It’s been more than a year since Amazon Go launched to the public in Seattle. Using AI technology with cameras and sensors laid out across the store and shelves, Amazon Go allows customers to walk in, pick up items, and walk out of the store without the need to queue or pay at the till.
Since that launch, there has been a burst of activity in the space from internet companies, traditional retailers, and startups putting their stakes in the ground:
- Several stores like the Amazon Go are popping up in China. Brands include both large internet players like Alibaba (Tao Café) and JD.com (JD Daojia) and startups like BingoBox, F5 Future Store, and Xingbianli.
- Traditional retailers like Walmart have responded with their own cashierless store in Sam’s Club Now, which doesn’t specifically use AI or camera-based technology, but relies on mobile-based product scans.
- A flurry of startups, including Sensei, Trigo Vision, and Standard Cognition, are supplying the technology platform for brick-and-mortar retailers to compete with Amazon. Other startups like Inokyo plan to open their own stores.
In the meantime, Amazon is keeping its cards close to its chest while it tests the technology in live stores and plans to expand them going forward both in terms of number of stores and the square footage. At the end of 2018, Amazon Go had 9 stores across the U.S. covering some of the major cities of Seattle, Chicago, and San Francisco, with plans to expand to more than 50 stores by the end of 2019. As per Bloomberg, Amazon has plans to open 3,000 stores by the end of 2022.
Questions and Business Models
Plenty of questions remain around the effectiveness of the technology and whether it can scale upwards from small stores (<1,000 SF) to large (>100,000 SF). We have yet to see any data around how many errors occur in Amazon Go stores and how many times human shop assistants are needed to resolve issues. There are also customer privacy issues around what data is being collected or used to make the cashierless experience more autonomous – or at some point even personalized.
Another issue revolves around the costs involved in deploying the technology, which includes expensive cameras and shelf-weight sensors. Estimates suggest that each Amazon Go store costs $1 million, although it’s not clear how much the AI technology and cameras contribute to this cost. Once trained on an inventory of products, the AI software can be reused in other stores, assuming the inventory largely stays the same. Amazon Go stores carry 1,000 items on average, all of which can be fed into an AI and deep learning recognition system. Most of these are carefully selected products where there is little ambiguity or chances of the system making an error. However, an average Walmart store can carry more than 300,000 products and multiple variations of the same product, which can cause issues for even the most advanced deep learning systems.
Amazon could also decide to sell the technology to other retailers, although that would be hard to do since retailers largely see Amazon as competition. Most likely, Amazon will stick to the current business model of operating its own stores, using Amazon Go stores as a testbed, and eventually rolling the technology out across its Whole Foods subsidiaries. Amazon Go stores could also become delivery pickup locations and hubs for delivery robots, drones, and autonomous delivery trucks to do their local delivery rounds. This model would possibly improve delivery times, especially for grocery items, since it would lessen reliance on warehouse capacity that is located (for the most part) outside city limits.
However, the growth trajectory of Amazon Go might face regulatory hurdles. Cities and states in the U.S. have started to ban cashierless stores to prevent low income communities that don’t necessarily have credit cards, bank accounts, or smartphones from being left out. New Jersey and Philadelphia have already banned cashierless technology, and San Francisco is known to be considering doing the same.
The Future of Retail
Despite the regulatory and technology hurdles, Amazon Go looks and feels like a technology from the future. We are moving toward a time where autonomous retail will become common in most metropolitan city neighborhoods. In 5-10 years, there will be different versions of autonomous stores, and the best way to think about it might be to use SAE autonomy levels for autonomous cars:
- Level 5: At the far end of the scale are stores like Amazon Go, which are fully autonomous stores (Level 5) where there is almost no need for a human.
- Level 2: At the lower end of the scale are stores like Sam’s Club Now and other cashierless handheld terminal technologies that have humans in the loop (Level 2-3) to assist with issues. These don’t necessarily use cameras but still do away with tills.
- Level 1: Stores that have physical tills for checkout but use some level of technology like self-checkout tills.
In other words, embedding intelligence into stores is where the future of retail lies. Whether that intelligence is used in collaboration with a human store assistant or without is something that will play out over time. However, we are surely advancing from physical stores to intelligent stores. This trend mostly applies to the grocery sector, but some elements could also be borrowed in clothing stores, for example.
It’s highly unlikely that the brick-and-mortar retail experience will stay static for the next 10 years. Amazon Go has shown us the possibility of building the store of the future: Level 5 autonomy. But the challenge lies in converting existing stores into semi-autonomous stores. Stakeholders need to find practical solutions to make that transition and reduce the cost of conversion.
Many stores today utilize footfall sensors, shelf sensors, RFID sensors, CCTV cameras, and other types of sensors – all which can be utilized to feed into an AI system. Amazon Go focuses on a full cashierless experience, but stores can also use AI technology and sensors to improve backend logistics and operations or simply improve security and reduce shoplifting. If you are a retailer, the best way to approach autonomous store technology is to think in terms of autonomy levels. Chart a path toward the first levels of autonomy and consider how you can reuse the same technology to think beyond the customer experience.