RBI Digital Lending Guidelines and NBFC Credit Technology

The RBI guidelines do not require publishing model weights. They do require a plain-language explanation for every application. This is not legal advice; involve your legal and compliance team for specific determinations.

The guidelines are public documents, not obscure rules. Their accountability requirements have concrete operational implications for any NBFC using algorithmic tools in its credit workflow.

The Core Accountability Rule

The RBI Digital Lending Guidelines, consolidated in the September 2022 circular and issued under it, state that the Regulated Entity, the NBFC, retains full responsibility for the lending decision, even when a third-party Lending Service Provider or algorithm assists assessment. Accountability cannot be outsourced to technology.

This directly affects how NBFCs work with credit technology vendors. If an external model produces a score, you must explain how it was produced and which factors shaped it. "The algorithm said so" does not satisfy a declined borrower or an RBI examiner.

The NBFC must maintain an audit trail for credit decisions. For automated assessments, record and retrieve the tool's output, the factors used, and the lender's final decision.

Practical Algorithmic Transparency

Algorithmic transparency does not require publishing model weights. It requires explaining, in terms a credit committee can understand, why a specific applicant received a specific score. The assessment system generally needs two capabilities.

First, the system must provide a factor explanation with each score. A number alone is insufficient for audit. The credit narrative should identify the leading positive and negative signals using underwriting concepts such as cash-flow adequacy, payment regularity, and income stability.

Second, the explanation must relate to the specific application, not a generic description of what the model usually considers. For each record, you should be able to show, using data available at assessment, why that borrower received that score on that date.

Well-designed systems can provide per-application credit narratives through methods such as SHAP values or rule-based explanations. The vendor should deliver them in a format aligned with your credit appraisal memo. Otherwise, translation creates documentation burden.

Borrower Disclosure Rules

Before loan disbursement, borrowers must receive a Key Fact Statement covering items including the Annual Percentage Rate and all charges. For decisions using alternative data or automated scoring, borrower consent and data usage disclosures under the DPDP Act 2023 also apply.

For NBFCs using Account Aggregator-sourced data, consent is captured and managed through the AA framework. The borrower authorizes sharing before data is fetched, and the consent record remains in the AA system. This is cleaner than some earlier bank statement collection methods and meets the RBI requirement for explicit consent before financial data is used in assessment.

A borrower declined partly on alternative data signals should receive an explanation of the contributing factors. This is where algorithmic transparency and borrower disclosure intersect: explain the relevant factors in language the borrower understands, not technical model outputs.

Accountability for Technology Vendors

When NBFCs use third-party credit assessment platforms, the NBFC must retain oversight and control. Its credit committee should review and approve the process, the technology should support rather than make the decision, and the contract should give the NBFC audit rights over the vendor's processes.

NBFCs should scrutinize vendor agreements. Do they provide access to model documentation? Do assessment outputs include factor explanations, not only scores? Are assessment records maintained to support your audit trail requirements?

These are compliance conditions, not merely vendor negotiation points. An NBFC is in a weaker position when a black box vendor prevents it from explaining its automated credit assessments.

Designing Compliance-Ready Credit Technology

These requirements point to several design principles for credit assessment systems used in RBI-regulated NBFC lending.

Explainability is mandatory. Each score or assessment output should include its leading signals in plain language, generated for that assessment rather than copied from a static model description.

Consent must come first. Borrower data should not enter assessment without explicit, documented consent. AA-mediated collection provides this by design. PDFs and manually entered data require separate consent documentation.

Audit logs must be complete and retrievable. Store assessment inputs, feature values, model output, and final credit decision with the application record for internal or regulatory review.

The NBFC retains final decision authority. A credit officer or committee must review the automated assessment before finalization, and the documentation must show that review.

Speed and Compliance Can Coexist

NBFC credit teams sometimes fear that compliance will reduce the speed advantage of automated assessment. That is a misreading of the requirements.

These requirements concern documentation and accountability, not manual steps for every application. An automated, explainable assessment with a complete audit trail can be delivered in two minutes and remain faster and more compliant than undocumented manual review. The rules define required output, not process duration.

Building explainability and audit logging in from the start costs less than retrofitting them after deployment. Address these requirements during vendor selection and integration design, not after an RBI examination exposes gaps.