From an undisclosed location in the town of Hudson, Amazon is planning its latest assault on traditional retailers.
This one involves using cameras and sensors to replace the cashier — and the lines — at a convenience store. Just grab the items you want, and the system automatically tallies up your bill and charges your credit card as you exit. Amazon’s tagline for it is “Just Walk Out” shopping.
While the first of these stores, called Amazon Go, is set to open to the public in Seattle later this year, a team in Massachusetts is developing much of the technology that will make it possible.
Led by former entrepreneur Jeremy De Bonet, the team has been growing since the fall of 2014 and is now estimated to number about 20 people, though Amazon spokeswoman Lori Richter wouldn’t confirm the team’s size — or even that it exists.
A recruiting flyer from 2015 alluded to an “Advanced Projects Group in the Boston Metro West area” working on “an ambitious project.” Around that time, De Bonet showed up to a gathering of computer vision experts — that’s technology to allow computers to “understand” images from the real-world — to pitch people on joining him. “He told people, ‘We have this big audacious project and it’s super-secret,’” says Samson Timoner, an organizer of the Boston Imaging and Vision group. “`I can’t tell you about it, but once you see it, it will be amazing.’” Anyone who was invited for an interview had to sign a non-disclosure agreement that prevented them from discussing what they’d been told.
A chipper video that Amazon released online in December showed how an Amazon Go store would work: scan your mobile device at the entry to let the store know you’ve arrived, and simply fill your basket with pre-made salads, sandwiches, sodas, or a cupcake. (Yes, you can decide to put an item back on the shelf.) Cameras in the store, possibly augmented by sensors in the shelves, track your path and “see” what you’ve chosen. Once you leave, the itemized receipt shows up on your phone. The store itself looks more geared to meals-on-the go than full-week grocery shopping.
“My initial thought when I saw the video was, ‘Wow, this is hard,’” says Timoner, who holds a doctorate in computer vision from MIT; he arrived at the computer vision department there not long after De Bonet departed without completing his PhD. “There are so many complications in a real-life store, where people are blocking some of your camera views.”
Despite the fact that the first Amazon Go store in Seattle has been open for several months to Amazon employees and other invited testers, Timoner says he suspects that the technology isn’t yet perfected. “I’d bet they’re up to the point where they’ve got it pretty good, but need to iron out a lot of things, and the test store will let them do that.”
Amazon hasn’t said anything about the role that the Hudson team is playing in the development of the Go technology, but the same was true for another Amazon research and development group in Cambridge, which built much of the software that allows Amazon’s intelligent speaker — a.k.a. Alexa — to respond to spoken commands, like, “Alexa, what was the Celtics score last night?”
But according to recent job postings, Amazon may be doing more than designing hardware and software for the Go store in Massachusetts. It also may plan to make some of the equipment here. An online ad for a manufacturing technician spelled out Amazon’s need for someone with “fabrication, machine tool operation, and assembly” experience. Setting up a manufacturing line in Massachusetts wouldn’t be unheard of — Amazon in 2012 acquired a North Reading company, Kiva Systems, that builds thousands of heavy-duty robots to schlep merchandise in Amazon warehouses.
Working alongside De Bonet on the Amazon Go team is Oded Maron, who earned a PhD in computer vision at MIT. Maron also worked with De Bonet on a startup called CoPiloted, which offered advice about planning for retirement.
There are still plenty of questions about how significant Amazon Go will prove to be. How many locations does the company plan to open? How will the prices compare with a Cumberland Farms or Whole Foods? And will it be worth spending millions of dollars on new technology just to eliminate a couple of minimum wage cashier jobs at each Go store?
“Clearly, their payroll model is very different from what it would be in a mom-and-pop corner store or grocery store, and it’s going to have an impact,” says Jon Hurst, president of the Retailers Association of Massachusetts, an industry group. “I’m not as worried about the big chains who have internal and external technology consultants who can replicate the same ideas. My worry is, is it affordable to the small retailer?” And Hurst says that as states discuss raising the minimum wage for hourly workers like cashiers, that elevates the cost of running a store just as “technology is helping the behemoths operate with fewer workers.”
Doug Stephens of the retail consultancy Retail Prophet says that “the payoff could be huge” for Amazon if it figures out how to run a bricks-and-mortar market with fewer workers — and no lines. And if Amazon Go’s sensors and cameras are truly able to track the location of every item in the store, keeping more accurate tabs on inventory and reducing theft may be side benefits.
“The grocery industry has a net profit margin of 1 percent,” Stephens says. “It’s an industry that is as close to being underwater as you can possibly be. If I’m Jeff Bezos, I’m willing to disrupt this industry — and even lose money in the process — so that I can take a good swath of the incumbent players completely out of the equation.” Stephens says that the technology developed for Go could also eventually be applied in other types of stores, like the bookstores Amazon has been opening in Dedham and Lynnfield.
Stephens says he doesn’t think of Amazon as a retailer. “They’re a data innovation and technology company that happens to sell things — and they have a huge war chest of cash to blow on experiments.”
The team out in Hudson will be key to the latest of those experiments.