Kakao Bank builds benchmark for Korean finance AI

(Kakao Bank)

Kakao Bank’s Financial Tech Lab has unveiled new research in artificial intelligence and data science at several international academic forums, highlighting advances designed to speed up explainable AI and improve Korean-language financial search, the internet-only lender said Thursday.

The studies focus on two long-standing hurdles in financial AI: enhancing model transparency and boosting accuracy in processing Korea-specific financial language and context.

At the International Conference on Information and Knowledge Management in Seoul in mid-November, Kakao Bank showcased a high-efficiency explainable AI technique co-developed with the Korea Advanced Institute of Science and Technology. Explainability is essential in regulated sectors such as finance, but many existing methods are too computationally heavy for real-time deployment. Kakao Bank said its new approach accelerates computation by up to 8.5 times while preserving explanation quality, a capability expected to strengthen compliance with Korea’s new AI Basic Act and improve systems such as fraud detection.

The lender also presented research on Korean-language financial information retrieval at the International Conference on AI in Finance in Singapore later that month. Because most large language models are trained primarily on English, they often struggle with Korean financial terminology, suffering in search accuracy and system reliability. To address this gap, Kakao Bank created a dedicated benchmark dataset to objectively evaluate AI performance on Korean-language financial queries — a resource the academic community recognized as foundational for future research.

“Kakao Bank’s recognition at global academic societies shows its technology is already competitive by international standards,” a bank official said. “We will continue advancing reliability and customer convenience through research that can be applied directly to everyday financial services.”