Online retailer says, after yearlong delay, it finally has trained its in-store algorithms
Nearly a year after it was promised, Amazon.com Inc.’s AMZN cashierless convenience store is slated to open to the public on January 22, 2018
The new Amazon Go store, located in the base of Amazon’s main headquarters in Seattle, uses computer vision and machine-learning algorithms to track shoppers and charge them for what they select, thereby eliminating checkout counters.
In an interview last week, Dilip Kumar, vice president of technology for Amazon Go and Amazon Books, said testing with employees has trained the technology to work in the store, an experiment that is part of the company’s broader effort to reinvent how consumers shop.
Kumar declined to say whether Amazon will expand the Go concept, although he said the company has developed the technology to scale.
Amazon announced the new Go store with fanfare in December 2016, and said it would open to the public in early 2017. The opening was delayed, however, as the technology proved more difficult to master than expected, with glitches occurring when too many people were in the store or were moving too quickly, The Wall Street Journal reported in March 2017.
Amazon didn’t explain the delay at the time. According to Mr. Kumar, while the store was originally expected to quickly open to the public to gain extra traffic needed for testing, the company decided it had enough employees to teach the system instead.
That training helped Amazon Go’s technology better identify objects and follow the different speeds and patterns of shoppers, tasks Mr. Kumar described as particularly challenging in a crowd.
Some people “move in very unpredictable ways,” Mr. Kumar said. “You’re always bending down, you’re examining items, you’re picking things up.”
The Go experiment shows how Amazon is trying to transform shopping in physical stores after decades of pioneering retail online. Since 2015, the company also has added more than a dozen Amazon Books stores, which encourage customers to pull out their phones to scan covers for prices.
In August, Amazon completed a USD 13.5 billion deal to buy grocery chain Whole Foods, adding 470 brick-and-mortar stores to its portfolio overnight.
Amazon Go’s technology uses cameras throughout the store to track shoppers once they are inside, though it doesn’t use facial recognition, Mr. Kumar said. A customer entering the store scans his or her phone and then becomes represented internally as a 3-D object to the system. Cameras also are pointed at the shelves to determine interactions with goods.
Among the challenges for the technology was telling the difference between similar looking products—say containers of vanilla and regular yogurt. Adding to the complexity, when customers pick up products, they usually cover the distinguishing aspects of the label with their hands.
Some store associates are still needed. For example, customers purchasing alcohol must show identification.
Former Amazon executives say it likely would be difficult to scale the system to track people in a bigger store, and that it could take years to make it work in a larger store footprint. Still, they say it may make sense one day for Amazon to try to implement the technology more widely—either via additional Go stores or even in Whole Foods.
Kumar said there are currently no plans to introduce the technology in Whole Foods. He added, however, that every project should be expandable.
“We have this unwritten rule that whatever it is you’re building, you have to be able to scale it so that it covers significantly amount of more load than what you would normally ever expect,” Mr. Kumar said.