Consider an illustrative case. A woman in Nagpur runs a small tailoring unit. Her UPI collections are steady. Her GST filings are clean. She has never missed a bill in her life. Still, when she applied for a working capital loan on a lending app, she got rejected in under a minute. No human read her file. No one explained why. A score decided, and the app moved on to the next applicant.
AI credit scoring in India is making such decisions faster and more data-driven. Banks and non-banking financial companies can use credit-bureau records, consented bank-account information, income indicators, GST data and other permitted inputs to estimate the probability that a borrower will repay. This can help lenders assess applicants who lack conventional salary slips or long credit histories. It can also reproduce historical exclusions, penalise borrowers with thin financial records or generate decisions that are difficult to explain and challenge.
RBI's own 2025 survey found that 20.8% of responding financial entities had already deployed AI across areas such as customer service, sales, cybersecurity and credit underwriting, while about 67% expressed interest in exploring AI use cases.
India has fair-lending and non-discrimination requirements, but it does not yet have a dedicated AI-lending regime that prescribes how models must be tested for disparate outcomes, what results must be disclosed or when human review is mandatory. RBI's Ombudsman Scheme alone received 13.34 lakh complaints in FY25, up 13.55% in one year. Loans and advances made up nearly 30% of them.
No regulator is actively auditing these models for bias today. AI credit scoring in India is moving faster than the oversight meant to check it.
Also Read | AI Bias in Indian Credit Scoring: Who Gets Excluded When Algorithms Decide?
How AI Credit Scoring Works in India
A customer applies for a loan on an app. The lender checks identity first, usually through Aadhaar e-KYC and PAN. Next comes financial data collection. This is where AI credit scoring in India has changed the most. Lenders no longer rely only on handwritten bank statements. They now pull:
- Bank statements, fetched digitally.
- UPI transaction history.
- GST filings, for small business owners.
- Credit bureau data from CIBIL, Experian, Equifax and CRIF High Mark.
- Account Aggregator data, shared with the borrower's consent.
- Digital footprint signals, such as app usage and device data.
- Alternate data, including utility bills and e-commerce spending.
This raw data goes through feature engineering. A data team turns messy numbers into clean inputs, like "average monthly balance" or "bounced payments in six months." These feed an AI model, often a machine learning model called a gradient boosted tree. It outputs a credit score or a default risk. Some lenders send high value or unclear cases to a human underwriter.
Many small loans, especially instant personal loans under Rs 50,000, get approved or rejected with no human involved at all.
Also Read | Why Responsible Interest Rate Regulation Can Strengthen India's Small-Ticket Lending Ecosystem?
India's Account Aggregator framework has made this pipeline one of the fastest in the world. By FY26, the network had handled over 500 crore data fetches and 45 crore consents. Daily consents crossed 7 lakh, per industry body Sahamati. Bank data now reaches a lender's AI model in 2 to 5 seconds.
That speed is the whole selling point. It is also where things can go wrong. A model working this fast rarely pauses to check if its own logic is fair.
The Bias Was Already There Before the AI Ever Ran
This is the part that surprises people. AI credit scoring in India usually turns biased not because of bad code, but because the training data was already unfair.
Take a simple case. If a model learns from five years of past approvals, and those approvals favoured salaried men in metro cities over gig workers and women in smaller towns, the model treats that pattern as normal. It is not taught to discriminate. It simply repeats old bias at massive scale, instantly, for every future applicant.
Then there is the proxy problem. A model may never see someone's religion, caste or gender directly. But pin code often maps to community clusters. Phone brand often maps to income level. Even UPI spending categories can hint at occupation, and occupation can hint at social background.
Finance professors Hurlin, Perignon and Saurin have shown that removing gender or religion from a dataset does not fix this. The model finds the pattern anyway through these proxies. Regulators call this "fairness through unawareness," and it almost always fails.
A third source is thin credit history. Women, first time earners, gig workers and rural applicants often have shorter digital footprints, simply because they entered formal finance later. A model trained mostly on urban, salaried borrowers can end up under scoring exactly the group financial inclusion is meant to help.
RBI's FREE-AI Committee flagged this in its August 2025 report. It found bias and fairness risk, from skewed training data, to be a central concern as AI adoption grows across Indian finance.
Also Read | Bima Sugam vs Policybazaar: What It Means for PB Fintech and Insurance Stocks
What the DPDP Act Covers—and What It Misses
Many people assume the DPDP Act, 2023 already controls how AI uses personal data for loans. That is only partly true. The law rolls out in three stages. The government formally established the Data Protection Board framework in November 2025, but its practical operationalisation remained underway in mid-2026, including the appointment of its chairperson and members.
Many of the DPDP regime’s substantive obligations are scheduled to take effect 18 months after the November 2025 notification—around mid-May 2027—while other provisions begin earlier.
