SEATTLE — Amazon Go is a go for the masses.
The retailer’s cashierless convenience store opened its doors to the public Monday, a debut that follows a nearly 14-month trial run open only to the Seattle company’s employees.
The store requires customers to scan their smartphone on the way in, tracks them with cameras and other sensors as they browse, and, when they take an item off the shelf, adds it to a virtual cart. Groceries are charged to the customer’s Amazon account when they leave with their goods.
At 7 a.m. Monday, the store opened to anyone with the Amazon Go smartphone app and a linked Amazon account.
Amazon Go is among the boldest efforts by the online retailer to reshape brick-and-mortar shopping. For a pocket-size, 1,800-square-foot convenience store that hasn’t technically opened to the public, the concept has had an outsized influence on the retail industry since its surprise unveiling in December 2016.
The concept, which Amazon has termed “Just Walk Out” shopping, sparked speculation that Amazon could use its high-tech concept as a beachhead to expand into convenience stores or perhaps other categories of physical retail.
It was also criticized by grocery-store workers’ unions, which feared an effort to automate the work done by cashiers, the second-most-common job in the U.S.
Amazon has said the goal isn’t to make retail employees redundant, but to offer convenience you can’t get from a sometimes-crowded deli or corner store.
A visit last week to the store, occupying a corner of the ground floor of Amazon’s Day One skyscraper at the corner of Seventh Avenue and Blanchard Street, offered a demonstration of the concept shortly before the lunchtime rush of Amazon employees.
A few workers stocked shelves; others milled about, ready to help customers with questions. One employee stood sentry near the corner with beer and wine, to check IDs.
In a kitchen in the back, a half-dozen staff members were at work slicing vegetables for the sandwiches, salads and other take-away lunch options on shelves at the front of the store.
Gianna Puerini, the Amazon vice president who oversees Go, said in an interview that the aim is for the store’s prices to match other markets.
At first glance, prices seem comparable to small stores. A 12-ounce can of Springdrift sparkling water was $1.25, a 20-count tea pack from Coffee Bean and Tea cost $7.75, and, in a long cooler in the front geared toward take-away meals, $5.99 got you house-made tofu banh mi.
The range of items available lands somewhere between that of a small gas-station convenience store and a mainline grocer. Amazon had four varieties of coconut water on hand, but there is no hot food selection, and few items beyond snacks, drinks and basic groceries.
“It takes selection, price and convenience,” Puerini said. “I think we’re delivering on all three.”
Of course, Amazon Go isn’t the company’s only experiment in physical grocery retail. Amazon spent about $13.5 billion last year to scoop up more than 460 Whole Foods locations.
Also last year, the company opened two Seattle sites for an experimental curbside grocery delivery program, and, though it was recently scaled back, the Amazon Fresh branded grocery delivery service is still being offered.
Few expect Amazon’s first Go store to be its last. “We’d love to open more,” Puerini said, but she stopped short of announcing any expansion plans. (Amazon may have foreshadowed its intentions anyway, posting a job opening for an experienced real-estate manager — travel required — last year.)
Amazon likely isn’t aiming for a Go store on every street corner. The company’s internal projections, according to someone familiar with the early stages of Amazon’s plans, determined that a store needed thousands of office workers within a few-block radius to make the investment worthwhile.
A focus on office workers would “cater to what the store does well,” Puerini said. “People pressed for time and hungry.”
Much of the technology that makes the place work lives above the store’s shelves.
Dozens of brick-shaped black devices, about the size of a paperback book, are suspended just below the ceiling. Set at slightly different angles, they are featureless except for a single aperture aimed at a portion of the store.
These devices use “a combination of sensor inputs,” Puerini said, likening it to the systems that help self-driving car prototypes identify the people and objects in their field of view. (That is, a combination of video cameras backed by technology built to analyze images, and laser arrays.)
Getting the system right took a while.
When the store was announced and opened to employees in December 2016, Amazon expected to open it to outsiders in early 2017. The company didn’t meet that deadline, with The Wall Street Journal reporting that the system crashed in tests when the store was crowded.
One engineer formerly assigned to Amazon Go said in an interview that the company struggled in its early efforts to get its computer vision algorithms to reliably track customers.
Behind Amazon’s automated people-tracking technology, the person said, teams of real humans had to review video footage to make sure the systems were functioning properly — at least initially.
The company evidently thinks it has ironed out those kinks. In fact, Bloomberg News reported in November that three Amazon employees at one point donned Pikachu costumes to try to fool the tracking system. Amazon’s algorithms charged the correct amount to each fuzzy character, the news service reported.
Puerini didn’t address any technical reasons behind the delay.
She said part of the reason for Amazon Go’s long incubation was the surprisingly high use of the store by Amazon employees, which essentially simulated real-world conditions well enough that the company didn’t have to open it up to the general public as soon.
How often does the system make mistakes? Puerini wouldn’t go into it, beyond saying “the system is highly accurate.”
If there is an error, Puerini says, customers can swipe on an item on their receipt within the Amazon Go app to remove a charge for something they didn’t take.