Banking Document Automation

Automate credit risk decisioning

Unstract reads a borrower's financial statements, tax returns, bank statements and credit report, spreads them into your template, checks the ratios against your credit policy, and sends the result to your loan origination system or decision engine.

loan-40817 · small business term loanrunning
Package4 documents, 3 years
Read & extract418 lines and balances
Spread & checkrevenue off by $245K
To decisionDecision engine
09:02:11RM upload · financials FY23–25, two 1120s, 12 bank statements, credit report
09:02:40Read · 418 lines and balances · lowest confidence 0.91 (FY24 1120 scan)
09:02:44Spread · DSCR 1.40x · debt/EBITDA 2.50x · current ratio 1.50
09:02:45Check · FY25 revenue $4.86M on statements, $4.615M on 1120 · held
09:31:08Analyst confirmed accrual timing, $245K · noted in credit memo
09:31:09Spread, ratios and memo sent to the decision engine

Trusted by lending and onboarding teams around the globe

G2 Leader, Fall 2026G2 Best Usability, Summer 2026
Moody'sEndpointMedaxionPinpointConvrSmart ClerkCyberSoftThoughtFocusZluriIOA
The problem

Every credit package is spread by hand before anyone can judge it

A borrower sends three years of statements in their accountant's layout, tax returns as scans and a year of bank statements as PDFs. An analyst keys every line into the spreading template before credit analysis can even start.

FY25 SPREADMARCREDIT COMMITTEE

An analyst keying three years of statements the week before committee

FORM 1120RECEIVED 03/04

A tax return scanned crooked, with a stamp across gross receipts

FIN STATEMENTSREVENUE4.86MFORM 1120REVENUE4.615M

Revenue on the statements that the tax return doesn't support

LOAN 40817 · DOCUMENTSMONstatementsTUE1120 FY25THUbank stmtsWK 2⚑ FY23 missing

A package that trickles in over two weeks, one year short

DEBT SCHEDULEEQUIP NOTE184,000FY24OPENEQUIP NOTE184,000FY24⚑ DUP

An equipment note keyed twice, which pushes leverage past policy

Each package costs an analyst hours of keying and tying out, so the borrower waits and the credit memo starts late.

Unstract reads the package as it arrives, spreads it into your template, works out the ratios against your credit policy and holds only the packages that don't reconcile.

How it works

From the borrower's package to the decision engine, without the keying

Five steps, run on their own for every credit package. Parsing by LLMWhisperer, extraction by Document Studio, the whole run by Flow.

1 · Package intake

Pick up the credit package wherever it lands

The relationship manager uploads what the borrower handed over, the borrower adds the rest through the portal, and the credit report comes from your bureau pull. Financial statements, tax returns, bank statements, the debt schedule and the credit report are tied to one loan record as each one arrives.

The run starts with the first document, and the file shows what is still missing.

RELATIONSHIP MANAGERFinancials FY23–25BORROWER · PORTAL1120s FY24–25BORROWER · PORTAL12 bank statementsBUREAU PULLCredit reportLOAN 40817statements3 yearstax returns2bank stmts12 monthscredit1 report
2 · Read and extract

Read every statement and return, line by line

LLMWhisperer reads each document as text with its layout kept, so a scanned 1120 and an accountant's PDF come out equally clean. Document Studio then extracts every line of the balance sheet, income statement and cash flow for each year and entity, the balances and deposits from the bank statements, and the tradelines from the credit report.

The same field definitions run on every borrower, whichever accountant prepared the statements.

INCOME STMTfiscal_yearFY2025✓revenue4,860,000✓net_income361,000✓interest_exp118,000✓income_tax96,000✓depreciation167,000✓confidence0.97accountant PDF
3 · Spread

Map the statements into your spreading template

Each extracted line is mapped to its row in your spreading template: revenue and cost lines to the income statement section, assets and liabilities to the balance sheet, the debt schedule to its own rows. Every borrower's statements land the same way, so the years compare side by side whatever the original layout was.

