I finally visited the Amazon Go store in Seattle last week.  I thought the ability to walk out of the store without waiting in line was a great customer experience.  The visit also re-enforced my belief that computer vision and deep learning will revolutionize how retailers operate.

The Amazon Go store, located in downtown Seattle near Amazon’s headquarters, opened to the public earlier this year after several years of employee testing. The store features what Amazon calls “Just Walk Out Shopping,” which automates the checkout process so that you just walk out and Amazon bills you for what you take.  Amazon is reportedly planning on rolling the store format out to several additional markets, including San Francisco and Chicago.

Customer experience

The Amazon Go store is not merely a good experience.  It is a GREAT experience.  While the pace of adoption will depend on store format and size, it’s hard to imagine automated checkout won’t become the industry standard over the next 5-10 years.

The customer experience is both easy and fun:

• You scan a QR code via the Amazon Go app to enter the store (the app is linked to your Amazon Prime account).
• You grab what you want (I put some items into a reusable shopping bag and some into my pockets).
• When you are done, you just walk out of the store (see figure 2). It feels a bit odd to check out this way at first – it almost feels like stealing.
• Amazon then charges you for what you bought.

My receipt was spot on. For example, Amazon charged me for the bagel I stuffed into my pocket but didn’t charge me for the Cheerios I picked up and eventually put back before leaving.

No one shops for the checkout experience.  Checking out is a necessary evil required to tally your purchases and pay. By automating the checkout process entirely, Amazon is able to speed up the shopping trip and reduce operating costs. While retailers have experimented with other options such as scan and bag, mobile POS, self-checkout, nothing I have seen beats the speed and convenience of Amazon Go.

Merchandising

Amazon Go is a small format convenience store (1800 square feet) with a focus on fresh prepared foods (see Figure 2). In addition to prepared food and meal kits from Amazon and local vendors, the store also carries a mix of packaged goods, refrigerated items (e.g., yogurt, milk, cheese), frozen food, and Amazon logo items.  There was a small amount of Whole Foods 365 private label products as well.

The store also sells wine, and there was an employee stationed at the wine section to check IDs.  It will be interesting to see how this process gets automated over time – perhaps using facial recognition to confirm identify and age.

The store does not carry hot food (coffee, soup, etc.) or random weight fruits or vegetables, most likely because of the difficulty of accurately tracking and charging for these items using computer vision.

The selection is highly curated. I imagine over time Amazon will adjust the assortment mix based on sales, customer shopping behavior, and local demographics.

Technology

There is A LOT of technology in the store, particularly in the ceiling (see figure 3).  As the image shows, there are cameras every foot or so.

Computer vision uses AI and deep learning to analyze digital images and videos.  According to their website, Amazon Go uses this technology coupled with sensors to track each person in the store, what they take off the shelf, put back, and carry out of the store.

Given the amount of hardware in the store, it’s hard to imagine Amazon Go is scalable or affordable for most retailers as it is currently constructed, or the right choice for consumers who are looking for fresh fruit and veggies.  However, it is likely that Amazon and startups targeting the space will rapidly innovate to bring down costs and increase effectiveness.

But the technology works well today.  Amazon charged me only for the things I bought.  It did not charge me for items I picked up and put back – or for free items (including a reusable shopping bag and cream cheese for my bagel).

Interestingly, it took about 10 minutes from the time I left the store to receive my receipt (I did two trips into the store and this happened both times).  I am not sure why.  It could be as simple as a slow email server, but my guess is that either a) the technology takes several minutes to process the video from your trip to figure out what you actually bought and/or b) there are humans reviewing the results from the video analysis to ensure accuracy.

In summary, Amazon Go is a great customer experience.  And while the current technology set up might not be scalable, automated checkout is likely to become the industry standard in the coming years at a store near you.