Kakao’s First ‘AI Asset’ Is Something Naver Doesn’t Have

Kakao is developing KakaoTalk Gift data as an AI asset instead of using raw KakaoTalk conversations. KakaoTalk messages are private conversations and cannot be used for AI training, but KakaoTalk Gift leaves behind users’ tastes, relationships and purchase intent as transaction data. / Image generated by ChatGPT KakaoTalk Gift is a relationship-based commerce service that lets users search for products by recipient, gifting situation, price range and theme. When users select and purchase products for a specific purpose, KakaoTalk Gift accumulates data on tastes, relationships and purchase intent. For Kakao, which faces limits in directly using raw conversations, this serves as a data foundation inside KakaoTalk that can be expanded into AI recommendations, search and advertising. KakaoTalk conversations cannot be used as AI training data because they are private conversations. Scatter Lab, the company behind the AI chatbot “Lee Luda,” previously used messenger conversations collected through a relationship analysis service to train AI and was fined and sanctioned 103.3 million won by the Personal Information Protection Commission for privacy violations. Kakao’s decision to design “Kanana in KakaoTalk” as an on-device AI service is also seen as a move that reflects the limits on using raw conversation data. For Kakao to advance its AI, it needs various types of data other than KakaoTalk conversations. KakaoTalk Gift contains long-tail data showing to whom users sent gifts, for what purpose and at what price range. Long-tail data refers not to mainstream data generated commonly by large numbers of users, but to granular data scattered across different tastes, situations and relationships. A theme such as “funny gifts” on KakaoTalk Gift can provide data on what kinds of situations prioritize humor over practicality and what products users choose in those situations. Naver has argued for the importance of long-tail data earlier than Kakao. This appears...