Machine Vision in Retail after the Unmanned Store Hype Cycle

For a period of time, the unmanned store was deemed to be the “next big thing” in technology. Chinese e-commerce giants such as Alibaba, Suning, and JD.com as well as Amazon in the United States have set up retail stores that are fully automated in major cities, equipped with various technologies such as electronic shelf label, Radio-Frequency Identification (RFID), computer vision, and facial recognition. According to ABI Research’s Smart Retail Market Data (MD-RET-102), published in October 2018, the total number of unmanned stores was forecasted to grow from 393 in 2019 to 44,138 in 2023 globally.

Since then, market momentum has stalled. Recent news reports have announced that the recent boom of unmanned stores has ended, and that deployment has slowed down, as the stores struggled to move fresh groceries. In the past three years, Alibaba is reported to have set up more than 200 unmanned stores in China, but many have since gone out of business. Investments in full-scale automated retail stores are very expensive. Online retailers such as Amazon, Alibaba, and JD.com are willing to invest in these technologies because they see the value beyond the retail experience. These cloud Artificial Intelligence (AI) giants amassed large amounts of customer insights and data that allow them to identify the correct mix of in-store items and preferred User Experiences (UX), further optimizing and improving their AI models. For now, the costs still outweigh the benefits.

However, not all is lost. The technological advancements propelled by the unmanned store boom have found a second life. Unmanned store is an early concept that was ahead of its time, but many of its key components are not. Instead of being deployed in a greenfield environment, computer vision and facial recognition have now been integrated into the existing retail landscape, from storefronts all the way to the backend. Due to their rising maturity, these technologies can work seamlessly with existing layouts and product selections, alongside other more established retail technologies such as RFID for source tagging, automated self-checking Point of Sale (POS) terminals, and smart vending.

According to ABI Research’s  Artificial Intelligence and Machine Learning (MD-AIML-104) Market Data, the total installed base of AI-enabled camera systems in retail is expected to grow from 420,000 in 2020 to 2.37 million in 2024. Trax, a Singapore-based startup, provides in-store AI-based image recognition solutions for retailers to monitor their shelves and predict future demands. Pensa Systems goes a bit further; by deploying its shelf monitoring solution on mobile apps and in-store drones, it is able to reduce stockouts, develop visual representations of the shelf display, identify pricing strategies, and optimize inventory systems.

In the case of autonomous checkout, California-based leading retail technology startup Standard Cognition continues to strengthen its AI-powered computer vision platform in Brick and Mortar (B&M) stores through the acquisition of DeepMagic, a deep-learning computer vision startup, and Explorer.ai, an indoor mapping solution provider. Similarly, AiFi, another California-based startup, introduced a scalable autonomous solution with major retailers such as Carrefour (France), Albert Heijn (Netherlands), Valora (Switzerland), and Żabka (Poland). Both startups rely on a combination of AI-based machine vision, cameras, edge computing, and sensor fusion technology. On the security front, U.K.-based startup ThirdEye is developing context-aware alert systems to detect relevant events in CCTV streams for retailers.

Interestingly, even after the unmanned store fever has died down, retailers are still actively investing in this domain. Amazon Go, the concept that started the hype, launched its first full-blown cashier-less supermarket. The supermarket now features about 5,000 items and 10,000 square feet of space, with a fresh produce section, a bakery, and an alcohol section. There are still employees in the store to serve as greeters and stock associates, but core functions, such as store monitoring, checkout, and payment, rely heavily on AI technologies.

In Asia-Pacific, there are also signs of development outside of China. In November 2019, Singapore-based Octobox started trialing an unmanned store in the National University of Singapore. At the same time, Saha Group, Thailand’s leading consumer goods manufacturing conglomerate, has committed nearly US$1.7 million to digitalize its business and bring the first unmanned store to Thailand.

It is definitely too early to pronounce the death of the unmanned store. ABI Research believes e-commerce vendors will continue to refine their technology stacks and the market will rebound in foreseeable future. Rising wages, labor crunch, and even an epidemic like the recent COVID-19 pandemic expose the weaknesses of labor-intensive retail supply chains. Unmanned stores augmented with robotics and machine vision solutions might present themselves as the solution to these challenges.