Back to Resources
CASE STUDY

How FinanceFlow Automated 10,000 Loan Applications per Month

A mid-sized lender reduced manual document review time by 85% and cut processing costs by $2.1M annually using Retriv.ai's Extract API.

By Sarah MartinezNovember 10, 20246 min read
85%
Reduction in Processing Time
$2.1M
Annual Cost Savings
10,000+
Applications per Month

The Company

FinanceFlow (name anonymized) is a digital lender specializing in small business loans and personal mortgages. Founded in 2018, they grew rapidly during the pandemic as businesses sought fast, online-first lending solutions.

By early 2024, FinanceFlow was processing over 10,000 loan applications per month. But their operational costs were unsustainable.

The Problem

Each loan application required manual review of 8-15 documents: pay stubs, tax returns, bank statements, employment verification letters, and more.

  • 22 full-time underwriters dedicated to document data entry
  • 45 minutes average processing time per application
  • ~12% error rate in manual data entry
  • 3-5 day turnaround time from submission to decision
"We were growing fast, but we couldn't hire underwriters fast enough. And even if we could, the manual data entry was killing our margins. We needed automation, but every OCR tool we tried was too unreliable for production."
— David Park, COO of FinanceFlow

The Search for a Solution

FinanceFlow's tech team evaluated several document processing vendors:

Traditional OCR (Tesseract, AWS Textract)
❌ Rejected
Raw text output required extensive post-processing. Tables were completely unusable. Would still need humans to structure the data.
Template-Based Solutions
❌ Rejected
Required weeks of setup per document type. Broke every time formats changed. Not scalable for small teams.
Custom ML Models
❌ Rejected
Quoted $300K+ for training custom models. 6-month timeline. Ongoing maintenance costs.
Retriv.ai
✅ Selected
API-first, no templates needed, worked out-of-the-box on their documents. 15-day POC showed 97% accuracy.

Implementation

FinanceFlow's engineering team integrated Retriv.ai in just 3 weeks. Here's how they did it:

1

Week 1: Schema Design

Defined extraction schemas for each document type (pay stubs, W-2s, 1099s, bank statements, etc.). Retriv's team provided consultation on best practices.

{
  "employer_name": "string",
  "employee_name": "string", 
  "gross_pay": "float",
  "ytd_earnings": "float",
  "pay_period_start": "date",
  "pay_period_end": "date"
}
2

Week 2: API Integration

Built a microservice that receives uploaded documents, calls Retriv's Extract API, and pushes structured data into their Encompass LOS (Loan Origination System).

3

Week 3: Human-in-the-Loop Review

Built a UI for underwriters to review and correct extracted data before final submission. Used Retriv's visual grounding to show exactly where each field was extracted from.

Results

FinanceFlow launched their automated document processing in April 2024. The results exceeded expectations:

85%
Faster Processing
Average application processing time dropped from 45 minutes to 7 minutes.
$2.1M
Annual Savings
Reduced headcount from 22 to 8 underwriters. Repurposed staff to higher-value work.
97.4%
Extraction Accuracy
Down to just 2.6% error rate, less than half the previous manual rate.
24 hrs
Faster Decisions
Turnaround time from submission to decision dropped from 3-5 days to 1-2 days.
"Retriv.ai didn't just save us money — it fundamentally changed our business model. We can now compete with the big banks on turnaround time while maintaining better margins. Our NPS score from borrowers went up 18 points because we're so much faster."
— David Park, COO of FinanceFlow

What's Next

FinanceFlow is now expanding their use of Retriv.ai:

  • Processing commercial loan applications (more complex documents)
  • Building automated fraud detection using Retriv's metadata analysis
  • Extracting data from title reports and appraisals for mortgage processing
  • Creating a RAG-powered chatbot for loan officers to query historical applications

Key Takeaways

  • API-first beats custom models for most lending use cases
  • Human-in-the-loop workflows balance automation with accuracy
  • Visual grounding is critical for building user trust
  • 3-week implementation is realistic for mid-sized lenders

Want Similar Results?

Schedule a demo to see how Retriv.ai can transform your document processing.

SM
Sarah Martinez
Head of Customer Success @ Retriv.ai
Previously: Product @ Plaid