Analysts open a finished spread, not a stack of PDFs.

FIN-STMTS-FY25.PDF3 YEARS · 22 PAGESINCOME STMTrows 1–18✓BALANCE SHEETrows 19–44✓CASH FLOWrows 45–60✓DEBT SCHEDULErows 61–66✓
4 · Ratio and policy checks

Check the ratios and the tax return against your credit policy

DSCR, leverage and liquidity are worked out from the spread and checked against your policy thresholds. Revenue is tied to the tax return, reported cash to the bank balances, and the debt schedule to the tradelines on the credit report. A package that ties keeps moving. One that doesn't stops in a review screen with the figure, the documents and the gap side by side.

Set the conditions once (a policy exception, statements that don't reconcile, a missing year) and only those packages reach an analyst.

LOAN 40817 · HELD⚑ REVENUE FY25FIN STATEMENTSFORM 1120BANK DEPOSITS$4.86Mincome stmt · p. 3$4.615Mgross receipts · 1a$4.70M12 months totalDSCRDEBT/EBITDACURRENT1.40x · min 1.25x2.50x · max 3.50x1.50 · min 1.20CONFIRM + NOTEREQUEST CPA NOTEOpen the page →
5 · To the credit decision

Send the spread and ratios to the credit decision

The spread, the ratios and any flags go to your decision engine or loan system, such as FICO Platform, nCino or Moody's CreditLens, and the completed spreading template is saved with the loan. Each note an analyst added along the way travels with the file into the credit memo.

Nobody rekeys a spread, and every figure traces back to the page it came from.

LOAN 40817DSCR 1.40x1 exception, clearedDECISION✓ sent 09:31Decision engineLOAN SYSTEM✓ attachedSpread and ratiosMEMO✓ draftedAnalyst note FY25

Watch your last ten credit packages spread themselves.

Bring the statements, returns and bank statements exactly as borrowers sent them. We run them live against your template and policy and show every line, ratio and exception.

Supported documents & fields

Everything a borrower hands over for a credit decision

Documents

Fields captured

Borrower legal nameEINEntity typeFiscal year endRevenueCost of goods soldOperating expensesNet incomeInterest expenseDepreciation and amortizationTotal current assetsTotal current liabilitiesTotal debtOwner's equity
Connectors

Our agents are extroverts.

Cloud storage like S3, Google Drive, Dropbox and Box. Databases and warehouses like Snowflake, BigQuery, Postgres and SQL Server. Accounting, ERP and CRM systems like QuickBooks, Xero, SAP, NetSuite, Salesforce and HubSpot. Project trackers like Jira, Asana, ClickUp and Linear. And anything with an API.

100+
Connectors ready to go right out the box
Amazon S3
Google Drive
Dropbox
Box
Microsoft Azure
Google Cloud Storage
Snowflake
BigQuery
PostgreSQL
SQL Server
MySQL
Amazon Redshift
Oracle Database
SAP
QuickBooks
Xero
Sage
Odoo
HubSpot
Zoho
Jira
Asana
ClickUp
Linear
Getting live

Live on one loan product first

Your credit team doesn't write code, and your credit policy doesn't change. The run is set up on your own packages and template, then proven on one loan product.

01

Book a consultation

We go through ten recent credit packages with you: what borrowers send, how your analysts spread it, and which policy thresholds decide a hold.

02

Built on your spreading template

The fields, the template mapping, the policy checks and the hand-off to your decision engine are set up on your own borrowers' statements and returns.

03

Pilot one loan product, then widen

One product, such as small business term loans, runs live within days. Lines of credit and commercial real estate follow once the spreads match what your analysts would have keyed.

Who it's for

Built for the people who sign off on credit

Commercial and small business lenders

Small business packages spread the day they arrive, so a relationship manager can give the borrower an answer while the deal is still warm.

