An SME lender's credit analysts spent hours retyping bank statements into spreadsheets before any real underwriting began. We built an OCR pipeline that parses statement PDFs, categorizes every transaction, computes key ratios, and delivers a ready workbook in 12 minutes.
Typical results
12 min
Upload to finished workbook
3x
Analyst throughput
2 days
To credit decision, from 6
Figures reflect typical results for this kind of automation.
The challenge
Every loan application arrived with three to six months of bank statements in inconsistent PDF formats. Analysts manually keyed transactions into Excel, categorized them line by line, and built cash-flow summaries by hand — about 3.5 hours per file before assessment even started. Applications queued for days, and rushed data entry produced costly categorization errors.
Before and after
Prep time per application
−81%Time to credit decision
−67%Files per analyst weekly
3.1×The automation
We built a document pipeline that ingests statement PDFs from the application portal and email, runs them through OCR, and uses an AI model to classify every transaction — revenue, recurring costs, loan repayments, owner drawings. The automation assembles monthly cash-flow tables, computes average balances and debt-service ratios against the lender's credit model, and generates a formatted underwriting workbook. Twelve minutes after upload, the analyst gets the finished file with anomalies already flagged.
How it works
- 01
Statements arrive
Applicants upload bank statement PDFs via the portal or email; every file is captured automatically.
- 02
OCR parsing
Each statement is parsed and every transaction extracted with dates, amounts, and counterparties.
- 03
AI categorization
Transactions are classified into revenue, recurring costs, repayments, and owner drawings; anomalies get flagged.
- 04
Ratios computed
Monthly cash-flow tables, average balances, and debt-service ratios are calculated against the credit model.
- 05
Workbook delivered
A ready underwriting workbook reaches the credit analyst about 12 minutes after upload.
Starts from
- Bank statement PDFs
- Application portal uploads
Delivers
- Underwriting workbook
- Cash-flow ratios
- Analyst notification
The impact
Underwriting prep dropped from 3.5 hours to 40 minutes per file — an 80% cut — and each analyst now assesses 28 applications a week instead of 9. Credit decisions go out in 2 days instead of 6, and categorization is consistent on every single file.
Questions about this automation
Which bank statement formats can it read?
It handles statement PDFs from different banks, including scanned pages, uploaded through the application portal or sent by email. OCR extracts every transaction with its date, amount and counterparty. Statements with missing pages or unreadable sections are flagged so the analyst can request a clean copy before the file moves on.
How are transactions categorized?
An AI model classifies each transaction as revenue, recurring cost, loan repayment, owner drawing or another category the lender defines. Rules handle known counterparties consistently, and unusual items such as large one-off transfers or bounced payments are flagged as anomalies for the analyst to review.
What does the credit analyst receive at the end?
A formatted underwriting workbook with monthly cash-flow tables, average balances and debt-service ratios calculated against the lender's credit model, plus the list of flagged anomalies. The analyst reviews it and makes the credit call. Simple automations like this are usually live within days once the workbook template and categories are set.