Underwriting analysis a bank can put in front of an auditor.

Three pressures at once. Turnaround expectations have collapsed from days to hours while files got thicker: multi account, multi bank, 24 months deep. Fraud has industrialised, with consumer PDF editors and generative tools making convincing fakes cheap, and the July 2024 RBI Master Directions raising what supervisors expect lenders to detect. And every number a bank acts on now has to be explainable later, to committee, audit and inspection.

Manual assembly cannot meet all three at once, and parser era tools stop at extraction. The gap is an analysis layer that is fast, forensic and auditable at the same time.

• Statement analysis at committee grade. 95%+ categorisation accuracy validated over 200,000 transactions, segment specific logic for salaried, self employed and SME borrowers, CAM ready output your analysts do not rework.

• Fraud screening before analysis. A tampering verdict on every document and segment tuned transaction rules, every flag citing the exact page. Flags reach analysts with evidence; nothing is auto declined.

• Bureau cross checks. EMI outflows matched against tradelines in both directions, exposing undisclosed obligations and stale tradelines before sanction. A companion GST view benchmarks declared turnover against banked credits for business files.

• Evidence cited answers. Copilot lets the committee question any case in plain language, with every claim citing its statement page or cross check.

As a layer, not a rip and replace. Analysed output flows into your LOS through unified REST APIs and webhooks, in Excel or JSON, in up to two weeks of integration work. Your BRE keeps applying your credit policy to better inputs, maker checker flows stay intact, and report fields and categorisation rules are configurable to your policy. Teams that want zero integration start on the browser dashboard, onboarded within 48 hours.

This is where Fiscus is built differently. Every figure in every report carries the page, position and confidence it came from, so provenance is a property of the data rather than a reconstruction exercise. Fraud flags carry the rule and the page. Copilot answers carry citations. When the RBI Master Directions on Fraud Risk Management ask for Early Warning Signals and Red Flagging frameworks that are integrated and data driven, statement level screening and continuous monitoring supply the transactional indicators with the evidence attached.

The same analysis keeps working through the life of the loan: inflow and balance deterioration flagged month on month, new leverage spotted in the banking before it reports to the bureau, alerts ranked by exposure and severity, and collections queues ranked by capacity to pay. Underwriting, monitoring and collections read from one picture instead of three vendors.

Deployment is in your environment. Data residency is in India. Handling follows DPDP principles ahead of the May 2027 enforcement. Evaluation runs in parallel with your current process on live cases, so nothing changes in production until your reviewers have compared the output on your own book. The full capability detail lives on the bank statement analysis, fraud detection and Copilot pages.