workflow

Intelligent Document Processing (IDP) - workflow Intelligent Document Processing (IDP) - workflow
1

Data and Infrastructure Assessment (Foundation)

We conducted a comprehensive audit of the firm’s existing document archives, data infrastructure, and legacy systems to determine data readiness, governance needs, and integration points.

2

Data Processing and Cloud Setup

We ran several batches of documents through computer vision (OpenCV-based) OCR tools to digitize them. Our experts then applied data-cleansing rules and standardization algorithms to normalize entity formats (e.g., dates, currencies, nomenclature) across the entire archive.

This step also involved setting up an AWS environment, defining the microservices architecture, setting up VPC (virtual private cloud) security, and configuring the Amazon RDS (PostgreSQL) instance to serve as the single source of truth for this structured data.

3

IDP Module Development

We built a proprietary IDP module to automatically ingest, categorize, and standardize it. This module was deployed on AWS Lambda using custom APIs managed by Amazon API Gateway. This module extracted data and instantly classified it based on key attributes using custom-trained (on the client’s proprietary data) TensorFlow models.

Extracted attributes:

  • Gross Revenue/Sales
  • Net Income/Loss
  • EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortization)
  • Operating Expenses
  • Existing Debt Obligations
  • Working Capital
4

Process Orchestration

We then developed a process automation engine that calculated the preliminary financial risk score automatically from these insights. Loan files meeting pre-set, low-risk thresholds were immediately fast-tracked, and automated notifications were triggered to alert analysts.

5

Testing, Optimization, and Ongoing Support

We implemented an automated MLOps pipeline on SageMaker to retrain and optimize the TensorFlow models, validating accuracy at every iteration. This ensured the system consistently maintained a 99.5 percent data extraction accuracy benchmark. Our team also ran controlled A/B evaluations on new document layouts and regulatory changes to keep the model robust and compliant.

Project Outcomes

20-30% reduction in operational IT costs

35% higher underwriter throughput

60% reduction in manual effort needed for audit trail generation

90% reduced data entry rate with automation

Technology Stack

Component / Layer

Technology Used

Cloud Platform
  • Amazon Web Services
    Amazon Web Services (AWS)
API/Compute
  • AWS Lambda
  • API Gateway
Primary Databases
  • Aws Rds Streamline Icon: https://streamlinehq.com
    Amazon RDS (PostgreSQL)
  • AWS DynamoDB
Intelligent Document Processing (IDP)
  • OpenCV
  • Custom TensorFlow Models
CONTACT US

Looking to Automate Some of Your Processes?

We build production-grade workflow automation platforms that enhance efficiency, reduce manual workloads, and standardize operations across multi-site environments. Contact us to learn more about our business process automation services.