24/7 AI Retail Agent: The Yamada Denki Case with GPT-Realtime

The timeline: from zero to 30,000 interactions in fourteen days

The project grew out of a collaboration between avatarin, a Japanese startup specializing in telepresence and physical agents, and OpenAI. The goal was clear: eliminate downtime in customer service at Yamada Denki stores. Yamada Denki is one of Japan's largest consumer electronics chains, with customers of many different languages and age groups.

Avatarin integrated GPT-Realtime — OpenAI's model optimized for low-latency voice conversations — into a physical agent deployed in stores. The result is documented directly by OpenAI in its official case library: 30,000 interactions in the first two weeks, with a 92% satisfaction rate among those who answered the post-interaction survey.

The system also handles requests in multiple languages without manual setup for each one. This matters a lot: Yamada Denki serves a growing number of tourists and foreign residents. A multilingual agent reduces reliance on specialized staff and cuts the cost of scaling.

The solution's architecture: why GPT-Realtime suits physical retail

GPT-Realtime is designed to handle streaming audio with lower latency than standard models. In retail, every second of perceived waiting affects the experience. So choosing a real-time model isn't a cosmetic decision: it keeps the conversation flowing.

Avatarin's agent combines three technology layers. The first is the language model, which interprets the request and generates the answer. The second is the physical presence layer — the robot or terminal — which gives the agent a visual identity. The third is the system that connects to product data and shelf availability.

In fact, an in-store conversational agent without access to real inventory would be useless. The real technical complexity lies not in the language model but in integrating with existing management systems. Every retailer has to tackle this before considering a similar deployment.

To explore AI architectures that work in the Italian context, the SHM Studio — AI services team offers a preliminary assessment of the integrations required.

Winners and losers: who really gains from this model

The most obvious winner is the retailer. A 24/7 agent removes the cost of night or holiday staff for basic information tasks. It also scales without significant marginal costs: the fiftieth simultaneous user costs the same as the first.

But there are less obvious losers. The first is the retailer who adopts the technology without a clear customer journey strategy. A poorly configured agent, with generic answers or no link to real inventory, creates frustration rather than satisfaction. Yamada Denki's 92% isn't automatic: it's the result of careful design.

The second potential loser is sales staff who aren't retrained. The AI agent handles standard information requests. As a result, the value of human staff shifts toward complex advice and closing the sale. Those who don't evolve their role risk gradual marginalization.

Finally, there's a third scenario: the competitor who stands still. In mid-complexity retail — electronics, furniture, DIY — the first to roll out a reliable AI agent builds a perception advantage that's hard to catch up with.

SHM Studio's take: real scalability or an isolated case?

The question Italian marketing managers are asking is a fair one: can a Japanese case be replicated in Italy? The answer is nuanced.

On one hand, the underlying technology is identical. GPT-Realtime is available through the OpenAI API to any developer or agency. So the technical barrier isn't geographic. On the other hand, the context is different: Italian customers have specific expectations around relationships, and an agent that feels cold or mechanical can harm the brand instead of supporting it.

At SHM Studio, we see that successful cases share one thing: experience design comes before the technology choice. First you define what the agent should do, which questions it must handle, and what tone it should use. Only then do you select the model and build the integration.

According to McKinsey's research on the economic potential of generative AI, retail is one of the sectors with the highest value to capture from conversational automation. But capturing that value takes integration with existing processes, not just installing a chatbot.

What it means for Italian retail: three concrete scenarios

For marketing leads weighing a similar initiative, it helps to distinguish three levels of implementation maturity.

• Basic level — digital information agent: a chatbot on the e-commerce site or app, powered by GPT-Realtime or equivalent models, that answers questions about availability, shipping and return policies. Limited investment, impact you can measure within weeks. Suited to retailers with well-established digital traffic.
• Intermediate level — omnichannel agent: the same agent integrated across the digital channel and the physical store, with real-time inventory access. It requires integration with ERP or POS systems. The Yamada Denki case sits here.
• Advanced level — proactive agent: the system doesn't just answer requests but anticipates needs based on browsing behavior or purchase history. This level requires a quality data foundation and an integrated digital marketing strategy.

At every level, the critical factor isn't the AI model but the quality of product data and how well the tone of communication fits the brand. An agent that responds in a way that clashes with the retailer's positioning creates cognitive dissonance for the customer.

Metrics to track after go-live

Yamada Denki's 92% satisfaction is an output KPI. But process KPIs matter just as much to understand whether the agent is really working.

First, the containment rate: how many requests the agent resolves without human escalation. A rate below 60% points to training or scope definition problems. Second, the average resolution time: it must be lower than the equivalent human channel, otherwise the perceived value is zero.

It's also essential to track the conversational abandonment rate: how many users drop the session before getting an answer. A high rate signals perceived latency or irrelevant answers. Finally, the post-interaction conversion rate: in retail, the agent must contribute to sales, not just information.

To build an effective measurement system, Google Ads campaigns and SEO activities can be integrated with the agent's data to create a funnel that works consistently from acquisition to conversion.

What nobody talks about: the cultural context problem

The Yamada Denki case takes place in Japan. That's not a neutral detail. Japanese culture has a very high threshold for technology acceptance, especially when it comes to robots and automated agents in public spaces. The average Japanese customer is used to interacting with automated systems in settings where an Italian customer would expect a human being.

So replicating this case in Italy requires a change management phase aimed at customers, not just at internal staff. The launch communication needs care. The agent should be presented as an added service, not a replacement. The tone should be warm, not technocratic.

According to Harvard Business Review, the negative perception of AI agents in physical settings is often tied not to the technology itself, but to a lack of transparency about the system's automated nature. Stating openly that it's an AI agent builds trust rather than eroding it.

This is a principle we at SHM Studio also apply when designing content and digital communication: transparency is a competitive asset, not a weakness. If you want to explore how to build an effective narrative around these initiatives, our SEO copywriting and LinkedIn campaigns services offer concrete tools to communicate innovation to stakeholders and B2B clients.

Next moves: what to evaluate over the next six months

The avatarin-Yamada Denki case is a useful benchmark, not a model to copy. Every retailer has a different context: customer mix, catalog complexity, internal digital maturity.

So the recommended first step is an audit of your current customer experience. Where do repetitive requests pile up? Which questions get slow or inconsistent answers? Which channels go uncovered when fewer staff are on hand? These three questions define the right scope for a first AI agent.

Next, the prototyping phase should be short and measurable: two to four weeks, one channel, a limited set of intents. Yamada Denki's time-to-value of two weeks is not a technological miracle. It's the result of a well-defined scope from day one.

If you want to explore these opportunities with the support of a specialized team, the starting point is a consultation with SHM Studio. We work alongside marketing teams from defining the strategy through to go-live, with a focus on measurability at every stage. More resources and use cases are available on the SHM Studio blog and in the web development section, for those considering deeper integrations with existing systems.