What did RBI's August 2024 amendment to NBFC-P2P Directions require for AI matching algorithms?

The Reserve Bank of India issued a significant amendment to the NBFC–Peer to Peer Lending Platform (Reserve Bank) Directions, 2017 (DNBR(PD)CC.No.078/03.10.124/2017-18) on August 16, 2024. Among the key changes, the amendment requires NBFC-P2P platforms to: (1) adopt a board-approved lender-borrower matching algorithm that is non-discriminatory; (2) ensure the matching algorithm does not preference borrowers from affiliated entities, co-branded channel partners, or connected parties (the closed-user-group prohibition); (3) prohibit guaranteed returns or assured yield marketing — P2P platforms cannot use AI-generated performance projections to market assured returns to lenders; (4) ensure T+1 fund deployment (lender funds deployed to borrowers within one business day of lender commitment). The non-discriminatory matching algorithm requirement is unique globally — no equivalent obligation exists in EU, US, or UK P2P regulatory frameworks. It creates a board-level AI governance obligation: the matching algorithm must be approved at board level, documented as non-discriminatory, and testable as such during RBI examination.

How does MRMF apply to P2P credit assessment AI models?

RBI MRMF June 2026 applies to all AI/ML models used in credit decisions by RBI-regulated entities. NBFC-P2P platforms are RBI-regulated entities (classified as NBFCs under the 2017 Directions). Their credit assessment models — scoring borrower creditworthiness to determine matching eligibility and risk-based pricing — are material credit AI models under MRMF. MRMF obligations for P2P credit AI: (1) Model inventory: every credit assessment model used in borrower evaluation must be in the platform's MRMF model inventory with a model card documenting purpose, scope, data sources, performance metrics, and validation status; (2) Independent validation (Section 3): the model must be validated by a team independent of the model development team; for a small NBFC-P2P, this may require commissioning external validation; (3) Performance monitoring (Section 6): ongoing PSI/CSI monitoring for the credit assessment model; MRMF PSI threshold above 0.25 requires model review; for P2P platforms, monitoring must cover both the overall borrower population and the specific borrower segments that the matching algorithm routes to different lender categories; (4) Third-party model governance (Section 7): if the P2P platform uses a credit bureau AI score or a third-party credit assessment model, MRMF Section 7 applies — the third-party model must be in the inventory, independently validated, and monitored. The MRMF obligation creates a documentation package that also partially satisfies the 2024 amendment's non-discriminatory algorithm disclosure obligation, since MRMF documentation includes demographic performance analysis.

What is the closed-user-group prohibition and how does it create an AI origination-channel independence obligation?

The 2024 amendment prohibits NBFC-P2P platforms from operating as closed-user-group platforms — arrangements where borrowers are sourced from affiliated entities or co-branded channel partners and given preferential matching. The prohibition prevents a scenario where a fintech's P2P platform preferentially matches loans to borrowers from its own lending or NBFCs in the same corporate group. For AI matching algorithms, the closed-user-group prohibition creates an origination-channel independence requirement: the matching algorithm must not use borrower origination channel (where the borrower was acquired from, which partner referred them) as a feature that influences matching priority or credit score. An AI model that was trained on historical matching data where affiliated-entity borrowers were systematically matched faster or at better rates will have learned a proxy for the closed-user-group preference even if origination channel is not an explicit feature — related entities' borrowers may share other characteristics that the model uses as proxies. Governing this requires: (a) feature importance audit to confirm origination channel is not used as a direct or proxy feature; (b) backtesting to compare matching outcomes for affiliated-entity borrowers versus open-market borrowers with equivalent credit profiles; (c) documentation in the model card that the non-discriminatory matching obligation has been tested and confirmed; (d) ongoing monitoring of matching rate disparities by borrower origination source.

What are the T+1 disbursement pipeline governance obligations for P2P AI?

