Pay stubs
Gross pay, pay frequency, year to date, employer and pay dates. Monthly income is calculated for any pay cycle.
Document intelligence
Documents read, checked and flagged before a person touches the file
Algebrik's Agentic Document AI reads what members upload, compares it with the application, and clears or flags it. Your underwriters keep the final say.
Fig 1Document scanner
A scan bar reads the page and lights each field5 of 5 fields read
Illustrative: a sample pay stub with invented data.
What it does
Document AI reads every upload, checks it against the application, and flags what does not match. It never makes the credit decision. Your people do.
What it reads
Every document a loan file needs
Members upload from the portal, an emailed link or their phone camera. Staff can upload in the Lender's Cockpit.
W-2s and tax returns
W-2s, 1040s and Schedule K, read and compared with what the member stated.
Bank statements
Read and cross-checked against the application.
Government IDs
The barcode is read and checked against the printed front, and mismatches are flagged.
Insurance documents
Declaration pages, cards and binders. Policy number, carrier, coverage dates and, for vehicles, the VIN.
Appraisals and title reports
For home equity: appraisal values and dates, owner names and subject property details.
Utility bills
Read, classified and matched to the application.
Photos and mixed PDFs
Camera captures are accepted. A single PDF of mixed documents is split into separate, typed documents.
Accepted formats: PDF, PNG, JPG, JPEG and DOCX, plus camera capture on mobile.
How it works
Six steps, from upload to cleared or routed
Every step is logged. Anything outside your tolerances goes to a person.
1
Upload
A member uploads from the portal, an emailed link or a phone camera. Staff upload in the Lender's Cockpit.
2
Classify
The system identifies the document type. Mixed PDFs are split into separate documents.
3
Extract
Structured data is pulled out: income, dates, VIN, ID fields, employer and coverage.
4
Validate
ID barcodes are checked against the printed front, and pay stubs for internal consistency. A fraud confidence score is assigned.
5
Cross-reference
Extracted data is compared with what the member stated. Differences become structured alerts.
6
Clear or route
Within tolerance, the stipulation clears automatically. Outside it, an underwriter sees the discrepancy.
What it flags
It flags what a person would have to hunt for
Every flag shows the discrepancy, so your team starts from the answer, not the search.
An ID barcode that does not match the front
The barcode on a government ID is read and checked against the printed data. A mismatch is flagged for review.
A pay stub that contradicts itself
Pay frequency that does not fit the year-to-date total and the date range is flagged as an inconsistency.
Figures that differ from the application
Extracted values are compared with what the member entered. Any difference surfaces as a structured alert.
A VIN that does not match the collateral
The VIN and coverage on an insurance card are read and checked against the collateral record.
A document with a low confidence score
Every upload gets a fraud confidence score. Anything below your threshold is flagged for review.
People stay in charge
Routine clears itself. Exceptions go to a person.
You set the tolerances. Document AI applies them the same way every time.
- 1. Document uploaded and readClassified, extracted and validated.
- 2. Checked against your tolerancesApplied the same way every time.
- 3a. Within toleranceThe stipulation clears automatically and the file moves on.
- 3b. Outside toleranceRouted to the underwriter queue with the discrepancy shown. Your team decides.
One engine, every loan
The same reading, wherever documents show up
HELOC
- Appraisal values and dates
- Title report details: owner names and property address
- Insurance policy number, carrier and coverage dates
Direct and indirect
- VIN and coverage read from insurance cards
- Checked against the collateral record
- Dealer paperwork uploaded straight into the application
Personal and signature loans
- Pay stubs for any pay cycle
- Government IDs and bank statements
- Stipulations that clear within your tolerances
Built for oversight
AI that assists, and stays accountable
Supervised AI only
Document AI assists underwriting. It does not replace it.
People at every checkpoint
Manual override is available at every stage.
A full audit trail
System actions are timestamped and attributable to a user or a process.
SOC 2 Type II, US-only data
Algebrik is SOC 2 Type II certified, and all data is stored in the United States.
Questions leaders ask
Straight answers
Does Document AI make credit decisions?
No. It extracts, validates and flags. Every exception and judgment call stays with your underwriting team.
Which documents and formats does it handle?
Pay stubs, W-2s, tax returns, bank statements, government IDs, insurance declaration pages and cards, and utility bills, plus appraisals and title reports for home equity. It accepts PDF, PNG, JPG, JPEG and DOCX, and camera capture on mobile. Mixed PDFs are split automatically.
What happens when something does not match?
Differences between a document and the application become structured alerts. Inside your tolerances, the stipulation clears automatically. Outside them, the file goes to an underwriter queue with the discrepancy shown.
Can we set the tolerances?
Yes. Stipulations clear within the tolerances you configure, and extraction can be configured for your credit union.
How do members submit documents?
Through the member portal, an emailed link or their phone camera. Staff can also upload in the Lender's Cockpit.
See Document AI in action
A short briefing for your lending and operations teams.
Book a Document AI demo