← Agent catalogue

AI Agent Profile · LendingIQ · Singapore

Chief Risk Officer AI

Invoked via: internal orchestration APIRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens

DivisionRisk division

Resume

What this agent does

The CRO AI operates at the institutional level. It reads macro signals, constructs and oversees stress scenarios, articulates risk appetite, and produces board-quality risk reports and draft regulator communications. It prepares the risk intelligence and documentation that the CRO and Risk Committee need to set strategy, satisfy regulators, and govern the portfolio.

Primary functions

Risk Appetite Strategy

Annual cycle · Ad-hoc refresh

Invoked when: RAF review cycle opens OR macro environment shifts materially

  • Reads the current Risk Appetite Framework (RAF) document and portfolio MIS, then identifies where actual exposures sit relative to stated limits — by sector, product, geography, and borrower segment.
  • Synthesises macro signals (Regulator's rate stance, credit growth data, Banking sector stress indicators) and narrates what they imply for the institution's risk tolerance in the next 12 months.
  • Drafts revised RAF language — appetite statements, tolerance bands, and limit schedules — calibrated to the board's stated risk philosophy and current capital position.
  • Flags internal contradictions in the existing RAF: where a limit is structurally impossible to hit, or where two clauses conflict under stress.
Output: Revised draft RAF document with tracked changes, an executive summary of what changed and why, and a table of current-vs-proposed limits for board review.

Macro Signal Reading

Monthly · Triggered on RBI event

Invoked when: monthly MPC decision / NBFC circular / macro data release

  • Reads injected macro data — RBI MPC decisions, CPI, IIP, NBFC credit growth, NPA sector aggregates, GST collections — and narrates the credit cycle implication for the institution's portfolio mix.
  • Identifies which sectors, products, or borrower segments in the portfolio are most exposed to the macro shift, and quantifies the exposure by reading the portfolio MIS passed to it.
  • Does not access live data autonomously. Reads whatever structured data files are injected at invocation — the risk team controls what the agent sees.
  • Produces a ranked "watch list" of portfolio segments whose risk profile has worsened relative to the last reading, with the macro evidence behind each flag.
Output: Macro signal brief — 1-page narrative with key observations, portfolio segments on watch, and a recommended management action for each flagged segment.

Stress Test Oversight

Quarterly · Regulatory cycle

Invoked when: RBI stress test requirement / internal ICAAP cycle / board request

  • Takes stress scenario parameters defined by the risk team (e.g., 200 bps rate shock, 15% NPA migration in MSME, 20% haircut on collateral values) and applies them across the portfolio data injected into context.
  • Estimates the P&L and capital impact of each scenario in plain language — it reasons over the data rather than running a statistical model. Returns ranges, not point estimates.
  • Drafts the stress test narrative section of the ICAAP or board pack: scenario rationale, assumptions, results, management responses, and capital adequacy conclusion.
  • Flags where the stress results breach RAF thresholds and what management action the RAF requires in response.
Output: Stress test report — scenario definitions, impact analysis by segment, CRAR sensitivity, breach flags, and a draft management response section.

Board Risk Reports

Monthly / Quarterly board cycle

Invoked when: board pack preparation window opens

  • Synthesises the portfolio MIS, macro brief, stress outputs, and capital position into a board-format risk report — structured to the bank's standard template injected at invocation.
  • Writes in board-appropriate language: declarative, jargon-minimal, with clear "so what" statements. Not a data dump — a narrative with data support.
  • Generates the risk committee dashboard: key risk indicators (KRIs), RAG status against RAF limits, material changes since last board, and management actions in progress.
  • Can answer specific board questions in natural language — "What is our concentration in real estate above ₹10 Cr?" — by reasoning over the data passed to it.
Output: Board risk report in the bank's prescribed format — executive summary, KRI dashboard, portfolio narrative, stress summary, outlook, and appendix data tables.

Regulator Liaison Support

On-demand · Inspection cycle

Invoked when: RBI inspection query / NBFC supervisory letter / ad-hoc regulator request

  • Reads the regulator's query or inspection observation and maps it to the relevant RBI Master Direction, IRACP norm, or scale-based regulation — citing the exact circular and clause.
  • Drafts a structured response: acknowledgement of the observation, the institution's current position, corrective actions taken or planned, and the timeline for compliance.
  • Cross-references the institution's data (injected into context) against what the regulator is asking about — identifies any gap between what the regulator expects and what the data shows.
  • Does not sign or submit responses. Produces a draft for review, editing, and formal sign-off by the CRO and Compliance team.
Output: Draft regulator response — structured per the query, with regulatory citations, factual position, corrective action plan, and a compliance timeline.

Knowledge base

RBI Master Directions & Circulars

IRACP norms, NBFC scale-based regulation, PCA framework, ICAAP guidelines. Loaded as grounding corpus; current circulars injected via RAG.

Risk Appetite Framework v3.1

Institution's live RAF document, retrieved via RAG at invocation. Always reads the current version — not a cached summary.

Basel III / IV & ICAAP Framework

Capital adequacy (CRAR), LCR, NSFR, Pillar 2 requirements, stress testing norms. Applied when assessing capital impact of scenarios.

Macro Credit Cycle Knowledge

Pre-training knowledge of Indian macro indicators, NBFC credit cycles, RBI policy transmission, and sector-level stress patterns.

Sector Benchmarks

Industry NPA rates, leverage norms, and concentration benchmarks for MSME, real estate, agri, and services — used to contextualise portfolio data.

Live Portfolio Data (Injected)

MIS extracts, capital reports, CBS outputs — passed at run-time. Not stored. Not recalled across sessions. The risk team controls the data feed.

Hard guardrails

Will notSet or approve the Risk Appetite Framework unilaterally. It drafts and analyses — the board approves the RAF.
Will notSubmit any regulator communication. All responses are drafts for CRO and Compliance sign-off before submission.
Will notProduce actuarially calibrated PD/LGD numbers for regulatory capital purposes. It provides directional stress estimates — use validated models for ICAAP capital submissions.
Will notRetain portfolio data between sessions. Each invocation starts fresh — no memory of prior MIS runs, prior stress tests, or prior board packs.
Will notAccess live data autonomously. It reads only what the risk team injects into its context — data governance remains with the institution.

Known limitations

Stress impact estimates are directional, not actuarially validated. The agent applies scenario parameters to portfolio data using reasoned approximation — not a calibrated credit risk model. Use a validated PD/LGD engine for regulatory capital submissions.Treat stress outputs as management intelligence, not regulatory-grade modelling.
Macro knowledge has a training cutoff. The agent may not know about RBI circulars issued after its training date unless the document is injected into context. Always supply the current circular via RAG for regulatory drafting tasks.Never rely on the agent's baked-in regulatory knowledge for compliance-critical outputs.
Board report quality depends entirely on MIS quality. The agent cannot detect errors in the data it is given — it will reason confidently over incorrect inputs. Upstream data validation is the institution's responsibility.Garbage in, board-quality garbage out applies fully.
Cannot read scanned or image-format documents. All MIS, policy documents, and regulatory letters must be machine-readable text. Scanned inputs must be OCR-processed before passing to the agent.Invest in an upstream OCR layer for document ingestion from regulators.
May be over-literal on RAF limit language. Ambiguously worded appetite statements produce ambiguous breach assessments. Precise RAF drafting produces precise monitoring.Invest in clean, unambiguous RAF language — the agent cannot infer intent from vague limits.
Agent Profile · Chief Risk Officer AI · LendingIQ · BengaluruLast updated May 2026 · For internal use

Important Reads

Learn more about how to deploy Chief Risk Officer AI to your lending workflow.