Korean banks and credit card companies are rapidly expanding AI-powered customer service through chatbots and call bots, but actual consultation accuracy falls short of expectations. While these systems handle simple inquiries such as account balance checks and transaction history, they frequently provide incorrect responses to complex questions about card and loan product recommendations or enrollment requirements.
According to a review of major financial institutions' AI services conducted on May 10, AI call bots and chatbots responded relatively accurately to simple keyword-based inquiries but often failed when questions became slightly more complex.
For example, when asked to recommend cards with KTX discount benefits, Shinhan Card's chatbot responded with "Here are cards offering telecommunications fee benefits" and provided a link to the "KT Family Satisfaction DC Shinhan Card." When subsequently asked to recommend the card with the highest public transportation cashback benefits, the chatbot recommended a "card for customers applying for livelihood recovery consumption coupons."
In many cases, inquiries to AI agents connected through customer service phone lines ultimately required transfer to human agents or directed customers to search the company website themselves. When asked to summarize the benefits of a credit card currently in use, Shinhan Card's AI agent simply provided a link to the card company's website without explanation. NongHyup Bank's AI agent also transferred calls to human agents when asked about savings rate comparisons or automatic transfer setup methods.
Errors were also found in AI chatbots embedded in mobile applications. When asked about application requirements for "Woori WON Jeonse Loan," Woori Bank's AI loan consultant explained first-time homebuyer requirements. When asked whether the jeonse loan was only available to first-time buyers, the AI agent confirmed this was correct. However, the actual application requirements for Woori WON Jeonse Loan contain no such restriction.
Product recommendation functions were also limited. When asked about the highest-interest savings account, KB Kookmin Bank's chatbot did not identify the top-rate product but instead listed various savings products. Woori Card's chatbot asked for clarification when requested to recommend cards with low annual fees. A KB Kookmin Bank spokesperson explained, "Since preferential rate conditions vary by customer, we operate by providing general savings product information rather than definitively presenting the highest-rate product."
Kakao Bank's chatbot provided detailed responses even to lengthy questions such as "How much can I borrow with an annual salary of 60 million won and a credit score of 870?" answering that "the credit loan limit is 300 million won, but the final amount depends on debt status and other factors." However, accuracy declined for matters subject to policy changes, such as jeonse loan guarantee limits.
The low accuracy of AI consultation services also affects user satisfaction. According to a Deloitte survey of 2,027 U.S. bank customers conducted in January last year, 74% of respondents preferred human agents over AI chatbots even for simple tasks. Chatbot usage was primarily for simple tasks including technical problem resolution (60%), account inquiries (53%), and card payment processing (29%).
Within the financial industry, current AI consultation services are assessed as remaining at the level of simple inquiry processing. One financial industry official said, "AI consultation often operates based on predetermined keywords or scenarios, making it difficult to handle complex product inquiries. The direction is for AI to handle simple tasks while human agents focus on complex consultations."