The dirty secret of online shopping is that product recommendations are often terrible. They’re built on what you clicked, not why you clicked it. avatarin, an AI firm spun out of ANA Holdings, just proved there’s a better way by transforming the expertise of Yamada Denki’s sales associates into a 24/7 voice agent that had a 92% positive rating from roughly 30,000 users in a two-week trial.
The agent, called Kurashi-Marugoto AI Agent, runs on OpenAI’s GPT-Realtime. That choice was deliberate. avatarin’s CEO, Fukabori, wasn’t looking for a faster keyword bot. He needed a single model that could handle speech, text, and visual understanding with low enough latency for a real conversation. “For us, the performance of a single model went beyond what we had seen from specialized speech recognition systems,” he said. The goal was to flip the script—instead of forcing shoppers to adapt to a machine’s rigid rules, the technology would bend to the messy, uncertain way people actually talk.
What makes this different from a standard chatbot is the agent’s proactive design. A customer might say, “I need a refrigerator for a family of four, but my kitchen is small. Which one should I choose?” A conventional chatbot would choke on that. avatarin’s system, grounded in a retrieval-augmented generation setup for product accuracy, asks follow-up questions to uncover real needs and navigate shifting preferences. It’s not just answering questions; it’s guiding product discovery in a way that mimics the best floor associate who knows when to listen and when to nudge.
The trial data is compelling, but the deeper insight is what Fukabori calls the beginning of an “interface revolution.” Every conversation generated data on what made shoppers hesitate, what they cared about, and what helped them decide—insights that are invisible in a traditional clickstream. The fact that people were willing to discuss budgets and uncertainties with an AI agent after store hours suggests a new channel for retail that isn’t just about efficiency, but about candor. Whether that translates to actual sales conversion at scale is the next, much harder question.
