Kakao Mobility uses Gemini Nano on-device to reduce costs and boost call conversion by 45%

Kakao Mobility, a leading South Korean mobility service, faced challenges with improper bike and scooter parking leading to violations and safety concerns. They also sought to improve the address entry process for their delivery services to reduce manual work for drivers. Initially, they considered a cloud-based image recognition model for parking but found it too costly and a privacy risk. Similarly, improving address extraction traditionally would require significant machine learning expertise and data. To overcome these hurdles, Kakao Mobility utilized Gemini Nano via ML Kit’s GenAI Prompt API. This on-device AI solution offered a cost-effective and privacy-preserving method. For bike parking, they used Gemini Nano's multimodal capabilities to analyze photos and detect violations, providing real-time feedback to users. This feature protects user location data by processing images directly on the device. For the delivery service, the GenAI Prompt API streamlined entity extraction from natural language orders. Instead of complex machine learning, a simple prompt allowed the extraction of recipient details like name, address, and phone number. This significantly reduced developer overhead and development time. The implemented address entry improvement has already shown substantial results, reducing order completion time by 24% and increasing conversion rates by 45% for new users. The team also noted a substantial increase in AI-powered orders during peak seasons. Kakao Mobility plans to further enhance their parking feature and explore other on-device AI applications. They highly recommend ML Kit’s GenAI Prompt API for its efficiency, security, and cost-effectiveness.