Kakao Unveils 'AI Easy Settlement' Feature to Automate Corporate Card Reimbursements

Kakao's internal information systems department developed the 'AI Easy Settlement' feature to reduce employees' repetitive corporate card reimbursement tasks. When users upload multiple receipt images, the AI analyzes the content, automatically matches it with transaction details, and recommends expense categories.

To overcome the limitations of traditional OCR, Kakao introduced LLMs for contextual understanding and error correction. In particular, by applying its proprietary 'Kanana' model, it achieved a high text extraction accuracy of over 98% even on low-resolution images.

Additionally, the system analyzes users' past settlement patterns and transaction history to proactively suggest items such as 'overtime meal expenses' or 'team dinner costs.' During development, the team secured an 80% accuracy rate using only vector similarity search as a pre-step to LLM inference, and improved result quality by more than twofold through prompt engineering.