Skylark Group serves up better dining experiences with generative AI and Azure OpenAI

In-store generative AI delivers better customer experience with fresh points of contact

With over 20 famous Japanese brands like Gusto, Bamiyan, and Jonathan’s, Skylark Group controls around 3,000 family and buffet restaurants, confectionery stores, delis, and other food-related businesses. Its stores have expanded from Japan to Taiwan, Malaysia, and the United States, delighting around 350 million customers every year. IT-based initiatives, including the introduction of digital menus in 2020 and a cat-shaped delivery robot in 2022, are a key part of its success. 

The group recently took up new challenges to create fresh customer contact points using generative AI. It introduced a system consisting of Azure OpenAI Service and other components at Gusto Akihabara and conducted a proof-of-concept project called “Co Store Manager.”

“We’ve been promoting IT-based information integration to streamline store operations,” says Yutaka Ikeda, Director of Skylark Holdings’ Customer DX Group. Although the initiatives have improved customer convenience and streamlined store operations, they’ve left fewer points of contact between customers and workers. “The purpose of Co Store Manager is to enhance customer experience with exceptional human service while using generative AI to complement it,” explains Ikeda.

A concept for the solution emerged in December 2023. Skylark Holdings started building a system in March 2024 and began trials at Gusto Akihabara in September. The figure below shows an overview of this system, in which in-store tablets work with Azure services to provide digital menus.

“Our basic policy is to always compare multiple options to choose the best,” says Ikeda, explaining the Azure selection. “Azure OpenAI Service is evolving the fastest of the three mainstream generative AI services, and the speed of this evolution is extraordinary. We were already using ChatGPT on a daily basis, and the advent of GPT-4o significantly reduced hallucinations. It was worth trying out in a brick-and-mortar store.”

A chatbot that understands the subtle nuances of Japanese for polite and clever conversations

Co Store Manager has two main functions. The first is AI Robo, which talks with customers and recommends meals via the digital menu. The other function automatically creates daily reports by summarizing conversations with customers. Of the two, the AI Robo is particularly interesting.

Usually, when a customer opens the Gusto digital menu, two options appear on screen: “Select meal from menu photos” and “Select meal with menu number or coupon number.” The Akihabara station store has an extra option—”Talk to AI Robo to select meal.” This option lets customers choose their food by chatting with AI Robo via an on-screen keyboard or voice input. 

Although the digital menu is available in Japanese, English, Chinese, and Korean, conversations with generative AI are limited to Japanese or English. Yoshie Fujimoto, Leader of the Marketing Division’s Menu System Design Team, describes the conversation content and use of Azure OpenAI Service below.

“We continually checked conversations before in-store deployment. Azure OpenAI Service demonstrated better comprehension than other generative AI and grasped the subtleties of Japanese. Its conversations are friendly and polite. It seems to understand situations brilliantly and you really feel like you’re speaking to a human staff member.”

The source data for conversations is a combination of the generative AI model’s own learnings and master data of Gusto menus. Azure Cosmos DB stores this menu master data, with RAG searches implemented through Azure AI Search.

Conversation logs are checked daily and prompts are tuned as needed. “We have a significant number of tokens as a result,” says Takahiro Kurebayashi, a software architect in the Customer DX Group. According to Kobayashi, his team is currently organizing and simplifying prompts.

Entertaining conversations with AI Robo attract repeat customers

“The robot presents three products for every recommendation, which really shows the intelligence of Azure OpenAI Service and how it doggedly pursues its mission,” says Kurebayashi. Even more interestingly, many customers enjoy conversing with AI Robo.

Its character, animation, and backstory are a big part of its success. One day, an Arctic sea snail entered a tablet glittering in the ocean, journeyed through the sea of the internet, and started working at Gusto, hence its “in-training” badge.

“We tried to make AI Robo likable and slightly less quirky by making it seem a little bit hopeless,” says Manami Nakazaki, a UI/UX worker from the Web and Applications Direction Team. Because of the robot’s approachability, customers often ask it about its origins or request an interesting story.

According to Nakazaki, “Some customers have become regulars because of the opportunity to chat with AI Robo. They say ‘Thank you’ or ‘I'll be back’ and tell it their own nicknames to use in the conversation. Some customers even play quizzes and word games with it.”

AI Robo can also reprimand customers who use inappropriate language, some of whom apologize immediately. The emotional exchange between humans and a robot is like a sci-fi movie.

“If we encourage people to casually chat with AI Robo, we’ll get more feedback on our stores than ever before,” says Fujimoto “We also believe that having someone to talk with while waiting for an order will improve customer experience.”

Ongoing pursuit of the possibilities of generative AI

Although this use of AI Robo is creating new customer experiences, according to Ikeda, “It’s still too early to say that it’s actually enhancing customer experience. However, marketing teams and store crews are gaining a new awareness of generative AI.”

“There has been a vague unease about using generative AI because of hallucinations and the like,” Ikeda continues. “However, in-store use showed we could control it properly. We initially aimed to improve customer service, but generative AI’s excellent language conversion abilities make it versatile for many use cases. Even store workers have proposed new uses.”

One idea is to use the technology in kitchens. Because cooks need to memorize new recipes upon menu changes, they’re keen to use generative AI to provide the information. Developers have already begun working on a kitchen version of AI Robo to deliver a solution.

There are also other challenges on the horizon, which Kurebayashi explains below.

“We’ve implemented a mechanism which summarizes conversation logs into daily reports, and we’re beginning to abstract that content. Additionally, real-time conversations would be more interesting than question-and-answer type chats. Azure OpenAI Service is evolving fast and always ahead of the competition. We’re also excited about OpenAI o1, which has amazing reasoning capabilities.”

After the trial concludes, the company will consider expanding the solution to more stores. It will be intriguing to watch the improvements in store operation, customer service, and worker support that generative AI will achieve.