avatarin's GPT-Realtime AI Agent Served 30,000 Yamada Denki Customers

Japanese startup avatarin has deployed a retail AI agent built on OpenAI's GPT-Realtime API in stores of the major electronics chain Yamada Denki. In its first two weeks, 30,000 shoppers used the agent, and 92% of survey respondents rated the interaction positively, according to a case study published on the OpenAI Blog.

What the agent does in Yamada Denki stores

The avatarin AI agent advises customers 24/7 without any live staff involved. The system covers the consultant's role at night, on holidays, and during peak periods when stores are short-staffed. It runs on OpenAI's GPT-Realtime API, an interface designed for real-time conversations with minimal latency and support for voice and text input.

The agent's key advantage is multilingual support. This matters a great deal for Japanese retail: the country receives tens of millions of foreign tourists from Asia, Europe, and North America every year, and the language barrier has traditionally been a serious obstacle to serving them. The agent lets shoppers speak to it in their native language, with no interpreter or specially trained employee required.

"Avatarin uses

OpenAI's GPT-Realtime to give Yamada Denki customers round-the-clock multilingual support," the official OpenAI Blog post about the project says.

Key numbers from the first two weeks

The avatarin agent drew 30,000 users in its first two weeks at Yamada Denki. According to the OpenAI Blog, the results are as follows:

• 30,000 shoppers used the agent in the first two weeks
• 92% of survey respondents rated the interaction positively
• Operating mode: around the clock, with no breaks or days off
• Multilingual support without hiring additional staff

A 92% approval rate is a strong result for any automated customer-service channel. Traditional retail chatbots often score poorly because of canned answers and limited understanding of the context of a request. According to OpenAI, the GPT-Realtime API allows a more flexible dialogue: the system picks up on the nuances of a question and responds to the situation at hand.

Why GPT-Realtime instead of a standard chatbot

OpenAI's GPT-Realtime API is built specifically for low-latency scenarios: voice agents, live consultations, and interactive real-time assistants. A standard pipeline first converts speech to text, then processes it with a language model, then synthesizes the reply as speech again. GPT-Realtime instead processes the audio stream directly. This cuts latency and makes the conversation feel more natural to the user.

For avatarin, choosing this API simplified the system architecture and sped up the launch. According to the OpenAI Blog post, the team built a production-grade solution on a compressed timeline, skipping the long enterprise rollout cycle typical of traditional customer-service systems. This approach lowers the technical barrier for companies that want to add a voice AI agent but are not ready to build their own speech-recognition pipeline from scratch.

What this means

The avatarin and Yamada Denki case is one of the first public confirmations that the GPT-Realtime API works under the heavy traffic of real-world retail, with measurable results. 30,000 interactions in two weeks and 92% positive ratings are not a lab pilot but a production result in a working store chain.

Retail chains facing staff shortages, a need for round-the-clock coverage, and a language barrier now have a concrete example of how the GPT-Realtime API addresses all three problems at once. avatarin has shown that the path from idea to a production AI agent with high user satisfaction has become significantly shorter.