Bank Statement Analysis for Underwriting

Analyse bank statements from PDFs, scans and Account Aggregator data to surface income, obligations, bounces and red flags, with every figure traceable to transactions.

Capability:: Document Intelligence
Industry:: NBFCs & Lending

Book a 20-minute call WhatsApp us

Who it is for

Chief Credit Officer
Underwriting Team Lead
Head of Risk Analytics
Fraud Control Unit Manager

The problem

For self-employed and thin-file borrowers, the bank statement is often the most honest picture of cash flow. Underwriters read months of transactions to estimate income, spot other EMIs, count cheque and NACH bounces and judge whether balances are genuine, while fraudsters edit PDFs or route funds in circles to inflate turnover.

Formats that vary by bank

Statements from different banks, password-protected PDFs, scans and Account Aggregator data each need separate handling before analysis can begin.

Income hard to separate from noise

Salary, business receipts, transfers between own accounts, loan disbursals and cash deposits look similar in narration text.

Hidden obligations

EMIs to other lenders, informal loans and recurring payments are spread across narrations that differ by bank and channel.

Edited statements and circular flows

Altered PDFs, recalculated balances and round-tripping between related accounts are hard to catch by eye.

What we build

We build a statement analysis service that accepts e-statement PDFs, scanned copies and Account Aggregator data. It parses transactions, recomputes running balances and checks PDF integrity, then categorises each transaction using rules and a language model that reads narrations in context, including UPI, NEFT, NACH and cheque patterns. The service computes the metrics your credit policy uses, such as average balances on set dates, monthly credits net of own-account transfers, existing EMIs and bounce counts, and highlights red flags like sudden credit spikes, circular transfers or cash-heavy months. Every figure links to its underlying transactions, so the underwriter can verify it before using it in the credit memo.

Parses statements from major Indian banks, scanned copies and Account Aggregator FI data
Running-balance recomputation and PDF metadata checks to flag possible tampering
Transaction categorisation for salary, business receipts, transfers, EMIs, bounces and cash
Credit-policy metrics such as average bank balance, net monthly credits and existing obligations
Red-flag detection for circular transfers, related-party flows, credit spikes and window dressing
Underwriter view with every metric drilled down to its transactions

How it works

01Receive statementsStatements arrive as uploaded PDFs, scans or consented Account Aggregator data linked to the application.
02Parse and verifyTransactions are parsed, balances recomputed and document integrity checked, with failures flagged.
03Categorise transactionsRules and a language model categorise each transaction and identify counterparties and recurring patterns.
04Compute policy metricsMetrics defined in your credit policy are computed per account and across accounts, net of own-account transfers.
05Flag risksTamper signals, bounces, undeclared EMIs and unusual flows are listed with linked transactions.
06Underwriter reviewThe underwriter reviews metrics and flags, corrects categories where needed and uses the output in the credit memo.

Bank Statement Analysis for Underwriting: workflow diagram — Open full size

Architecture

Intake
- PDF and scan upload with password handling
- Account Aggregator FIU data receiver
- Application linking
Parsing and integrity
- Bank-specific and generic table parsers
- OCR for scanned statements
- Balance recomputation and PDF metadata checks
Categorisation
- Narration rules for UPI, NEFT, NACH and cheques
- LLM categoriser for ambiguous narrations
- Counterparty and recurrence detection
Analytics
- Credit-policy metric engine
- Red-flag rules
- Cross-account netting
Review and integration
- Underwriter workbench with drill-down
- LOS and credit memo adapter
- Category correction feedback

Inputs and outputs

Inputs

E-statement PDFs, including password-protected files
Scanned or photographed statements
Account Aggregator FI data shared with consent
Credit policy metric definitions
Applicant's declared income and existing obligations

Outputs

Categorised transaction ledger per account
Credit-policy metrics with transaction drill-down
Bounce, EMI and obligation summary
Tamper and red-flag report
Underwriter corrections captured for model improvement

Where people stay in control

Underwriters review every metric and flag before it enters the credit memo, and the fraud control unit reviews integrity failures. The service never sets eligibility or limits; it provides verified evidence.

Integrations

Loan origination system
Account Aggregator ecosystem through the lender's FIU integration
Credit bureau data for obligation cross-checks
Fraud or risk case management system
Document management system

Guardrails

Metrics computed by deterministic code; the language model only categorises narrations
Integrity failures always routed to fraud control, never auto-rejected
Account Aggregator data used only for the consented purpose and duration
Statement data masked in logs and retained per policy
Category corrections reviewed before retraining

Compliance notes

Account Aggregator data is processed within the consent artefact's purpose and retention terms under RBI's NBFC-AA framework.
Supports data minimisation under the DPDP Act 2023 and need-based data collection under RBI digital lending rules.

A typical pilot

Duration: 3 weeks

Statements from historical self-employed loan files across five or six common banks, compared with underwriter workings, plus a set of known tampered statements.

Deliverables

01Metric definitions encoded from your credit policy
02Working analysis service on historical statements
03Underwriter workbench
04Accuracy report against underwriter workings
05Tamper detection results on known cases

What we measure

Parsing completeness: Parsed transactions and closing balances reconciled with statement totals for each file.
Metric agreement with underwriters: Computed metrics compared with underwriter workings for the same files, with differences explained.
Categorisation accuracy: Categories checked by underwriters on a sample of transactions per file.
Tamper detection on known cases: Known altered statements flagged or missed, and genuine statements falsely flagged.

FAQs

01Do you support Account Aggregator data?

Yes. AA data arrives as structured FI data, so parsing errors disappear and the same categorisation and metrics apply. PDFs remain supported for borrowers who do not use AA.

02Can it detect edited statements?

It checks PDF metadata, fonts, layout consistency and running balances, and flags statements that fail. These are signals for fraud control rather than proof, so every flag is reviewed by a person.

03Can we use our own credit policy definitions?

Yes. Metrics such as average balance dates, income netting rules and bounce definitions are configured from your policy and versioned when it changes.

04Does it replace our underwriters?

No. It removes manual parsing and counting so underwriters spend their time on judgement. Every figure is traceable, and they can correct categories when context calls for it.

Related use cases

Founded by Akhilesh Mittal. 8 years building a security platform for a New Zealand company. IIT ISM Dhanbad and IMT Ghaziabad. Builds production agentic AI.

Test one lending workflow on your own files

Book a discovery call to scope a short pilot, from statement analysis to a conduct-safe reminder agent, and judge it on measured results before scaling.