What AI governance obligations do RBI's Digital Lending Guidelines 2022 create for algorithmic underwriting?

RBI Circular RBI/2022-23/111 (September 2, 2022) — Guidelines on Digital Lending — applies to all RBI-regulated entities (REs) engaging in digital lending: commercial banks, co-operative banks, NBFCs, NBFC-P2P platforms, and NBFC-Account Aggregators. It creates four specific AI governance obligations for algorithmic credit underwriting: (1) Auditability of credit algorithms: algorithms used by REs and their Lending Service Providers (LSPs) for credit scoring must be auditable — able to demonstrate minimum underwriting standards and surface potential discrimination factors in credit pricing. RBI did not define 'auditable' but the inspection context makes clear that an AI credit model whose decisions cannot be explained to RBI examiners fails this standard; (2) Non-discriminatory training data: credit scoring algorithms must be trained on diverse, non-discriminatory data. This is India's first RBI requirement for training data quality in AI credit models — creating an obligation to document training data composition, demographic representation, and bias testing outcomes; (3) RE accountability for LSP AI: when an RE uses a Lending Service Provider (LSP) — a fintech partner — for digital loan origination, the RE bears full accountability for AI decisions made by the LSP's platform. An RE cannot treat LSP AI as outside its regulatory perimeter just because the LSP built and operates the model; (4) Key Fact Statement (KFS) explainability: the KFS must be provided before loan execution and must include the annual percentage rate (APR) and all fees in a standardized format. Where AI determines the interest rate and fee structure through dynamic pricing algorithms, the KFS must reflect those AI-determined terms — creating an implicit explainability obligation: the RE must be able to explain why the AI assigned a specific rate to a specific borrower.

How does the LSP accountability model under RBI Digital Lending 2022 create AI governance obligations for bank-fintech partnerships?

India's digital lending ecosystem operates through a bank-fintech partnership model: banks and NBFCs (the Regulated Entity) partner with LendingTech fintechs (Lending Service Providers) who provide the digital origination platform, credit scoring AI, and customer-facing app. The LSP-RE accountability structure under RBI's 2022 guidelines creates a specific AI governance challenge: the RE bears regulatory accountability for AI it does not own, operate, or have full visibility into. Four governance obligations follow from this structure: (1) LSP AI in RE's model inventory: the RE cannot exclude LSP-operated AI models from its own model risk governance. RBI's Digital Lending Guidelines make the RE accountable for LSP AI decisions — the RE must therefore govern LSP AI as if it were internal, which requires including it in the model inventory, conducting or commissioning validation, and monitoring ongoing performance; (2) Contractual AI governance rights: an RE that does not contractually secure the right to audit the LSP's credit scoring model, receive model change notifications, and obtain validation evidence cannot satisfy its accountability obligation. Standard fintech partnership agreements typically do not include these provisions — the 2022 guidelines effectively require RE legal teams to retrofit them; (3) Fair lending AI monitoring: if an LSP's credit scoring AI systematically denies credit to certain demographics (low-income borrowers, women, rural borrowers), the RE faces the regulatory liability under RBI's non-discrimination requirements — not the LSP. The RE must monitor LSP AI outcomes for disparate impact; (4) KFS accuracy for AI-dynamic pricing: if the LSP's AI dynamically prices loans based on real-time risk assessment, the RE must ensure the KFS generated for each borrower reflects the AI's specific pricing for that borrower — not a generic rate range. AI-generated KFS discrepancies create consumer protection violations attributable to the RE.

What borrower profiling AI consent requirements arise from RBI's Digital Lending Guidelines 2022?

RBI Digital Lending Guidelines 2022 Para 7.1 requires capture of the economic profile of the borrower before loan disbursal — explicitly requiring documentation of income, assets, liabilities, and financial behavior. In digital lending, this economic profiling is AI-driven: income inference from bank statement analysis, employment inference from transaction patterns, spending behavior scoring, and alternative data aggregation (social media, e-commerce, utility payments). The AI-driven nature of borrower profiling under the 2022 guidelines creates specific consent and governance obligations under both the digital lending framework and the DPDP Act 2023: (1) Consent architecture for AI profiling: digital lending AI that accesses alternative data sources beyond the standard credit bureau requires explicit, purpose-specific consent under DPDP Act Section 6. A consent given for loan processing cannot automatically extend to AI behavioral profiling for creditworthiness — especially where the profiling draws on data from sources the borrower may not know are accessed (e-commerce purchase history, utility payment patterns); (2) Minimum data principle for AI models: RBI's requirement to capture economic profile 'for the purpose of extending credit' — combined with DPDP Act Section 8(2) data minimisation — means digital lending AI cannot harvest unlimited alternative data simply because more data improves model accuracy. The data footprint must be justified by the credit assessment purpose; (3) AI profile transparency: when a digital loan application is declined based on an AI-generated economic profile, the borrower has a right under RBI's customer service guidelines and DPDP Act to understand what data was used and how. AI credit profiling models that use unexplained alternative data features (device type scores, social graph metrics, app usage patterns) face a transparency deficit that RBI inspections increasingly surface; (4) Account Aggregator intersection: where AI-driven profiling uses data obtained through the AA framework, the consent obtained through AA for data sharing must cover the specific AI processing purpose — AA consent given for 'financial analysis' may not extend to AI behavioral scoring.

