Part 09 — Data & Integrations

Bank Statement Analysis

What a bank statement analyser really does

Bank statement analysis (BSA) is the workhorse of Indian SME underwriting because bank conduct often reveals what tax returns and unaudited financials hide. A statement analyser converts PDFs, net-banking pulls, CSVs or Account Aggregator (AA) data into a structured ledger, then builds underwriting variables: cash-flow stability, average bank balance (ABB), bounce history, obligation load, counterparty concentration, circular transactions, cash intensity and fraud indicators.

Named vendors in India include Perfios and FinBox, alongside bureau/vendor stacks that embed similar capability. Perfios describes its document analysers as extracting, curating and analysing data from AA, bank statements, audited financials, ITR, payslips and GST, with reports shared into LOS/LMS systems and built-in fraud checks for manipulated PDFs or transactions (Perfios document analysers). FinBox BankConnect describes itself as a bank statement analyser integrated with AA and says it detects borrower salary, obligations, recurring expenses and 500-plus parameters, with 150-plus fraud checks (FinBox BankConnect). FinBox’s developer documentation says enriched data can be fetched through APIs after statement upload, including transactions and identity details (FinBox BankConnect docs).

Ingestion and parsing

The first job is not analytics; it is producing a reliable transaction table. Inputs arrive as digitally generated PDFs, scanned PDFs, downloaded Excel/CSV, net-banking statements, AA JSON/XML, or partner settlement data. The parser must extract account holder name, account number, IFSC, bank name, branch, statement period, opening balance, closing balance and every transaction date, narration, debit, credit and balance.

Quality checks start immediately:

Are all pages present?
Does the running balance reconcile after every transaction?
Does opening plus credits minus debits equal closing balance?
Are fonts, metadata, page objects or row positions inconsistent with that bank’s template?
Does the account number/name match the applicant or declared business?
Are there unexplained gaps between statement periods?

Perfios says its products support more than 4,000 document formats from more than 1,000 financial institutions and include built-in fraud checks (Perfios document analysers). That scale matters because Indian current-account statements are messy: bank-specific narrations, merged cells, vernacular descriptors, debit/credit reversals, value date versus transaction date, password-protected PDFs and current-account OD limit displays.

AA changes ingestion quality but not all interpretation. AA data avoids PDF tampering, but it still carries raw bank narrations. Counterparty classification, self-transfer detection and business-income judgement remain lender logic.

Core variables lenders extract

Average bank balance (ABB) is the simplest liquidity proxy. Lenders calculate monthly ABB from daily closing balances where available, or from periodic snapshots where policy permits. Some market tools mention the six-date method using balances on the 1st, 5th, 10th, 15th, 20th and 25th, but that is a policy convention and not a universal RBI formula. For SME current accounts, minimum balance, month-end balance and negative/overdrawn days are often more predictive than plain average.

Turnover and inflow quality separates operating credits from noise. Good credits include customer receipts, card/UPI settlements, NEFT/RTGS from buyers, marketplace settlements and anchor payments. Weak credits include self-transfers, cash deposits without business explanation, round-number related-party transfers, loan disbursements, reversals, cheque returns, GST refunds and temporary accommodation entries. For merchant loans, daily settlement consistency matters more than total credits.

Obligation detection identifies EMIs, NACH debits, standing instructions, credit-card auto-debits, gold-loan renewals, app-lender collections and informal finance proxies. FinBox says its analyser gets insights on obligations and recurring expenses (FinBox BankConnect). The analyst should compare detected EMIs with bureau tradelines and borrower declarations. Missing obligations are common when the borrower pays from another account.

Bounce and return behaviour is a key conduct marker. The analyser tags inward cheque returns, outward cheque returns, NACH mandate failures, ECS returns, UPI autopay failures, charges for insufficient funds and reversal entries. One customer cheque returned due to drawer error is different from repeated outward NACH bounces on existing loans. Count, amount, reason and recency all matter.

Cash intensity measures cash deposits and withdrawals as a share of total credits/debits. High cash deposits can be normal for kirana, restaurants, transport and small retail. It is suspicious for B2B manufacturing or IT services unless explained. Cash withdrawal spikes soon after loan disbursement can indicate diversion.

