Checking Out Amazon Go, The First No-Checkout Convenience Store

Every part of the U.S. has a different local term for a convenience store: the bodega, the corner store—even “the Wawa,” a chain name that Northeasteners use generically. Now Amazon wants to extend its brand to the notion of a grab-and-go shop with Amazon Go, a store that literally lets you grab and go. On Monday, more than a year after the company unveiled the concept and began a beta-test phase open only to its own employees, the first Amazon Go in Seattle will welcome all shoppers.

In December 2016, when the company first teased the automated store–which eliminates cashiers and checkouts in favor of AI and cameras that detect the products you select—it said that it was coming in early 2017. Last March, however, the Wall Street Journal’s Laura Stevens reported that Amazon was having trouble getting its “Just Walk Out” technology to work as the place filled up with customers. Amazon Go’s VP of technology, Dilip Kumar, who gave me a pre-opening tour, acknowledges that Amazon had expected to let in non-Amazonians earlier. But he says that the delay was because its employees embraced the beta program, providing enough willing participants until the store was ready to exit its beta phase.

For now, Amazon Go has a single location deep in Amazon’s own world. It’s on the ground level of Amazon’s Day 1 skyscraper, next to its Spheres, which are sort of giant-scale terrariums. (Kumar says that the company expects to roll out more locations on an unspecified timeline.) The store has been highly visible from the sidewalk since it opened to employees, with windows into the kitchens that let passersby watch staff make fresh prepared food and meal kits.

Amazon Go works like—well, like a physical manifestation of Amazon’s 1-Click checkout, where you “click” by taking an item off a shelf. On arrival, you launch the Go app, which comes out today for iPhones and Android phones and connects to your Amazon account. It displays a 2D code that you scan at one of several glass security gates. The code identifies you to the store and opens the gate. (You can also check in other people—a spouse, a kid, a friend—whose purchases will be added to your tab.) Once you’re in, AI algorithms start to track you and everything you pick up and keep. You can bag your items as you go if you so choose, and need interact with an employee only if you’re buying alcohol, in which case an associate standing in the liquor area will check your ID.

Using a phone provided by Amazon, I tested the system by picking up a can of LaCroix water and leaving the store. It was a non-event, which is sort of the point. The experience doesn’t feel like an act of advanced technology unless you scan the ceiling and notice the hundreds of matte-black cameras surveying the shopping floor below. “You can just walk in, take what you want, and leave,” says Kumar.


Just how the store keeps track of you and your purchases isn’t clear. Kumar deflected questions about AI back to the customer experience, and the company has provided little detail about how the Go technology works behind the scenes other than saying it’s developed software to perform tasks such as identifying products—whether they’re on a shelf or off it, and even if they’re partially obscured by something else. (The store doesn’t use more conventional technologies such as RFID tags to track items, but can take a cue from a normal barcode on a product.)

“You use machine learning and use computer vision in a way that makes this experience completely seamless,” says Kumar. “We have spent a lot of time figuring out how to make our algorithms and our sensors reliable, highly available, and very efficient so that you get things right and we’re very accurate.”

Despite the Wall Street Journal’s report of early glitches, Amazon’s back-end system is ostensibly now powerful enough to handle a store full of shoppers, up to maximum capacity as established by the fire code. Kumar says that Go won’t impose any limit lower than that. During my tour, a constant stream of Amazonians passed through the store. They weren’t plants, and I didn’t see any frustrated purchasers.

The necessary processing capacity and software to track dozens of people and their purchases would have sounded implausible a few years ago. But compared to driverless cars—already an impending reality, involving chaotic uncontrolled situations with unpredictable numbers of people, vehicles, and other stuff—monitoring known items in a fixed location seems a lot more straightforward.

from Fast Company: