How avatarin built a 24/7 retail agent with GPT-Realtime

Japan’s home-appliance retailers face a persistent challenge: extending expert sales support beyond store hours while staffing remains tight. avatarin, an AI customer service company spun out of ANA Holdings, partnered with Yamada Holdings to turn experienced associates’ knowledge into a 24/7 multilingual shopping agent.

Built on OpenAI’s GPT‑Realtime, the Kurashi-Marugoto AI Agent supports natural voice conversations and guides shoppers from product discovery to purchase decisions. In a two-week public campaign on Yamada Denki’s online store, approximately 30,000 people used the agent; 92% of survey responses were positive.

avatarin had worked with OpenAI well before the Yamada Denki project, using the OpenAI API for speech recognition, inquiry analysis, and employee training. Those earlier efforts helped the team understand what it takes to apply AI in practice. The shopping agent was its first major opportunity to bring those lessons directly to customers.

OpenAI worked with avatarin to structure complex prompts, optimize API costs for always-on voice service, and share implementation best practices. That support helped avatarin focus on translating Yamada Denki’s service model into an experience that felt accurate, responsive, and distinctly on-brand.

For Fukabori, one of the key advantages of GPT‑Realtime was the ability to bring voice, text, and visual understanding together within a single conversational experience.

“For us, the performance of a single model went beyond what we had seen from specialized speech recognition systems. It can work across speech, text, and images with low latency. That is why we chose GPT‑Realtime. Instead of asking people to adapt to the rules of a machine, the technology can adapt much more naturally to the way people communicate.”

A conventional chatbot waits for the right keyword. avatarin’s agent listens for context instead. It can respond to a shopper’s uncertainty, changing preferences, and the details that turn a generic answer into a useful recommendation.