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.

“This is not an extension of the conventional chatbot. It is an attempt to expand what customer service can be with AI. We see it as the beginning of an ‘interface revolution’ in retail.”

Akira Fukabori, CEO, avatarin

From answering questions to understanding people

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.

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.

Fukabori describes the difference with an everyday shopping problem.

“Customers do not want a chatbot. They want intelligence. ‘I need a refrigerator for a family of four, but my kitchen is small. Which one should I choose?’ Real customer service means being able to answer that question.”

Giving retail expertise a voice

avatarin designed the agent around three essential qualities:

- Keeping product information accurate without slowing the conversation. A retrieval-augmented generation system grounds the agent’s responses in product information, while GPT‑Realtime keeps the interaction responsive. This allows shoppers to have a fluid voice conversation while still receiving information based on relevant product data.

- Translating Yamada Denki’s sales expertise into conversation design. The information a sales associate needs to gather varies significantly by product category. avatarin incorporated Yamada Denki’s customer service knowledge into the agent’s conversation flows and prompting. The system is designed to adapt when customers change their requirements or move temporarily off topic, while guardrails help keep the conversation focused on the shopping experience.

- Designing an agent that asks, rather than only answers. Many traditional chatbot experiences wait for the customer to provide the next instruction. avatarin took a more proactive approach. The agent asks follow-up questions to uncover the customer’s actual needs, helping move the conversation from a simple question-and-answer exchange toward guided product discovery.

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.

“I’m excited by how naturally it understands our intentions,” Fukabori says, “and by glimpses of intelligence that sometimes go beyond what people can do.”

When a conversation becomes an insight

The public campaign made the impact visible in just two weeks:

- Approximately 30,000 users

- 24/7 multilingual support by voice and text

- 92% positive survey responses

The most important discovery was not the scale alone. Every conversation revealed what shoppers cared about, why they hesitated, and what might help them decide. “With conventional online shopping, those insights were difficult to see,” Fukabori explains. “With an AI agent, they become part of the conversation.”