#Shoptalk18: 9 lessons from Amazon Go

Amazon Go opened its doors to the public on January 22 2018, with the launch of a 1,800 sq ft checkout-less store in Seattle. After a year spent testing the ‘Just Walk Out’ technology on employees at its HQ down the road, the eCommerce juggernaut was confident its computer vision software was ready for real-life convenience shoppers.

The grocery store is unique for not having a checkout. Shoppers scan their mobile device with the downloaded Amazon Go app as they walk into the store and a combination of shelf sensors and cameras keep track of any items they pick up. Products are automatically detected and placed in a customer’s virtual basket. They simply walk out of the store and payment is made through the Amazon app, without the shopper having to even swipe their smartphone screen.

At Shoptalk 2018 in Las Vegas Gianna Puerini, VP Amazon Go and Dilip Kumar, VP technology, Amazon Go and Amazon Books, discussed the challenges and lessons learnt from launching Amazon Go. While the pair remained tight lipped on further roll outs (although we learn that the technology will not be extended to Whole Food stores), we did discover more about the technology which lies behind the checkout-less store, as well as how customers influence the store format going forward.

Here we share nine insights from Amazon Go:

1. Computer vision, not RFID

Amazon decided to use computer vision technology to monitor the movement of products around its Amazon Go store, rather than RFID tags.

Kumar said the main reason for this was because of Amazon’s investment in machine learning – “we kept coming back to what sets of technologies could leverage machine learning the most” – likening the technology to self-driving cars in a grocery store setting.

He said the operational impact of using RFID was too high, because in a fast-moving convenience store every product would have to be manually tagged before hitting the shelves. “We wanted it to be a little more seamless,” he said.

2. Teaching computers to recognise products isn’t easy

“How do you do this in a way which means it is effortless and the technology blends into the background – it is a lot harder than it seems,” said Kumar. He said with many people shopping in close proximity it was difficult to design the computer vision and machine learning algorithms, which would allow the technology to identify the products moving around the store.

“The holy grail of computer vision is to be able to take what is seen and to be able to interpret it,” explained Kumar. He gave the example of two cans of Red Bull, one with and one without sugar, or a jar of strawberry jam sat next to a jar or raspberry jam, which both look very similar and may only have one distinguishing character, which humans would look out for. But teaching this to computers is much more difficult than teaching a shopper.

“Computers take a lot of information to recognise items, they don’t have inbuilt knowledge humans have.”

3. Thousands of Amazon employees work on machine learning

When asked how many people work on machine learning at Amazon, Kumar said the technology is being used all across the business, not just in Amazon Go.

“There are thousands of people on machine learning, it is one of those things, we don’t have a specialised team,” he explained, say it is used in voice, fulfilment centres and Amazon Go to name just a few. “Pretty much everything you see or do at Amazon.”

4. Think customer first when rolling out technology

Amazon Go was created to add value to the existing Amazon customer. Amazon Go’s VP, Puerini, said before retailers think about deploying similar checkout-less technology they should first establish a sustainable business model.

“It comes down to who is your customer? What can you do that adds value to their life and what are you uniquely positioned to offer to them?” she said. “And if not, are you willing to build or buy it?”

“It comes down to who is your customer? What can you do that adds value to their life and what are you uniquely positioned to offer to them? And if not, are you willing to build or buy it?”

Amazon offered customers convenience online, but could not provide an in-store experience, so it decided to build stores including Amazon Go and its bookstores. The retailer also went on to buy Whole Foods in the middle of last year to add value to its customer.

She said the Amazon approach is to start with the customer and work backwards. The eCommerce giant even wrote the press release for Amazon Go before it knew how it was going to build the proposition – to ensure it was thinking about the customer first, not the technology.

5. Customers aren’t comfortable just walking out of a store

Puerini said the store was designed to make the shopping experience as natural as possible, rather than teaching customers how to shop differently.

“But everything in the store is new,” she said. “And what we didn’t necessarily expect was how many people would stop at the end of their first trip or two and ask ‘is it really OK if I just leave?’ It tends to wear off after the first or second trip.”

6. Store associates are repurposed

While the Amazon Go store lacks a point of sale (POS), the store associate who would have theoretically been manning the till point is now restocking shelves or helping customers on the shop floor.

While Amazon Go is a convenience store, Puerini said the inventory turnover was one thing that surprised her. Due to the increased speed of shopping, more customers and get in and our faster, which means shelves are constantly being restocked – especially during peak periods including breakfast, lunch and after work.

“We didn’t know how quickly people would move through, how frequently they would shop – it’s very much day one, and who knows what will happen as we keep learning,” admitted Puerini. “Keeping the shelves stocked is the most important thing in any store, whether it has technology or not.”

“We didn’t know how quickly people would move through, how frequently they would shop – it’s very much day one, and who knows what will happen as we keep learning,”Gianna Puerini, VP Amazon Go

While a chicken sandwich has been the best seller since doors opened, meal kits prove popular at 5.30pm when customers are on their way home from work, while fresh fruit and products from local Seattle bakeries are favourites during the morning peak.

7. Whole Foods won’t be getting ‘Just Walk Out’ technology

Amazon wouldn’t reveal its roll-out plans for Amazon Go, but interestingly it was quick to say it would not be rolling out the technology to any of Whole Food’s 450+ stores.

“Whole Foods is fantastic at what they do and we want them to keep doing what they do,” said Puerini.

8. Offer and price are key metrics of success…

Like many traditional retailers sales and frequency of visits was a key metric of success for Amazon Go, but additionally it relies on making sure it has the right products at the right price.

“We knew from the get go, convenience doesn’t matter if the customer doesn’t like the food assortment or the pricing.”

9. …so Amazon relies heavily on customer feedback

While Amazon has been spoiled up to now with the freedom of unlimited shelf space online, the retailer needs to make sure it has the right products on its shelves… at the right price.

The e-tailer listens to customer feedback given on the app and makes changes accordingly.

Kumar pointed to its salad offering which includes sunflower seeds and cheese, with a dressing available on the side. After receiving feedback from vegan customers, Amazon quickly decided to put the cheese as a side option as well.

“Customers are very passionate about food,” he said.