Counterparty concentration lists top payers and payees. A borrower with 70 percent of credits from one anchor is exposed to anchor termination but may be suitable for supply-chain finance. A borrower with many small UPI credits may be granular but operationally harder to reconcile with GST.

Tax and statutory payments identify GST, TDS, EPFO, ESIC, professional tax and income-tax challans. Absence of statutory payments is not always fatal for micro units, but it should align with employee count, GST registration and financial statements.

Circular transactions and balance dressing

Circularity is the classic BSA trap. The same money moves from Account A to related Account B and returns as a fresh credit, inflating turnover. The analyser looks for matching amounts, near-matching amounts, repeated counterparties, same-day or next-day reversal loops, round figures, narration similarity and group-account names. In SME cases, circularity can be benign treasury movement across branches, but it must be deducted from eligible turnover unless the business purpose is clear.

Balance dressing is a related pattern. The borrower brings funds into the account just before month-end or before statement submission, keeps balances high for a few days, then withdraws them. A good analyser reports day-wise balance trends, minimum balances, peak balances, end-of-month spikes and large one-off credits. The underwriter should ask for source evidence when the largest credit in six months appears two days before the application.

FinBox publicly says its fraud checks include document fingerprinting, accounting-discrepancy checks by tallying starting, ending and running balances, and suspicious activity such as round-tripping of funds to temporarily inflate balances (FinBox BankConnect FAQ). That is exactly the control a credit team needs, but the final conclusion should remain human-reviewed for higher-ticket SME loans.

Surrogate income and eligibility

For unsecured business loans, BSA often feeds banking-surrogate eligibility:

eligible monthly turnover = operating credits minus self-transfers, loan credits, reversals and abnormal one-offs;
gross margin or net margin proxy = policy percentage by industry;
assessed monthly income = eligible turnover multiplied by margin;
obligation capacity = assessed income multiplied by allowed fixed-obligation-to-income ratio (FOIR), then reduced by detected EMIs;
loan amount = affordable EMI converted to principal using product rate and tenor.

For working capital, BSA supports drawing-power discipline and conduct rather than term-loan EMI capacity. Analysts look at utilisation, cheque returns, inward collections, supplier payments, inventory purchase cycles, GST dates and whether bank credits support the requested limit.

For loan against property (LAP), BSA helps prove repayment capacity, but collateral valuation and legal title remain primary. A LAP file with weak bank conduct may still be approved at lower loan-to-value, more guarantors or escrowed rentals, but the deviation should be explicit.

Vendor output and integration

The BSA vendor usually returns three layers:

Raw parsed data: account profile and transaction ledger.
Derived variables: ABB, monthly credits/debits, EMI count, bounce count, cash percentage, top counterparties, fraud flags, salary/business income, GST/tax payments and recurring obligations.
Report pack: PDF/Excel/JSON summary, often with visual charts and lender-specific scorecards.

In architecture terms, BSA sits between the data capture layer and the business rules engine (BRE). The LOS sends a statement file or AA reference, the analyser returns parsed JSON and report URI, the BRE consumes selected variables, and the credit analyst sees both machine tags and raw transactions. For audit, store the source file hash, parser version, ruleset version, exceptions raised, analyst edits and final variables used in eligibility.

Do not let the BSA output become a black box. Credit policy should define which tags can drive auto-decisioning and which require analyst review. For example, “three outward bounces in last three months” may be an automated decline for unsecured loans, while “cash deposit share above 50 percent” may require industry explanation.

Edge cases in Indian SME statements

Many borrowers operate multiple bank accounts. One account may receive sales while another pays EMIs. Ask for all primary current accounts and loan repayment accounts. AA can improve account discovery only where the borrower links the relevant FIP accounts.

Proprietors mix household and business transactions. Rent, school fees and personal transfers may appear beside supplier payments. The analyst should not pretend there is clean corporate separation; instead, compute sustainable surplus after personal drawings.

GST and banking rarely match exactly. Differences arise from credit sales, cash sales, exempt sales, multiple GSTINs, inter-branch transfers, delayed collections, refunds and invoice cancellations. The task is reconciliation, not forced equality. Cross-link this with underwriting data and cash-flow assessment.

Finally, beware perfect statements. No cash withdrawals, no bank charges, perfectly round daily credits and no failed transactions over twelve months can be genuine for a disciplined digital merchant, but in many small-business contexts it deserves a tamper and completeness check.