From instant noodles to AI: inside Ting Hsin's smart retail revolution

The Convenience Store & Restaurant Chain Business Group (BG) of Ting Hsin International Group manages numerous brands, including FamilyMart, Dicos, Master Kong, and Belray Coffee. With more than 6,000 stores nationwide, this BG engages in diverse sectors, including food services, convenience stores, food production and logistics, business intelligence, e-commerce, food ingredient procurement, and cross-border procurement.

The New Retail Service Company primarily provides membership and system-related services, including membership systems, point platforms, system services, software development, data analysis, and consulting, for Ting Hsin's various brands. In 2014, the company entered the membership management sector by establishing the Jixiang Alliance convenient life service platform. It also offers point-based membership service solutions for the Group's internal and external customers. Furthermore, the company offers integrated technical services spanning software development, system platform customization, and O&M.

How can AI create value across different retail businesses and scenarios?

Regarding food services: Key AI applications in this domain are self-service kiosks and smart add-on sales recommendations, both on apps and at restaurants. The latter application means that after customers place an order, the system will automatically recommend related products that customers may be interested in, which encourages customers to buy more products. This drives higher average spending per customer and helps improve store revenue.

As for convenience stores (FamilyMart): We see 2024 as the first year of AI adoption at FamilyMart stores. Through internal discussions, we identified multiple scenarios that were ready for AI, and focused on three to five core scenarios to deploy AI. Among them, intelligent ordering was identified as a key application.

Intelligent ordering helps staff determine what products to order every day?

Yes. Each store needs to order products of all categories every day, including standard products (e.g., bottled water) and short-shelf-life products (e.g., oden, cakes, and sandwiches). The AI-assisted ordering system takes into account multiple indicators—such as a store's historical sales, neighborhood stores, customer profiles (including ratios of online/offline purchases), and weather conditions—to calculate each store's recommended order quantities for all categories of products every day.

If store managers find system-recommended orders not appropriate, they can adjust them manually. Currently, the system's accuracy is 70–80%, and we are continually optimizing it with the aim to increase accuracy to over 90%. The system has already cut daily ordering times from 1.5 hours to 30 minutes, which gives store managers and staff more time to work on tasks like customer service and store operations.

Your New Retail Service Company primarily provides membership and system-related services, including membership systems, point platforms, system services, software development, data analysis, and consulting, for Ting Hsin's various brands. Currently, the New Retail Service Company is implementing the "AI Brain" project, aimed at improving business efficiency through digital and intelligent technology. How will the AI Brain project help Ting Hsin strengthen your competitive advantages?

We consider 2024 our first year of AI adoption, split roughly into three phases:

• Phase 1: We spent about six months building AI infrastructure. We worked closely with Huawei in key areas like computing power, algorithms, and servers and built a solid backbone platform. I want to thank Huawei for their great help and support.
• Phase 2: We explored and piloted AI applications at stores and restaurants, such as the intelligent ordering and add-on sales recommendation systems we've talked about.
• Phase 3 (2025 and beyond): We will integrate AI into more operating scenarios at retail and convenience stores. AI investment is strategically prioritized and fully supported across the Group. It will remain a key strategic area for us in the next three to five years.

Every industry is looking for the right scenarios to apply artificial intelligence. As a consumer-facing industry, retail focuses more on using AI to provide consumers with personalized shopping experiences. How do you think you can use AI to strike a balance between personalized experiences and the efficiency of mass production?

We can consider this question from two aspects.

Experience: An example of applications used to provide customized experiences is the AI-powered personalized recommendation function we've launched on the FamilyMart app. The app now uses algorithms to recommend different products to different consumers, based on each consumer's purchase history.

Hou Jun, General Manager, New Retail Service Company, Ting Hsin International Group

Production and supply chain: We are implementing projects such as AI-powered raw material procurement and inventory management. We plan to use AI to optimize the entire process from raw material procurement to store ordering. In this process, AI-assisted systems provide data-backed recommendations, and then people make final decisions based on a combination of the recommendations and their own experience.

While some companies pursue fully automated stores, we position AI as a tool to assist in decision-making. We believe that the key value of AI is to provide advice and support. For key business activities, we still need to combine human experience with AI to make the best decisions.

The example you just gave shows that streamlining sales data across convenience stores, supply chains, and production departments can enhance collaboration between production and sales teams. Is that right?

Yes. Currently, through the Group's internal systems, we have streamlined end-to-end data flows, from stores and logistics, to production and supply chains (including raw material procurement and factory production). This ensures data consistency and controllability within the New Retail Service Company.

Thanks to streamlined data and AI algorithms, the operating statistics of a single BG can be used to predict and guide the large-scale expansion of other affiliates within the Group. This in turn generates more accurate real-time data flows.

In addition, we can use the real-time sales data of stores to predict raw material demand levels for the next day or even week, and then make procurement and production plans accordingly.

How has the AI Brain project unified capabilities across your Group's diverse businesses, and how did you tackle data challenges?

Data governance started quite early at our Group. Back in 2018 or so, each business segment began building its own data warehouse. Then around 2021, they began conducting data cleaning.

In 2023, we hit a major milestone. After significant effort, we finally managed to streamline our data in different business segments and companies and completed the Group's data lake. During this process, Huawei teams supported us with data streamlining and cleaning. This was an immense task that took us nearly six months to finish.

Now, the quality and governance of our core data has been significantly improved, whether it's at FamilyMart, Dicos, or our production and logistics centers. Today, 80% of the Group's data has been integrated. Our focus now is analyzing core data and expanding AI applications.

What is your Group's stance on data governance and data asset utilization? For instance, would you consider opening up and sharing the data?

In the first half of this year, the Group discussed issues related to data governance and data assets. The Group views data as one of our most important assets. In addition to the BUs responsible for daily operations, the Group also has two core assets: the extensive membership system covering all our brands and the Group's entire data system.

We are actually building internal platforms to try out data sharing and utilization across business segments. For instance, the analysis of Dicos' store location and other data could help FamilyMart with new site selection and operational analysis. However, such data is too sensitive to be directly shared. So, we remove specific details from the raw data, and only share relevant general information with FamilyMart. This way, we make the most of the data while protecting the core information of related businesses.

Long term, the Group remains open towards data assetization and is closely tracking how this idea develops in China. I understand that some companies are already including data assets in their balance sheets, and are even trying to monetize them. We are open to external partnerships, but we have not taken any concrete steps yet. Right now, our main focus is still on building a good data sharing platform and mechanism within the Group.

Data desensitization, which you mentioned earlier, is a good approach, but it is still technically demanding. Would implementing this require a lot of work?

Yes, it will take some effort. The key is to find the right balance in desensitization. When sharing information between companies within our Group, we strictly prohibit sharing highly sensitive personal information like phone numbers. All information we share is anonymized.

How exactly is the data shared? Let me give you an example. Dicos provides general store location information, and the data platform supplements publicly available or processed foot traffic data from the surrounding area (this data itself is not sensitive information unique to Dicos). For operational data, we generally share broad store performance ratings, like "good", "average", or "poor", instead of exact daily sales figures.

Simply put, we convert raw, sensitive data into reference information or suggested values that do not involve specific details or categories. Our systems automatically handle the operations like tagging and grading, requiring minimal human involvement.