The AI credit underwriting copilot that shows its work.

It is the analyst level interface to an analysed file. After Fiscus has read the statements, categorised every transaction and computed the credit picture, Copilot lets anyone on the credit desk question that picture in plain language: is the coverage adequate, what changed in the last two quarters, what would worry you here. It augments the analyst. The decision, the overrides and the accountability stay with your team.

The distinction that matters is between a copilot and a chatbot. A chatbot talks about a document. A copilot works over a structured analysis: every number it quotes was computed by the same pipeline that produced the report, and every claim points back to its source.

The questions a committee actually asks. Two examples from a working capital case, condensed:

Both answers are auditable in the way committees need: the chips resolve to the exact statement page or computed cross check, not to a model explanation.

The failure mode every lender worries about with AI is a confident answer that is wrong. Fiscus treats that as a design constraint, not a disclaimer:

• Grounded scope. Copilot answers only over the analysed file: the categorised banking, computed metrics, fraud check results and bureau cross checks. It does not speculate past its sources.

• Citations on every claim. Each factual statement carries a reference to the statement page, metric or cross check behind it. Verification is a click, not an investigation.

• Auditable record. Questions, answers and evidence persist with the case, so what the committee relied on is reconstructable later, the way credit files are supposed to work.

Copilot is the last mile of the same pipeline that runs bank statement analysis and fraud detection. Analysts get the CAM ready report and the dashboard; the credit committee and everyone who does not live in spreadsheets gets Copilot. Nothing is auto decided: flags and answers reach people, with evidence, and the lender decides. After sanction the same data feeds monitoring through the life of the loan.

Entirely. The underlying analysis reads statements from 1000+ Indian banks in native PDF, scanned and Account Aggregator JSON form, applies segment specific logic for salaried, self employed and SME borrowers, computes FOIR the way Indian credit policies define it, and cross checks obligations against bureau data. Data residency is in India, deployment is in your environment, and handling follows DPDP principles ahead of enforcement.

The dashboard onboards within 48 hours with no integration, and unified REST APIs and webhooks connect your LOS in up to two weeks. The honest evaluation is the one on your own book: bring decided cases, ask the questions your committee asked, and compare.