Bank statement fraud detection built into the analysis.

Banks reported 10,114 fraud cases worth Rs 48,021 crore to RBI in FY 2025-26, with advances related fraud dominating by value. The headline value needs a caveat: around Rs 30,199 crore of it came from legacy cases added back after a 2023 Supreme Court judgement, so the jump is mostly reclassification rather than new fraud. The underlying direction is still clear, and the bank statement sits at the start of it: it is usually the first document in a loan file, and with consumer PDF editors and generative tools it has never been easier to fabricate one that looks right.

Manual review catches the crude fakes. The convincing ones are arithmetic problems and pattern problems, and those need machines.

Before analysis begins, every document passes forensic checks and receives a tampering verdict:

• Balance trail reconciliation. Opening balance plus credits minus debits must equal every running balance on every page. Broken arithmetic is the single most reliable tamper signal, and Fiscus points to the exact page where the trail breaks.

• Metadata analysis. Producer and creator fields naming editing software instead of the bank generator, modification dates that postdate creation, and generation patterns inconsistent with the claimed download.

• Font and layout anomalies. Overlaid text in a different typeface, spacing or rendering than the rest of the document, the classic signature of a figure edited after the fact.

• Template tamper validation. Document structure checked against the known layout of the claimed bank, with digital signature mismatches flagged. Fiscus knows the formats of 1000+ Indian banks, including co-operative and regional banks.

A statement can be genuine as a document and still describe managed money. Fraud check units run rule sets over the transactions themselves, tuned per borrower segment because fraud looks different in a salary account than in a business current account:

• Circular flows. The same counterparty on both debit and credit sides, funds exiting shortly after entry, and high throughput with the net balance unchanged.

• Window dressing. Round figure credits clustering just before period ends, balances propped on statement dates, and deposits timed to the review calendar.

• Velocity anomalies. UPI and NEFT day limits exceeded, cash and cheque activity on holidays, months with abnormally low transaction counts, and dormant accounts that suddenly activate before an application.

• Segment tuned rules. For salaried profiles, a large debit immediately after every salary credit. For micro businesses, round figure credits above threshold. For SME accounts, nil credits between the 20th and the 5th while claiming steady trade.

The strongest detection layer never looks at the statement in isolation. Fiscus cross examines the banking against independent records:

• Bureau against banking. EMI outflows matched to bureau tradelines in both directions: recurring debits with no tradeline expose undisclosed borrowing, and tradelines with no matching outflow expose closed or disputed accounts still being claimed.

• Declared turnover against banked credits. For business borrowers, the companion GST view benchmarks GSTN declared turnover against revenue actually landing in the accounts, with variance computed and explained. Inflated banking rarely survives that comparison.

• Income consistency. Salary credits checked for regularity and against employment benefit flows such as EPFO, so a fabricated salary line has to fake an entire ecosystem, not one row.

A flag is evidence, not a verdict. Each one carries the rule that fired, the exact page and the values involved, so an analyst can adjudicate in seconds instead of rereading a 24 month file. Nothing is auto declined: unusual but genuine borrowers exist, and the decision stays with your credit team, with the full trail preserved for audit. During evaluation, teams typically run Fiscus in parallel on cases they have already reviewed to calibrate the rules against their own book.

The RBI Master Directions on Fraud Risk Management of July 2024 require commercial banks and NBFCs in the upper and middle layers to operate board approved frameworks for Early Warning Signals and Red Flagging of Accounts, integrated with core systems and supported by data analytics, with indicators approved by the risk management committee of the board. Statement level screening produces exactly the transactional indicators those frameworks run on, with the evidence trail regulators expect. The same data continues to work after sanction, feeding continuous loan monitoring on the live account.

A careful analyst can spot a mismatched font. No analyst can reconcile every running balance across 11 accounts and 24 months, compare every EMI debit to every tradeline, and apply forty pattern rules identically on the five hundredth file of the month. That is machine work, and doing it by machine is what returns analyst time to judgement: lenders using Fiscus cut per case analyst time by 50% or more, with every check applied the same way every time. The screening is part of the same analysis pass that produces the credit picture, not a separate tool bolted on.