Korea Deep Learning, IBK Cut Voice Phishing Paperwork Time 88%

Korea Deep Learning and IBK Industrial Bank of Korea have completed a pilot in which a vision-language model (VLM) AI agent automated paperwork for voice phishing victim relief. According to the companies, document processing time fell by 88% compared with manual handling.

The trial ran from April to August and was carried out with the bank's Financial Consumer Support Department. Applications and ID documents arrive by fax in the relief process, and until now bank staff had to check each one and type the details into internal systems. That slowed down the response to victims.

Beyond traditional OCR

The system is built on Korea Deep Learning's finance-focused agent, DEEP Agent for Finance. It classifies document types, extracts key data and infers the type of crime from the content. Where conventional optical character recognition (OCR) focuses on reading characters, the VLM also analyzes the visual layout and context of a document.

In the pilot, recognition accuracy reached 100% for printed text and more than 97% for handwriting. Document classification and crime-type inference both scored 100% under the evaluation criteria, and accuracy on some extraction items was up to 20.4 percentage points higher than with existing AI-OCR.

The design cross-checks AI-OCR and VLM results. Depending on confidence, a case is processed automatically or sent to a staff member, and bank employees make the final decision. The system runs on-premises inside the bank's internal network, so financial data is not sent to an external cloud.

Recognition and next steps

The project won an excellence award in the AI service proof-of-concept category of the 2026 Public Institution AI Innovation Challenge, hosted by the Ministry of Economy and Finance and the Ministry of Science and ICT. IBK received a ministerial award for it.

IBK plans to apply the technology to live relief work in the first half of next year. Korea Deep Learning intends to expand document automation for financial and public institutions. The bank said the faster handling frees up response time that matters in voice phishing cases.