I grabbed a tuna wrap from the refrigerated display, shoved it into my pocket and made straight for the exit as casually as I could. On my way out, a store employee stopped me. I paused for a long second. “You should try the Mediterranean chicken,” she said. “It’s really good.”
No, the store was not offering opening-day five-finger discounts. It’s a first-of-its-kind concept from Amazon called Amazon Go, designed to do away with the most inconvenient part of the convenience store experience: waiting in the checkout line. It’s also the wedge end of a long lever that the retail and data services giant could use to upend legacy practices across the brick-and-mortar landscape.
Small Store, Big Impact
To enter the store, I first had to download the Amazon Go app on my phone, connect it to my Amazon account and choose a payment method. This generated a QR code on my phone that I scanned at a station at the front of the store. Ceiling-mounted cameras and sensors throughout the store tracked any items I took off the shelf and charged me for them automatically when I left.
The Amazon Go store has been in private beta for nearly a year, serving Amazon employees who live and work around the company’s campus in the South Lake Union neighborhood of Seattle. On Monday, Jan. 22, it opened to the public for the first time, drawing a steady crowd to its 1,800-square-foot location at the base of Amazon’s “Day 1” building.
The selection was what you’d expect from a mid- to high-end downtown convenience store: plenty of ready-to-eat fare, plus staples like milk, yogurt, coffee, snacks and cereal. There was also a beer and wine section where a store employee was checking IDs.
Unlike Amazon’s book and electronics stores, which require users to scan an item to see the price and still have checkout stands where employees collect payment, Amazon Go has prices printed on shelf tags. Although there were no cashiers in the store, there were plenty of staff members around, including a team of people preparing sandwiches and various folks to help customers.
It’s About the Intelligence, Stupid!
So what’s going on with Amazon Go? Has the global retail behemoth decided to conquer 7-11? It’s always a bad idea to undersell Amazon’s ambitions even at face value, and there is a lot going on under the hood that has much wider implications for the industry at large.
Amazon describes the technologies at work in the store as “computer vision, sensor fusion and deep learning.” That means not only that a combination of cameras and sensors tracks where every customer is in the store, but also that the system is able to associate who you are with what you are buying. It needs to recognize that you, not the person standing a foot to your left, is the one reaching for the yogurt. It needs to understand that you’ve had second thoughts about that bag of Cheetos and decided to place it back on the shelf after putting it in your bag. Then it needs to charge you, not the person who left the store 10 seconds before you, for the items that you purchased. Get any of that stuff wrong, and you might as well just hire a cashier.
That’s just to deliver on the customer experience. There are additional layers of complexity to figuring out what products people buy at what times of day, when to reorder and restock on the shelves, how to present merchandise to best capture customer attention and optimize traffic flow around the store, and perhaps even how to spot troublemakers and shoplifters.
Of course, Amazon isn’t just a retailer. It’s a giant data warehouse that’s built one of the world’s largest cloud computing platforms – Amazon Web Services (AWS) – which accounts for a larger and larger share of the company’s revenue and is considerably more profitable than the relatively low-margin retail business.
All the data and technology that makes it possible for me to grab my tuna wrap and leave the store without taking out my wallet has much broader applications across the retail industry. But for those technologies to work properly, they need to be trained. They need real-world data to build and optimize models because what they are doing is very complex.
Even AIs Need A Little Bit of Employee Training
I spoke to Dr. Daniela Braga, a leading expert in the specialized data science discipline of training AI systems and the founder of a company called DefinedCrowd (disclosure: a client of my consulting practice). Although she has not worked with Amazon on this project, she says the capabilities and experience offered in the Amazon Go environment represent the cutting edge of what’s possible with today’s machine learning and AI systems.
“You can’t do it all with computer vision,” Braga says. “Even if they are coordinating between cameras, sensors, beacons and other systems, just correlating all that data in real time is a tremendous feat.”
However, one of the defining aspects of AI and ML is the ability of systems to learn and adapt. The more the system can observe its models in action, the more it can refine those models to produce more efficient results and avoid mistakes.
Most technology companies working on AI solutions don’t have access to good, relevant data sets for something as specific as a high-traffic retail environment. Most retailers don’t have access to the world’s best data cloud and processing capabilities. But in the case of Amazon Go, the prototype store is literally sitting underneath a skyscraper filled with engineers and computer scientists, and facing out onto a street packed with typical urban consumers. That’s a powerful combination.
It Starts With Go. Where Does It End?
Like the Amazon bookstore, Amazon Go appears on the surface to be a standalone concept that uses many of Amazon’s advantages to deliver on its mission of making everything on earth faster, cheaper and more convenient. And like everything that The Everything Store does, it raises plenty of questions about the future.
Will the company expand the “no checkout required” concept to its Whole Foods stores, since Amazon acquired the chain of upscale brick-and-mortar groceries last summer?
Is Amazon simply the first customer for its own technology, which it will then offer to others in the market? The company debuted Alexa on its Echo devices but is now licensing it to others, including some competitors, because there is greater value in owning the data than in monopolizing the means of distributing it. The same is surely true in the brick-and-mortar retail space.
Is this part of a larger “city of the future” play, in which Amazon has become a relevant player because of its high-stakes competition to host its new corporate campus? Amazon executives don’t have to look further than out the windows of their South Lake Union high-rises to observe how the new urban workforce of debt-ridden millennials is packing itself into shoebox-sized “microhousing” units that don’t accommodate a lot of extra storage, much less Costco-sized pallets of products. Upscale convenience stores serve this customer base better than traditional supermarkets. On its own, this represents an opportunity worth pursuing.
Is more personalization in the cards? The next time I scan my phone to enter the Amazon Go store, will I be met with a hologram that greets me by name and asks me if I’d like to try the chicken salad this time? Will it be much longer before Amazon, or someone, goes full Minority Report?
Finally, what is the impact on the already-endangered low-skill workforce if (when) this kind of technology does away with millions of service jobs? That’s not really a question for Amazon alone, but it plays a larger role in that conversation than most other companies.
Amazon is not commenting on the record, so we'll have to draw our own conclusions on those questions while we wait to see what they do. But the bigger question for all of us is: How much are we really willing to pay for the convenience of walking out of the store with a tuna wrap in our pockets?
Rob Salkowitz, Contributor