Your CIBIL Score Is Not What You Think It Is
Your CIBIL Score Is Not What You Think It Is
Part of “The AI You Don’t See” series by Akshay A. Walimbe
You probably know your CIBIL score. Or at least you have checked it once, maybe when you applied for a credit card or a home loan. It is a three digit number between 300 and 900. Above 750 and you are golden loans approved, good interest rates, banks calling you. Below 650 and doors start closing. Below 550 and you might as well not apply.
You have been told it measures your creditworthiness. Your financial discipline. How likely you are to repay a loan.
That is partly true. But there is a lot your CIBIL score is doing that nobody told you about. And a lot it cannot explain even to itself.
Let me start with what you probably know.
CIBIL TransUnion CIBIL, to use its full name maintains credit information on over 600 million individuals and 32 million businesses in India. Your score is calculated from your repayment history, credit utilisation, the length of your credit history, the types of credit you hold, and the number of recent inquiries on your file. Pay your EMIs on time, do not max out your credit cards, maintain a healthy mix of loans, and your score goes up. Miss payments, default, or apply for too many loans in a short period, and it goes down.
Simple enough. Transparent enough. Fair enough.
But here is the thing. That traditional CIBIL score is increasingly not the only number that determines whether you get a loan.
A significant portion of India’s population remains financially excluded. While the World Bank’s Global Findex 2021 found that about 78 per cent of Indian adults have bank accounts, many of those accounts are inactive or underused the Jan Dhan Yojana opened hundreds of millions of accounts, but meaningful banking activity remains limited for a large segment. Hundreds of millions of people have no formal credit history. No EMIs. No credit card repayment record. No data for CIBIL to score.
For traditional banks, these people simply do not exist in the credit system. A farmer in Madhya Pradesh who has reliably repaid informal loans from his cooperative for twenty years has a CIBIL score of zero. A street vendor in Hyderabad who has never missed a payment to her microfinance group has no credit file at all. They are invisible.
This is where India’s fintech revolution steps in. Companies like mPokket, KreditBee, and dozens of others offer instant digital loans to people who have no traditional credit history. Their stated goal is financial inclusion reaching borrowers that traditional banks will not serve. To assess creditworthiness without a CIBIL score, they use AI. And that AI looks at things you would never expect.
Your phone model. Your app usage patterns. Your mobile recharge behaviour. Your text messages. Your location data. Your PIN code. How often you charge your phone. What time of day you are active online. What websites you visit. Your typing speed.
Every digital footprint you leave becomes a data point in an invisible financial profile that determines whether you get a loan, how much, and at what interest rate.
As one industry analysis noted: “Every website visit, device choice, and typing habit becomes part of an invisible financial profile.”
Now I want to talk about one data point in particular. Your PIN code.
In most countries, your postal code tells a lender something about the neighbourhood you live in. Property values. Average income. General socioeconomic indicators. This is already controversial in the West using zip codes in credit decisions has been challenged as a proxy for race in the United States.
In India, your PIN code can carry even more. Due to historical residential patterns patterns shaped by caste, class, and community over generations your PIN code may correlate with your caste. It may correlate with your religion. It may correlate with your socioeconomic background in ways that go far deeper than income. This is not conspiracy. It is a consequence of how Indian cities and towns developed over centuries.
This is what researchers call a proxy variable. The algorithm never asks you your caste. It does not need to. Your PIN code tells it enough. Your surname tells it more. The school listed on your loan application government school versus private, English medium versus Hindi medium fills in the rest.
Nithya Sambasivan, then a researcher at Google, studied this phenomenon in the Indian context. Her research, “Re imagining Algorithmic Fairness in India and Beyond,” published at the ACM FAccT conference in 2021, found that “access to the Internet and technology has mostly been restricted to urban dominant castes in India, with only 50 per cent of people having access and the excluded half mostly comprising of Dalits, Adivasis and Muslims, resulting in algorithms that cater mostly to dominant groups and perpetuate systemic injustices.” (The internet access figure has improved since 2021, with India now at approximately 886 million active users according to IAMAI Kantar data for 2024, but the underlying representational imbalance in who generates AI training data persists.)
The excluded half. Let that sink in. Half the country is essentially invisible to the data systems that AI learns from. And the AI, learning from the half that is visible, builds its model of creditworthiness based on patterns that reflect urban, uppercaste, male, English speaking India.
But it gets worse. Because there is a feedback loop.
If your community was historically excluded from formal banking as Dalit, Adivasi, and Muslim communities disproportionately were then you have a thinner credit file. A thinner credit file means a lower or nonexistent credit score. A lower score means you are denied credit or given worse terms. Being denied credit means you cannot build a credit history. And the next generation of AI models sees even more evidence that “people like you” are higher risk.
