A newly installed machine by the entrance to the elevator lobby has become a regular stop for tea enthusiast Li Pei and her colleagues during tea breaks and overtime sessions.
The device is about the size of a standard freezer, with no high-definition display screen, replaced instead by a QR code for placing orders via a mini program. It offers a surprising variety of items, including mineral water, tea beverages, coffee, instant noodles, spicy snacks, bread, pastries, and dairy products, effectively functioning as a mini version of an unmanned convenience store.
With traditional vending machines, which require clicking a screen, selecting an item, scanning a code or inserting coins, and waiting for the item to drop, the process is lengthy and cumbersome. Today, in office parks, commercial buildings, and the central business districts across first- and second-tier cities in China, the older generation of vending machines is being replaced by a new and smarter alternative: vending cabinets.
These cabinets provide a broader product range and a more straightforward transaction process. In situations like impromptu meetings or when there’s no time to wait for delivery or go downstairs to a store, Li opts for an energy bar or a piece of bread from the cabinet. Users simply scan their face or the QR code, open the cabinet to select their items, and close the door to complete the transaction automatically. The system even suggests coupons based on consumption preferences, greatly satisfying immediate consumer needs.
According to reports by the Zero Power Intelligence Research Institute, in 2021, the market size for unmanned retail stores in China was approximately RMB 28.27 billion (USD 4 billion), reflecting a 24.6% year-on-year growth. By 2025, the market is expected to approach RMB 2 trillion (USD 280 billion), serving 250 million consumers.
But meeting user demands and preferences without human intervention—understanding consumer behavior and providing personalized services through a compact machine—relies heavily on artificial intelligence. Data analysis for optimizing product layout and recommendation strategies has become a critical challenge for the operation and development of smart vending cabinets.
How do they work?
Smart vending cabinets cater to instant retail needs.
Before food delivery became widespread, convenience stores were the go-to choice for quick purchases, requiring 15–20 minutes to grab a drink from downstairs. While food delivery resolved spatial limitations by offering a broader product selection, it extended waiting times to an average of 40 minutes.
For someone with only a ten-minute break, when delivery takes too long and convenience stores are far away, what are the options? Smart vending cabinets address these needs.
While many still associate unmanned retail with staircase-side vending machines, blind box machines at airports, or direct-sale cabinets from beverage companies, smart vending cabinets—the smallest units of offline retail—are becoming increasingly intelligent. They can analyze consumer preferences, predict future sales, and even recognize when it’s time to introduce new products.
Feng’E Zushi, a specialist in AI-powered retail, focuses on predicting market supply and demand and matching user needs.
The cabinet’s intelligence operates on two levels.
The first is frontend intelligence, which dictates the interaction experience of users, including smooth mini program usage, accurate camera recognition, and streamlined transactions.
The second is backend operational intelligence, where algorithms are applied to analyze and predict consumer behavior to adjust inventory and product structure in advance. Algorithms enable real-time decisions tailored to specific consumption scenarios, helping companies achieve stronger terminal competitiveness at the lowest cost.
For instance, in restocking, algorithms can replace the need for on-the-spot decision-making by logistics personnel. The supply chain simply follows pre-planned routes, delivering products tailored to consumer needs, leaving decision-making to the algorithm.
AI in unmanned retail optimizes product strategies and enhances operational efficiency. In smart cabinet applications, this means aligning supply with demand through historical sales data analysis. The “sell-to-buy” model, which adjusts purchases based on sales, replaces the traditional “buy-to-sell” approach, which pushes pre-purchased inventory to consumers.
“The fundamental difference is that the traditional model pushes products to consumers, while the new model pulls products based on consumer needs,” said Zhu Tao, COO of Feng’E Zushi, in an interview with 36Kr.
Algorithmic decisions depend heavily on accurate data and correct models, but some factors inevitably escape the algorithm’s reach. Therefore, practical applications must combine offline data and historical sales analysis.
The advantage of this data-driven model lies in its ability to provide real-time feedback. Users can interact directly with customer service via the system backend, offering genuine feedback that informs iterative upgrades of recommendation algorithms, delivering more accurate and timely products and services.
At the same time, the machine will require a diverse selection of products to meet instant needs that can differ depending on the time of day. This includes drinks, snacks, and ready-to-eat meals, all needing to be flexibly and rationally combined.
“In actual operations, many companies prioritize maximizing sales of high-margin products rather than enhancing the overall consumer experience. However, our users frequently tell us that the first impression when opening a Feng’E Zushi cabinet is the variety of products—almost like a mini supermarket,” Zhu said. “Our focus is to address broader needs, such as hunger, thirst, or craving for snacks, with a comprehensive and nuanced selection.”
Given the fixed capacity of smart vending cabinets, Feng’E Zushi’s R&D team introduced an expansion model to maximize space efficiency. This involves combining different types of products, adjusting displays, and tailoring offerings to the distinct preferences of users in different areas. As a result, nearly every vending cabinet has a unique product mix and capacity.
For example, in an office with 30–40 people, where meetings or overtime work occur sporadically, the cabinet might stock bread and instant noodles alongside beverages to cater to irregular mealtime needs. Meanwhile, in parks or sports facilities, tea, bottled water, and snacks are more popular.
“While selling only water might be more profitable in some locations, catering to the diverse needs of the audience ensures a better user experience. Effective algorithmic decision-making must balance rational data analysis with a consideration for human needs, creating optimal solutions that reflect care for all consumers,” Zhu said.
Lower costs, quicker expansion
The development of unmanned retail in China began in 2016. After Amazon Go’s groundbreaking entry with its grab-and-go shopping model, the market drew significant attention from investors and major retail giants.
However, rising operational costs, sluggish sales, and the challenges of supply chain complexity and regional management fueled widespread discussion about the cooling off of the unmanned retail market.
Following years of consolidation, the industry has now entered a relatively mature phase, with companies shifting their focus to AI-powered solutions.
Zhou Qi, an investor with long-term insights into retail, told 36Kr that the cost of a single vending machine typically exceeds RMB 20,000 (USD 2,800). “Every component—tracks, motors, delivery channels, and external order screens—adds to the cost. The equipment is also bulky, making relocation and repurposing difficult.”
In contrast, the smart cabinets used by Feng’E Zushi have significantly reduced costs. According to 36Kr, current market costs for smart vending cabinets have dropped to just a few thousand RMB, making them much more competitive than traditional vending machines.
“These cabinets have a similar seven-year lifespan as more expensive devices but leverage intelligent solutions to lower operational costs,” Zhu said. With self-service pickup, users can open the cabinet, take the desired item, and place orders faster. This simplified structure reduces mechanical complexity and costs while occupying less space.
The hardware integration employs standard refrigeration components such as compressors, sensors, and temperature controllers, similar to those used in traditional refrigerators. Utilizing supply chain resources from traditional manufacturing industries provides significant advantages in cost efficiency and large-scale production capabilities.
“Whether in terms of equipment, labor, or logistics, our approach allows deeper penetration into retail scenarios. This ultimately translates to staying closer to consumers than competitors,” Zhu said.
Over the past few years, the unmanned retail market has experienced intense competition, leading to consolidation among brands and companies. Many players have exited the market, but notable firms like Ubox, Nongfu Spring, and Genki Forest have aggressively expanded their presence to secure advantageous positions.
However, the competition for location dominance is not just about securing territory—it is fundamentally about fulfilling consumer demands. As user preferences grow increasingly diverse and personalized, the unmanned retail space expects to see more applications of AI-driven solutions, expanding its reach into new scenarios.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Huang Nan for 36Kr.