DPDP protects personal data well, but says little about how AI decisions should be explained. If an AI system rejects your loan, you may learn what data was used. You may still not get a clear answer on why the AI said no. That missing explanation could become one of India's biggest lending policy questions before 2027.
Where the Legal and Regulatory Gaps Actually Sit
Even though India has started regulating AI in finance, three major gaps still remain.
1. No law directly targets AI bias in lending.
The RBI's Digital Lending Guidelines (2022) ask banks and NBFCs to use reliable and auditable algorithms. However, they do not explain how AI models should be tested for bias, what information lenders must disclose, or what penalties they could face if an AI system unfairly rejects borrowers.
Also Read | RBI Bond Auction 2026 Brings Rs 32000 Crore to the Table
2. RBI's AI framework is still only guidance.
In August 2025, the RBI released its FREE-AI Framework, which recommends AI governance, accountability and transparency. But these are recommendations, not legally binding rules. Until the RBI turns them into mandatory directions, lenders face no AI-specific penalty for using a biased credit-scoring model.
3. Borrowers still struggle to challenge AI decisions.
The problem is growing. The RBI Ombudsman Scheme received over 13.34 lakh complaints in FY25, up 13.55% from the previous year. Loans and advances accounted for nearly 29% of all complaints, while NBFC-related complaints crossed 43,800. Yet the current system was designed for issues like interest rates or loan recovery—not for borrowers asking why an AI model rejected their loan despite a good repayment record.
These gaps show that while AI is already shaping lending decisions, India's rules for ensuring fairness, transparency and accountability are still catching up.
India, the EU and the US Compared
Europe and the US have moved faster on this exact question. The EU AI Act, in force since 2024, treats AI credit scoring as high risk. The EU AI Act classifies certain AI systems used to assess an individual’s creditworthiness as high-risk. European data-protection law also restricts solely automated decisions that produce significant effects, subject to legal exceptions and safeguards.
In the US, regulators say lenders cannot hide behind AI when a decision is unclear. A rejection like "insufficient income" is not enough if that is not the real reason. Borrowers must get the specific cause.
India has not made these steps mandatory yet. But RBI's FREE-AI framework already recommends similar measures, including bias testing and clearer borrower explanations. The next step is turning these ideas into binding rules every lender must follow.
Reading Between the Lines of the AI Lending Boom
Most coverage of AI lending focuses on the upside: faster loans, wider access, easier credit for gig workers and small businesses. A few facts get missed. One is the gap between technology and regulation. India's Account Aggregator network shares data in seconds and handles crores of consents a year. Yet the DPDP Act, meant to protect that same data, will not be fully enforced until May 2027. AI lending is growing faster than the rules meant to govern it.
Another is that RBI still leans on the Information Technology Act, 2000, a law written long before AI credit scoring existed, to handle many current AI risks.
There is also a question of accountability. A wrong decision from a person can be challenged and traced. An AI system rejecting lakhs of applications a month can hide unfair patterns at scale. The rise in Ombudsman complaints, up 13.55% in FY25, may reflect this exact problem: borrowers judged by systems that rarely explain themselves.
What Borrowers Can Do After an Unexplained Loan Rejection
A rejected applicant may not have access to the lender’s complete model, but that does not mean there are no practical steps available.
- First, ask the lender or lending app for the principal reason for the rejection. Contact its customer-support team and designated grievance officer in writing so that there is a record of the request.
- Second, obtain a copy of the relevant credit report and check it for incorrect accounts, duplicate loans, outdated balances, late payments that do not belong to you or personal details that have been wrongly recorded.
- Third, check whether the financial information supplied through an Account Aggregator, uploaded statement or GST record was complete and current.
- Fourth, ask whether the application can be reconsidered with updated information or through manual review.
- Finally, where a complaint against an RBI-regulated entity remains unresolved or the response is unsatisfactory after the applicable waiting period, the borrower may consider escalating it through the RBI Complaint Management System, subject to the Ombudsman Scheme’s eligibility requirements.
News4Bharat POV
AI-powered underwriting can make credit assessment faster, reduce paperwork and help lenders evaluate borrowers who do not fit traditional salaried profiles. Those benefits, however, depend on whether the systems are accurate, proportionate and accountable.
India does not have to choose between fast credit and fair credit. It needs rules that require lenders to deliver both. Every material automated rejection should generate an intelligible reason. High-impact models should undergo independent testing for data quality, performance and unfair outcomes. Borrowers should have a practical route to correct inaccurate information and seek human reconsideration when a decision appears unreasonable.
RBI’s FREE-AI framework has established a useful direction, but voluntary principles will have limited value unless they are translated into measurable standards, supervisory expectations and consequences for serious failures.
Sources:
- Reserve Bank of India
- Ministry of Electronics and Information Technology
- Sahamati
- European Union legal texts
- US Consumer Financial Protection Bureau
- Peer-reviewed credit-fairness research
This article analyses publicly available regulatory documents, industry reports and academic research. The opening example is illustrative and does not represent a named borrower unless otherwise stated.
