Korea Deep Learning cuts voice-phishing claim paperwork time by 88% with VLM
The proof-of-concept with IBK Industrial Bank of Korea won an excellence award at the 2026 Public Institution AI Innovation Challenge.
Korea Deep Learning said on October 30 that a proof-of-concept automating voice-phishing victim compensation work with IBK Industrial Bank of Korea cut document processing time by 88%. The project applied the company's finance-specific vision-language model (VLM) AI agent, DEEP Agent for Finance.
According to datanet.co.kr, the project was named an excellence award winner in the AI service proof-of-concept category of the 2026 Public Institution AI Innovation Challenge, and IBK Industrial Bank of Korea received awards from the ministers of economy and finance and of science and ICT.
In the demonstration, the system recorded 100% accuracy on printed-text recognition and more than 97% on handwriting. Document classification and inference of crime type each reached 100% accuracy, the company said.
Korea Deep Learning implemented an on-premise setup that can run on the bank's closed network, along with a structure that branches between automatic processing based on confidence scores and review by a staff member.
The demonstration ran from April to August 2025. Previously, voice-phishing compensation work involved staff manually checking documents received by fax and entering them into the system.
Plans for 2027
IBK Industrial Bank of Korea plans to apply the technology to actual operations in the first half of 2027 based on the demonstration results. In March 2025, Korea Deep Learning said its document-specialized VLM, KDL Frontier, ranked first worldwide in the English category of the OCRBench v2 benchmark.
Quick answers
What results did the Korea Deep Learning and IBK Industrial Bank of Korea demonstration produce?
Document processing time for voice-phishing victim compensation work fell by 88%, with 100% accuracy on printed-text recognition, more than 97% on handwriting, and 100% on document classification and crime-type inference.
When will IBK Industrial Bank of Korea apply the technology to real work?
The bank plans to apply it to actual operations in the first half of 2027 based on the demonstration results.
Which AI model was used in the demonstration?
Korea Deep Learning's finance-specific VLM AI agent, DEEP Agent for Finance.