"Re-evaluating customers who were denied loans"... Upfin and Seoul National University develop AI credit scoring technology

[For more diverse corporate information on the startups mentioned in this article, please visit Unicorn Factory's big data platform 'Data Lab'.]

On the 3rd, AI financial platform Upfin announced that it has jointly developed AI technology to improve credit scoring accuracy with Seoul National University's Industry-Academia Cooperation Foundation and filed a patent application.

This patent is the result of a joint research project on an "AI-based financial framework" conducted by Upfin and Seoul National University since last year. Both parties have been researching related technologies to improve credit scoring for customers with limited financial histories, such as those in developing countries.

The new technology addresses the difficulty of verifying the actual repayment ability of customers who were previously denied loans during the traditional credit evaluation process. While data on loan approvals and subsequent repayments or defaults accumulate for approved customers, no such data exists for rejected applicants since the loans were never disbursed. Consequently, even when AI is used to predict repayment likelihood, there has been limited data available to improve loan approval and rejection criteria.

To address this, Upfin and Seoul National University researchers developed a three-step process. First, the AI learns judgment based on existing credit evaluation criteria, then validates the assessment results through simulations using historical data. In this process, both default rates and factors such as interest income, principal recovery, losses, and costs are incorporated to analyze the profitability and risks associated with loan approval decisions.

Subsequently, among customers who were not approved under existing criteria, a select group deemed to have relatively high repayment potential is chosen for loan approval, allowing actual repayment data to be collected. This newly acquired data is then fed back into AI training to adjust credit evaluation standards. The company explained that this approach enables the gradual accumulation of repayment data from previously unapproved customer segments, which was difficult to obtain through traditional assessment processes.

Upfin particularly anticipates that this technology can be applied to evaluate customers whose limited financial transaction history made it challenging to assess their repayment ability using conventional credit scoring alone.

Upfin plans to gradually implement this technology in the credit evaluation system of India's fintech service "True Balance." The company also intends to utilize these technologies as it expands its business into other emerging markets in the future.

Shin Jae-hyuk, AI Lead at Upfin, stated, "Previously, we could train models to predict default rates, but there were limitations in the data available for improving actual approval and rejection criteria. With this new technology, we can now secure data from unapproved customers and utilize it to refine our assessment standards."

[MoneyToday startup media platform Unicorn Factory]