A Voice Agent Handled 30,000 Shopping Conversations in Two Weeks. 92% of Users Rated It Positively.

Roughly 30,000 people used a voice-based AI shopping agent during a two-week public campaign on Yamada Denki’s online electronics store, and 92% of survey responses about the experience were positive — a real adoption-and-satisfaction number, not a vendor projection. The agent, nicknamed “Kurashi-Marugoto” (“Total-Living”), was built by avatarin — an AI customer-service company spun out of ANA Holdings — using OpenAI’s GPT-Realtime voice model, and it translates experienced sales associates’ product knowledge into natural voice conversations that guide shoppers from discovery through purchase decisions. OpenAI’s role went beyond model access: the company says it worked directly with avatarin on structuring complex prompts and optimizing API costs for an always-on voice service, freeing avatarin to focus on translating Yamada Denki’s specific service model and brand voice into the agent’s behavior rather than solving infrastructure problems from scratch.

The 92% figure is notable given how unevenly voice AI tends to land with customers when it’s used as a cost-cutting front door rather than a genuine service upgrade. avatarin’s framing — replicate what a good sales associate already does, at scale, rather than replace the interaction with a worse one — appears to be doing real work here.

It’s also a useful contrast against how far behind adoption still trails comfort in adjacent service industries. A SevenRooms report on restaurant AI found 75% of diners are comfortable using AI for reservations and phone bookings, yet only 28% of restaurant operators have actually deployed it — even though 64% of diners prefer booking by phone and 40% of those calls currently go unanswered. avatarin’s 30,000-user, 92%-positive campaign is what closing that exact gap looks like when a company commits to voice AI as core infrastructure rather than an experiment: real usage volume, and real evidence customers accept — even prefer — the experience once it exists.

For consulting engagements evaluating retail or hospitality voice-agent deployments, avatarin’s numbers are a concrete benchmark for what “worked” can look like, and the SevenRooms gap is the reminder that most competitors haven’t gotten there yet.