Case Study
AI-Driven Bank Statement Analysis
Challenge
A non-QM lender that analyzes bank statements to estimate the income of self-employed borrowers identified a strategic advantage in completing the analysis in less time and being able to render a conditional approval to the borrower more quickly. The existing manual analysis was labor-intensive, required significant staff training, and therefore did not scale well during times of peak application demand. An AI-driven automated platform could replace the manual process, but data capture and fraud detection logic would need to be at near-perfect levels.
Industry
Financial Services / Mortgage
Data Type
Semi-Structured (.pdf, .png, .jpg, .tif)
Project Duration
7 Months
Ongoing?
No
Solution
A human-augmented document scanning processes enabled document data capture exceeding 99.9% accuracy by comparing captured data to the original input documents (bank statements) and editing incorrect or missed data via a machine learning loop, which continually increased the accuracy of the first-pass data capture. Manual statement analysis (using the data captured in the automated process) continued in parallel for 3 months, allowing underwriters to provide feedback to the fraud detection model via a separate machine learning loop.
Outcome
A labor-intensive manual process which added 3 business days to the initial underwriting of mortgage applications was replaced with a fully automated process which reduced the time to less than 3 hours per application. Underwriters were re-deployed from bank statement analysis to fullfile underwriting, increasing loan production capacity with no increase in headcount.
Download Case Study