What AI governance gap exists between RBI's Digital Lending Guidelines and actual NBFC-fintech practice in India?

India's digital lending market includes over 1,000 NBFCs, hundreds of fintech lending platforms, and a rapidly growing NBFC-P2P segment — most operating AI credit underwriting through LSP partnerships without formal AI governance frameworks. The gap between the 2022 guidelines' accountability requirements and actual practice is acute in four areas: (1) RE accountability awareness: many small and mid-size NBFCs operating through LSP partnerships have not operationalized the accountability implication — they treat the LSP's credit AI as the LSP's responsibility. RBI inspections post-2022 have increasingly cited REs for inadequate oversight of LSP AI systems; (2) KFS-AI consistency: dynamic AI pricing at NBFCs generates millions of borrower-specific KFS documents daily. Many digital lending platforms generate KFS from their core lending system (which may not reflect real-time AI pricing adjustments) rather than directly from the AI pricing engine — creating systematic KFS accuracy gaps; (3) Alternative data governance: many digital lending AI models use alternative data (device signals, app metadata, behavioral biometrics) that was never subject to formal data governance or consent architecture design. DPDP Act enforcement, when it begins, will likely surface consent architecture gaps at digital lenders; (4) Model drift in LSP AI: LSP credit scoring models are retrained periodically as lenders accumulate data. An RE that does not have contractual rights to receive notification of LSP model retraining will not know when the credit scoring logic has materially changed — and will not be able to conduct its own assessment of whether the retrained model still meets RBI's non-discrimination standards.

AI Governance · RBI · Digital Lending · Algorithmic Underwriting · LSP · India BFSI

RBI Digital Lending Guidelines 2022 AI Governance: Algorithmic Underwriting Accountability, Lending Service Provider AI Liability, Key Fact Statement Explainability and Borrower Profiling AI Consent India

RBI Circular RBI/2022-23/111 (September 2, 2022) — Guidelines on Digital Lending — applies to all RBI-regulated entities in India's $200B+ digital lending market: banks, NBFCs, NBFC-P2P, NBFC-AA. Credit scoring algorithms must be auditable, trained on non-discriminatory data, and any AI operated by Lending Service Provider (LSP) fintech partners is the RE's full regulatory liability. Key Fact Statement explainability for AI-dynamic pricing, borrower profiling consent under DPDP Act. Zero global competitor coverage of any RBI digital lending regulation.

Zytra Research Team · September 2022 · 11 min read

📖 Part of the India BFSI AI Governance Complete Guide 2026 — RBI, SEBI, IRDAI, NPCI & DPDP in one place.

India's digital lending market disburses over ₹3 lakh crore annually through platforms built on AI underwriting — PhonePe Lending, MoneyView, KreditBee, Slice, CASHe, Stashfin, and hundreds of NBFC-fintech partnerships. Every loan decision in this market is, at some level, an AI decision. RBI's Digital Lending Guidelines 2022 (RBI/2022-23/111) were the first RBI circular to directly address algorithmic underwriting accountability — requiring credit algorithms to be auditable, trained on diverse and non-discriminatory data, and held to full RE accountability even when operated by an LSP fintech partner.

The governance gap is operationally significant: most digital lenders deployed their AI credit models before the 2022 guidelines were issued, without the governance infrastructure those guidelines now require. The LSP-RE accountability model creates particular urgency — an NBFC that licensed a fintech's credit scoring platform in 2021 and has not renegotiated governance rights into the contract is operating an AI system it is fully accountable for but cannot adequately govern.

The LSP Problem: Accountable for AI You Don't Own

RBI's digital lending framework creates an unusual AI governance challenge: the Regulated Entity is fully accountable for AI decisions made by an LSP's platform, but the RE typically has no direct visibility into the LSP's model architecture, training data, validation history, or change management. Satisfying RBI's accountability requirement for LSP AI requires contractual AI governance rights that most current bank-fintech agreements do not contain — model audit rights, change notification obligations, validation evidence sharing, and disparate impact monitoring rights.

Digital Lending AI Governance Checklist

☐ LSP AI contracts: model audit rights, change notification, validation evidence, disparate impact monitoring rights — all contractually secured
☐ Credit algorithm auditability: model documentation sufficient for RBI inspection; decision rationale reconstructable for any loan outcome
☐ Non-discriminatory training data: data composition documented; demographic representation assessed; bias testing on India-relevant segments (gender, geography, income tier)
☐ KFS-AI consistency: KFS generated directly from AI pricing engine; dynamic pricing reflected per-borrower; KFS accuracy audit trail
☐ Borrower profiling consent: DPDP Act-compliant consent for alternative data AI processing; purpose limitation for credit assessment vs behavioral scoring; AA framework consent scope
☐ LSP AI in model inventory: all LSP credit AI models in RE's model inventory; risk tiered; validation status tracked
☐ Model drift monitoring: LSP retraining notification protocol; post-retraining bias assessment; RBI notification assessment for material model changes
☐ RBI examination readiness: LSP contract AI governance provisions, algorithm auditability evidence, training data documentation, bias testing results, KFS accuracy audit — packaged for RBI inspection