Risk Reporting Agent AI

Invoked when: daily data refresh is complete or a risk event threshold is breached

• –Assembles the daily CRO monitoring dashboard from the live data feeds: portfolio NPA rate and PAR-30, segment concentration versus limits, capital adequacy ratio, liquidity coverage ratio, early warning account count by tier, and model drift alerts. Each metric is shown with its trend (7-day and 30-day), its position relative to the risk appetite limit, and a RAG status — so the CRO AI can scan the dashboard in under two minutes and identify which metrics require attention.
• –Generates a trigger report immediately when any metric breaches a risk appetite limit — not at the next scheduled reporting cycle, but in real time when the breach occurs. The trigger report identifies the breached metric, the magnitude of breach, the last time it was within limit, and the contributing factors visible from the data. The CRO AI receives the trigger report and decides what management action is required.
• –Does not interpret what a breach means strategically or recommend management action. Threshold breaches are reported; the CRO AI assesses and the human CRO decides the response.

Invoked when: monthly Board Risk Committee reporting cycle is due, typically 5 working days before the committee meeting

• –Assembles the Board Risk Committee pack from the month's risk data, stress testing results, provisioning schedule, model risk register updates, and regulatory compliance status — formatting it into a board-ready document that follows the approved pack structure: executive risk summary, portfolio performance section, capital and liquidity section, model risk section, compliance section, and management action tracking from prior meetings.
• –Produces the executive risk summary — a 1-page narrative that tells the board the month's risk story: what has changed materially in the portfolio, which metrics have moved toward or away from risk appetite limits, what the stress tests show about portfolio resilience, and whether there are any matters that require board-level decisions or awareness. The narrative is drafted by the agent and reviewed by the CRO AI before distribution — the CRO AI may add qualitative context or strategic interpretation that the data alone cannot provide.
• –Tracks management action items from prior board risk meetings — whether actions committed at the last meeting have been completed, what their current status is, and where items are overdue. The action tracker is a governance accountability mechanism: the board's decisions are logged and their implementation is reported back to the board at the next cycle.

Invoked when: quarterly filing cycle is due or an ad-hoc regulatory return is required

• –Reads the current regulatory return formats and validation rules from the RBI reporting schema (RAG) — always the live version, because RBI periodically revises return formats and validation logic. Assembles the required data for each return — NBS-7 (financial data for NBFCs), ALM return (asset-liability maturity profile), ICAAP submission (capital adequacy assessment) — from the source systems, populates the return template, and runs the validation suite to check that all figures are internally consistent and agree with the source system data.
• –For each regulatory return, produces a pre-filing validation report: every data field in the return with its source, the source system value it was drawn from, and the validation check result (pass/fail). Where any field fails validation — the return figure does not agree with the source or fails an internal consistency check — the return is held and the discrepancy flagged to the human CRO. No return is submitted with a known validation failure.
• –Tracks filing deadlines — the XBRL submission window, the authorised signatory requirement, and the submission confirmation — and generates a filing status report showing which returns have been submitted, which are pending, and which are approaching their deadline. The human CRO authorises all regulatory filings before submission. The agent prepares and validates; the authorised official submits.