Credit risk and underwriting teams

Every spread mapped to the same template and every ratio checked against policy, with analysts working only the exceptions and the memo.

Credit unions and community banks

A lean credit department handles more member business loans without adding an analyst, and examiners see a trail behind every figure.

Private credit and direct lenders

Portfolio company financials spread every quarter in the same layout, so covenant tests run on fresh numbers instead of last quarter's Excel.

Outcomes

Spreads in minutes, exceptions caught before the memo

<3 minfrom a complete credit package to a spread in your decision engine
91%of credit packages spread without an analyst keying a line
54%fewer hours spent spreading borrower financials
2×faster credit answers back to the borrower
We used it to extract complex funding packets from credit unions… Instead of manually scanning through lengthy 60–70 page PDFs, staff could now view all extracted details through a simple, user-friendly interface.
G2Puneeth KumarSenior Developer, Digital Align · Reviewed on G2
1 / 3
Security

Borrower financials stay under your control

A credit package holds a company's full financials, its tax returns, a year of bank activity and the owners' personal credit history: the record a borrower trusted you with, and the one a breach would expose in full.

AICPA SOC 2ISO 27001GDPRHIPAA
  • Encrypted in transit and at rest.
  • No credit package is ever used for model training.
  • Access by role, loan product and document type, with every analyst decision recorded against the loan.
  • Retention you set: processed packages are deleted on the schedule your records policy requires.
  • Self-hosted deployment for lenders whose controls keep borrower data inside their own environment.
See how borrower data is handled in the Trust Center
FAQs

What credit officers ask before the first package

Look for one run that covers the whole path: picking up the package, reading every document, spreading it, checking it against policy and handing the result to your decision engine. Unstract does all five steps for each credit package and stops only where your policy says a person must look. The decision itself stays with your engine and your credit officers.

Unstract reads native PDFs, spreadsheets, scans and phone photos through one pipeline, and the layout is kept, so a column of figures stays a column. Each line maps to the same row of your spreading template whichever format it arrived in.

OCR turns a page into text. Spreading needs more: knowing which line is revenue, which is a one-off item, and where each figure belongs in your template. LLMWhisperer does the reading with the layout kept, and Document Studio's AI data extraction does the mapping, with a confidence score on every value so a doubtful figure stops for a person.

Automate the reading and tying out, and leave the judgment where it is. Unstract checks the tax return against the statements, the bank balances against reported cash and the debt schedule against the credit report. Credit underwriting stays with your underwriters, who start from a reconciled spread instead of a stack of PDFs.

Before it. Your credit risk management software scores and tracks the exposure once the numbers are in. Unstract produces those numbers from the borrower's own documents, so the software receives a checked spread instead of waiting on an analyst to key one.

The ratios your policy sets, such as DSCR, debt to EBITDA and the current ratio, against your thresholds. Alongside them: revenue against the tax return, cash against the bank statements, the debt schedule against the credit report's tradelines, and whether every year the policy asks for is present.

The package stops in a review screen with both figures, the page each came from and the bank deposits for the same year. The analyst confirms the difference with a note, such as accrual timing, or requests an explanation from the borrower's accountant. The note goes into the credit memo.

Your decision engine and loan system, such as FICO Platform, nCino or Moody's CreditLens, plus your own spreading template. Flow posts them as the last step of the run, with the documents and any analyst notes attached.

Unstract is SOC 2, ISO 27001, GDPR and HIPAA compliant. Statements, returns and credit reports are encrypted in transit and at rest, never used to train models, and kept only as long as your policy allows. Self-hosted deployment keeps them inside your own environment; see the Trust Center.

Days, not months, and your team writes no code. We build the run in a consultation around your template, your policy and your decision engine, then run one loan product live. Other products follow once the spreads match your analysts' own.

Next step

Bring ten recent credit packages to the consultation

The statements, returns and bank statements as your borrowers sent them, scans included. You watch each one spread, and every held package shows the figure that stopped it.