Amazon’s “GO” cashier-less retail store is now open to the general public and receiving much scrutiny. Overall, the concept seems both robust and well received by patrons and technology enthusiasts. Although a recent occurrence, in which a patron unwittingly received a free yoghurt, raises some important questions about cashier-less retail in general.
The motive behind cashier-less retail is compelling. For the consumer, it’s the full browsing experience without the inconvenience of the check-out line, and it fulfils the potential of cash-less transactions, through mobile-device payment. For the retailer, it offers novel un-attended retail formats, new ways of consumer engagement and potentially lower costs. Accordingly, many retailers like Kroger are evaluating the concept and many start-ups, like AiPoly, and retail technology providers are developing offerings.
On a technology level, open multi-user cashier-less systems require multiple elements: Authentication and Access - Connecting the physical shopper to a validated account and permitting access to the retail space; Tracking - typically a vision system to associate a shopper with a product selection by virtue of proximity; Inventory Monitoring - sensors, load cells, planograms, RFID tags and vision systems that help confirm the selection or de-selection of items by shoppers, and; Back-end Software for inventory tracking, order processing, cashless payment and data analytics.
On a structural and operational level, open multi-user cashier-less systems, like Amazon Go, or Suning’s Biu cashier-less store require consideration of layout and traffic patterns, shelving design, product placement and sensor/camera placement. Additionally, open systems will rely on human attendants for help, maintenance and inventory management. Closed ‘kiosk’ style systems like Bodega have an easier time, dispensing with customer tracking and always-present attendants.
On a system level, there are three security concerns designers have to address without unduly compromising shopper convenience; errors, theft and hacking.
Let’s go back to the free yoghurt and its implications. On opening day, an Amazon Go patron reported on social media that she was not charged for a yoghurt. Amazon’s official response was sanguine and appropriate: when the system makes a mistake in favour of the shopper, the benefit of the doubt goes to the shopper…enjoy a yoghurt on us. But what about the situation where the customer is mistakenly charged for a product they did not select or what if an incorrect shopper is charged for the lucky shopper’s yoghurt? Cashier-less systems are novelties and conventional habits are very well engrained. Without a cashier as the human interface for error adjustment, in a poorly designed system, the user risks a high degree of dissatisfaction, putting trust in the new model at risk. System developers and technologists need thoughtful selection of sensor platforms, clever and robust algorithms and consideration of human factors, usability and retail customer service. Here, the customer should always be right.
At the recent National Retail Federation’s Big Show in NYC, Beijing-based start-up Yi Tunnel displayed its AI-based vending system. This system requires development of a full training data-set on items for sale and relies on a camera to recognized images as they cross the threshold from vending machine to shopper. The system is impressively fast, with instant recognition of items moving in and out of the vending machine. I spent some time watching attendees trying the system and, almost without exception, people attempted to defeat the AI by obscuring the camera to sneak a beverage. The single sensor AI was easy to defeat. Shoplifting and related losses are an accepted reality of retail but cashier-less systems are at enhanced risk of theft without the familiar presence of a vigilant cashier and such systems may be an irresistible challenge for bad actors. A cost effective multi-sensor approach provides for a more secure system without compromising usability.
A quick survey of Amazon Go videos on YouTube shows an equal measure of reviews and creative efforts to ‘steal from Amazon’ and represents a dropping of the digital gauntlet for those seeking to defeat the sensing systems by attacking vulnerabilities at a system, communications or software level. A lot of work for a free soft drink, yes, but for motivated individuals, it’s about the notoriety and propagation of the hack, not the free goods. Design in adequate cybersecurity features and establish systematic responses to identify breaches and respond automatically and swiftly to devalue hacks.
It’s early days for cashier-less systems and far too early to judge the extent to which they will be implemented. Self-checkout systems, where shoppers use scanners or self-scanning systems, promised to deliver some of the benefits of fully cashier-less shopping but have fallen well short of mainstream adoption and so there must be some significant convenience or trust value with human interaction at check-out. With advances in sensing, vision systems and artificial intelligence, mobile personal device cash-less transactions and the emergence of a generation of shoppers who speak in terms of ‘time poverty’, cashier-less shopping will be a legitimate development. Success depends on thoughtful overall system design to identify and develop the correct balance between frictionless shopper convenience and retailer robustness.