Alternative credit scoring gives a credit score to people who have little or no credit history, using data other than past loans. Lenders call these people "new to credit" or "thin file". They include a young person in a first job, a rural woman who runs a small business, and many gig workers.
An ordinary credit score needs a history. The credit bureau can score a person only if that person has had loans or credit cards before. Many Indians have never had either. Without a score, most lenders decline them. So they cannot borrow, and so they never build a history.
Alternative credit scoring uses other data to break this circle. The data can include the money coming into and going out of a bank account, electricity and mobile bills paid on time, and UPI payment patterns. For farmers, it can include data about the crop and the land. All of it is used only with the borrower's consent. The bank data is often shared through the account aggregator.
The score is produced by a statistical or machine-learning model. The model is trained on past borrowers whose repayment is already known. It learns which patterns in the data went with repaying and which went with default, the failure to repay. Data scientists build such models with statistical software.
Say a woman in a village near Sitapur runs a tailoring business and applies for a ₹40,000 loan to buy a second sewing machine. She has never borrowed. With her consent, the lender sees that she has received UPI payments from customers every week for two years, about ₹14,000 a month. Her electricity bill has been paid on time every month. The model places her in a low-risk group. The lender's credit decision engine approves a smaller first loan of ₹25,000. When she repays it, she has a record at the bureau, and her next loan can be larger.
For farmers, crop monitoring from satellite images can show whether the crop on a field is healthy. This helps a lender judge whether a crop loan is likely to be repaid.
In India, rules limit the data a lender may use. Under the RBI's digital lending guidelines, a lending app must not read data such as contacts or photos from a borrower's phone. It must take the borrower's explicit consent for the data it does use.