Japan’s home‑appliance retailers have struggled to extend expert sales support beyond store hours while staffing remains constrained. To address this, avatarin — an AI customer service company spun out of ANA Holdings — partnered with Yamada Holdings to convert the knowledge of experienced associates into a 24/7 multilingual shopping agent.
The solution: Kurashi‑Marugoto AI Agent on GPT‑Realtime
The Kurashi‑Marugoto AI Agent, built on OpenAI’s GPT‑Realtime, supports natural voice conversations and guides shoppers from product discovery to purchase decisions. The system can interpret speech, text, and images within a single conversational flow, allowing it to respond based on context rather than waiting for keywords.
During a two‑week public campaign on Yamada Denki’s online store, roughly 30,000 people used the agent; 92% of post‑use survey responses were positive.
Why GPT‑Realtime?
avatarin had already used OpenAI technologies prior to the Yamada Denki project, employing the OpenAI API for speech recognition, inquiry analysis, and employee training. Those earlier efforts helped the team understand practical requirements for deploying AI in retail. According to Fukabori, a key advantage of GPT‑Realtime was that a single model delivered performance beyond what they had seen from specialized speech recognition systems: it handled voice, text, and images with low latency.
Fukabori noted that the technology adapts to how people communicate instead of forcing people to adapt to machine rules. That enables the agent to address shoppers’ uncertainty, changing preferences, and the contextual details that turn a generic reply into a helpful recommendation.
Design principles and implementation
avatarin designed the agent around three core qualities: accurate context awareness, natural voice interaction, and communication that aligns with Yamada Denki’s service brand. OpenAI worked with avatarin to structure complex prompts, optimize API costs for an always‑on voice service, and share implementation best practices. That support allowed avatarin to focus on translating Yamada Denki’s service model into an experience that felt accurate, responsive, and on‑brand.
Public trial and learnings
The public campaign made the impact visible within two weeks: every conversation revealed what shoppers cared about, why they hesitated, and what might help them decide — insights that were previously hard to surface in conventional online shopping. The service opened new moments for retail: customers could ask questions after stores closed and discuss budgets or uncertainty candidly without sales pressure. Each interaction concluded with a short voice survey, integrating feedback naturally into the shopping experience.
Users described the agent as “easier to talk to than an actual sales associate” and appreciated being able to ask questions again without stress. For avatarin, those conversations offered value beyond immediate sales, exposing opportunities to improve service, marketing, and future customer experiences.
Strategic vision
avatarin envisions a future where a single AI agent serves customers seamlessly across web, phone, and physical stores, carrying customer context forward instead of relying on separate systems per channel. For Yamada Holdings — whose businesses include home appliances, housing, financial services, and more — this approach could make AI a trusted companion across everyday life and embody the company’s hospitality in digital form.
Fukabori says AI has moved into a higher gear and expresses eagerness to build with multimodal intelligence. avatarin also emphasizes that it does not want all companies to sound the same: their goal is for “one intelligence” to reflect each company’s identity and brand across every interface. According to avatarin, the era of single‑purpose AI is over and a new era of general‑purpose, multimodal intelligence — led in part by OpenAI technologies — is beginning.
Conclusion
The Kurashi‑Marugoto AI Agent’s two‑week public trial, with about 30,000 users and 92% positive survey responses, demonstrated that a GPT‑Realtime‑based, voice‑enabled, multimodal agent can extend in‑store expertise beyond opening hours, surface richer customer insights, and deliver an experience aligned with Yamada Denki’s brand.