The 2024 amendment requires NBFC-P2P platforms to deploy lender funds to borrowers within T+1 (one business day) of the lender commitment. For AI-driven P2P platforms where credit assessment and matching are automated, the T+1 requirement creates a service-level obligation for the AI decision pipeline. If the credit assessment AI experiences downtime, elevated latency, or degraded performance that prevents timely matching decisions, lender funds are held beyond T+1 — creating regulatory non-compliance and potentially consumer protection exposure (lenders losing opportunity cost on delayed deployment). AI governance obligations arising from T+1: (1) SLA documentation for the credit assessment and matching AI pipeline — maximum acceptable latency per decision, average throughput capacity, and escalation process when batch volumes exceed normal capacity; (2) Fallback governance: if the primary AI credit assessment model is unavailable, what fallback process applies? The fallback must also comply with T+1 — a manual review fallback that cannot meet T+1 at scale is not an adequate contingency; (3) Drift-triggered review governance: when a PSI threshold triggers a model review under MRMF, the review must be completed without extending T+1 timelines for the affected period — the model must either be approved for continued use pending full review or a T+1-capable fallback must be activated; (4) Incident reporting: T+1 failures caused by AI system issues must be documented and disclosed to RBI as operational incidents, consistent with RBI's IT Framework Master Direction incident reporting obligations.

AI Governance · RBI P2P Lending · NBFC-P2P · Credit Matching · MRMF · India BFSI

RBI NBFC-P2P Lending Directions: AI Credit Matching Algorithm Governance, Non-Discriminatory Matching Obligation and Model Risk Under the August 2024 Amendment

RBI's August 2024 amendment to NBFC-P2P Lending Directions introduced the only board-approved non-discriminatory AI matching algorithm obligation in global P2P regulation. Combined with MRMF's application to P2P credit AI, the closed-user-group prohibition's origination-channel independence requirement, T+1 disbursement AI SLA governance, and the guaranteed-return prohibition — India's P2P AI governance is uniquely demanding. Zero global competitor coverage.

Zytra Research Team · October 2026 · 12 min read

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

India's Peer-to-Peer (P2P) lending market operates under the NBFC–P2P Lending Platform (Reserve Bank) Directions, 2017 (DNBR(PD)CC.No.078/03.10.124/2017-18), which classify P2P platforms as NBFCs and bring them under RBI supervision. The market includes platforms like LenDenClub, Faircent, i2iFunding, and Finzy — collectively managing billions in outstanding loans directly connecting individual and institutional lenders to borrowers.

The August 16, 2024 amendment to the P2P Directions was the most significant regulatory update since the original 2017 framework. It introduced stricter lender protection measures, the T+1 fund deployment requirement, the guaranteed-return prohibition, and — most consequentially for AI governance — the board-approved non-discriminatory matching algorithm obligation.

This obligation is unique globally. No P2P regulatory framework in the EU, US, or UK requires that the borrower-lender matching algorithm be explicitly approved at board level and documented as non-discriminatory. In India, it is a regulatory requirement. Combined with MRMF's application to P2P credit assessment AI, the full AI governance framework for NBFC-P2P is more demanding than most platforms have recognized.

What the 2024 Amendment Changed for AI Governance

The August 2024 amendment introduced four changes with direct AI governance implications:

Board-approved non-discriminatory matching algorithm: The platform's lender-borrower matching algorithm must be approved at board level and must be non-discriminatory. The board approval requirement brings the matching AI model into the same governance tier as credit policy — a material business decision requiring board oversight. Non-discriminatory matching requires that the algorithm not systematically disadvantage or advantage any borrower segment on protected characteristics (gender, religion, caste, region) or on characteristics that are proxies for protected status.

Closed-user-group prohibition: Platforms cannot operate arrangements where borrowers are sourced exclusively or preferentially from affiliated entities. This has direct AI implications: the matching algorithm cannot use origination channel as a feature that influences matching priority.

T+1 fund deployment: Lender funds must be deployed to borrowers within one business day. For AI-driven platforms, this is a service-level obligation on the credit assessment and matching pipeline.

Guaranteed-return prohibition: Platforms cannot market or imply assured returns to lenders — which prohibits AI-generated performance projections that present past returns or model-predicted returns as assured future yields.