The bias does not stay static. It compounds. Each cycle of the algorithm reinforces the pattern from the previous cycle. The machine does not know it is encoding centuries of caste based exclusion into a three digit number. It just sees patterns in data.
Research from Stanford’s Institute for Human Centered AI (HAI) found that credit scores for minorities are about 5 per cent less accurate in predicting default risk, and scores for people in the bottom fifth of income are about 10 per cent less predictive. (This research examined the US context, but the underlying mechanism that data sparse populations get less accurate scores applies anywhere credit scoring relies on historical data.) The score is literally less reliable for the people who need fair assessment the most.
And here is the part that should make every Indian uncomfortable.
When a bank denies your loan application, the RBI’s Fair Practices Code requires that they tell you why. But when the decision is made by an AI algorithm processing thousands of data points your phone model, your PIN code, your recharge patterns, your app usage what exactly do they tell you?
“Your application did not meet our credit criteria.”
That is it. That is what most people get. A form rejection. No explanation of which of the thousand data points tipped the scale. No way to know if it was your PIN code, your phone model, or the fact that you top up your prepaid mobile in small amounts instead of large ones.
The RBI is aware of these concerns. In 2022, the RBI issued Digital Lending Guidelines addressing data misuse and opaque credit practices. In May 2025, they released updated Digital Lending Directions with new requirements around transparency, neutrality in multi lender arrangements, and a ban on dark patterns. These are meaningful steps the 2025 Directions significantly tightened the regulatory framework.
But there is a critical gap. The RBI’s own Working Group had recommended requiring algorithmic transparency mandating that lenders document the rationale for their algorithmic features and conduct regular algorithmic audits.
Those specific recommendations were not incorporated. Not in the 2022 Guidelines. Not in the 2025 Directions. To be fair, implementing algorithmic audits at scale across India’s sprawling fintech ecosystem is not straightforward the technical infrastructure, standardised audit frameworks, and qualified auditors do not yet exist in sufficient numbers. But the gap between what the RBI’s own experts recommended and what was actually mandated remains a concern for consumer protection advocates.
In January 2026, the Delhi High Court stepped in, directing the RBI to file a counter affidavit in response to a writ petition alleging that digital lending apps are violating the data protection rights of borrowers. The case is ongoing, and the outcome is uncertain. But the fact that courts are now examining questions about algorithmic lending practices signals that this issue is moving from policy discussions into the legal arena.
This is not an abstract policy debate. Let me make it concrete.
You are 28 years old. You live in a tier 3 town. You work a steady job. You have never defaulted on anything. You need a personal loan for a family medical emergency. You apply through a fintech app because the nearest bank branch is an hour away and you need the money today.
The app asks for access to your phone. You say yes because you need the loan. The AI scans your device. It sees an Android phone, not an iPhone. It notes your PIN code. It sees your app usage mostly regional language apps, not English. It checks your mobile recharge patterns small, frequent top ups, not large monthly plans.
None of these things have anything to do with your ability to repay a loan. But each of them correlates, statistically, with demographic characteristics that the algorithm has learned to associate with higher risk. Not because people from your PIN code default more often. But because people from your PIN code have historically had less access to formal credit, which means less data, which means the model is less confident, which means higher risk score, which means rejection.
The computer said no. Nobody can tell you why.
Dvara Research, one of India’s most respected financial inclusion think tanks, put it starkly in their 2024 study on AI in Indian digital credit: “AI based systems can perpetuate biases present in data but do so at scale, without traceability and often with the impression that there is no bias given the automated nature of the system.”
Without traceability. That is the key phrase. The algorithm makes a decision that affects your life whether you can borrow money during a medical emergency, whether you can finance your child’s education, whether you can start a small business and there is no trail. No explanation. No way to challenge it. No way to even know what factors were considered.
NASSCOM, India’s own technology industry body, acknowledges the problem. In their report “Beyond Algorithms: Navigating Fairness in India’s Lending Landscape,” they define algorithmic discrimination as occurring “when an algorithm offers privilege to any individual or demographic over another in ways different from the system’s intended function, resulting in unequal outcomes.”
The intended function is to assess creditworthiness. The unintended outcome is encoding caste and class into lending decisions at a scale that no human loan officer could ever achieve.
The next time you check your CIBIL score, remember: that number is just the visible part. Behind it, a growing ecosystem of AI powered credit decisions is using data about you that you never knowingly shared, making judgments based on correlations you cannot see, and producing outcomes you cannot challenge.
Your PIN code. Your phone. Your language. Your digital footprint. All of it is being scored. None of it is being explained.
The computer said no. Nobody can tell you why.
I’m have written a book about exactly this how AI and automated systems make decisions about your life, where accountability disappears, and what we can do about it. If you want to know morea about this book or order a copy, you can do it here: https://akshaywalimbe.com/beyond-bias/