Non-Discriminatory Matching: The Board Approval Governance Architecture

Translating the board-approval obligation into AI governance requires building a governance architecture that a board can actually approve:

Algorithm documentation for board review: The matching algorithm documentation must be written at a level of abstraction that allows a board to assess its non-discriminatory characteristics without requiring technical model expertise. This means: the features used in matching (and what they represent), the outcome being optimized (matching speed, expected return, credit risk), the demographic performance analysis showing equal matching rates across protected and non-protected segments, and the conflict-of-interest assessment for closed-user-group compliance.

Board-approved non-discrimination policy: Complementing the algorithm documentation, the board should approve a formal non-discrimination policy for the matching algorithm that defines: which characteristics are protected in matching, how non-discrimination is measured (demographic parity, equal opportunity), what thresholds constitute material disparity requiring remediation, and the frequency of non-discrimination monitoring.

Annual board review: The non-discriminatory matching obligation is ongoing, not a one-time approval. The board should receive an annual report on matching algorithm performance including non-discrimination metrics, any demographic disparities detected and remediated, and any model changes with their non-discrimination impact assessment.

MRMF Application to P2P Credit Assessment AI

NBFC-P2P platforms are RBI-regulated entities. MRMF June 2026 applies to their credit assessment AI models — the models that score borrower creditworthiness to determine matching eligibility and risk-based pricing for lenders.

P2P credit AI has a specific characteristic that distinguishes it from bank credit AI: the assessment affects both the borrower (whether they match with lenders at all) and the lender (the risk they take on). A false positive in credit assessment — overestimating a borrower's creditworthiness — harms lenders who match with that borrower. A false negative — underestimating creditworthiness — harms borrowers who are excluded from the platform or routed to higher-cost lender categories.

MRMF PSI monitoring is particularly important for P2P credit AI because P2P borrower populations are more volatile than bank credit populations — they skew toward segments with thinner bureau histories and higher sensitivity to economic shocks. COVID-era P2P default events demonstrated that P2P borrower risk profiles can shift substantially faster than bank credit portfolios, creating drift monitoring requirements with shorter review cycles.

Guaranteed-Return Prohibition and AI Marketing Governance

The 2024 amendment's prohibition on guaranteed-return marketing extends to AI-generated content. A recommendation engine or marketing AI that generates lender-facing communications stating or implying assured returns — "lenders in this category historically earned 11–13% annualised" without adequate risk disclosure, "our AI model predicts returns of X%" — violates the prohibition.

AI governance obligations for P2P marketing AI:

Output filtering: any AI-generated communication to lenders must be screened for guaranteed-return language before delivery
Historical performance disclosure: if AI-generated content references historical returns, it must include compliant risk disclosure that past returns do not indicate future performance
Model-predicted return disclosure: if the platform uses AI to project expected returns for specific loan matches, projections must be presented as estimates with risk ranges, not guaranteed yields
Audit trail: marketing AI outputs to lenders should be logged for regulatory examination

NBFC-P2P AI Governance Checklist

☐ Board-approved matching algorithm: formal board resolution approving the matching algorithm with documented non-discrimination policy; annual board review scheduled
☐ Non-discrimination testing: demographic performance analysis for matching algorithm across protected segments; equal matching rate analysis by gender, geography, and bureau-thin status
☐ Closed-user-group compliance: origination channel not a direct or proxy feature in matching algorithm; backtesting confirms no systematic matching advantage for affiliated-entity borrowers
☐ MRMF model inventory: credit assessment AI in MRMF model inventory with model card; independent validation conducted
☐ P2P-specific PSI monitoring: shorter review cycle appropriate for P2P borrower population volatility; PSI thresholds calibrated to P2P portfolio dynamics
☐ Third-party bureau AI: if credit bureau AI score is used in credit assessment, MRMF Section 7 obligations satisfied (inventory entry, independent validation)
☐ T+1 AI SLA: credit assessment and matching pipeline SLA documented; fallback process that meets T+1 defined; T+1 incident reporting process in place
☐ Guaranteed-return output filter: marketing AI output filtered for guaranteed-return language; historical performance disclosure compliant; model-predicted return disclosure policy
☐ DPDP compliance: borrower credit data processing has lawful basis; purpose limitation for P2P credit assessment; data retention aligned with RBI KYC MD (not indefinite)
☐ RBI examination readiness: board-approval documentation, non-discrimination test records, MRMF model card, matching algorithm change history — packaged for examination