Amazon Go

Written by: Srinath Sitaraman

December 12, 2016 - How does Amazon Go work? And how does Amazon make their check-out process completely invisible and frictionless?

According to some reports [ref1], about 70% of online shoppers abandoned their shopping cart this past Black Friday / Cyber Monday shopping frenzy. On an average Friday, the average cart abandonment rate is not much different either, but the pinch felt by merchants is much more during a Black Friday weekend. This article [ref2] further estimates that about $4.5 Trillion worth of merchandise is left behind in abandoned carts every year for the last few years. That’s a lot of money left on the table and consequently online retailers, brands, acquirers, processors are investing substantially to remove the friction at the final step in a transaction, and part of that is simplifying the payment process. Needless to say, Amazon.com has been at the forefront of these efforts to simplify the purchase and checkout process on their websites. This blog is however not about eCommerce or mCommerce. Rather it is about Amazon’s recent announcement about Amazon Go.

The big news this week is that Amazon is taking a plunge into retail shopping with their ‘Amazon Go’ stores, to specifically cater to food and grocery needs of the fast moving urban population (their trial store is at the heart of Seatlle). The part that is interesting to me though is how Amazon makes the check out (payment) process completely invisible and frictionless – again catering to the needs of their target demographics. Here’s a nice little video from Amazon, on how this service works:

Quick disclaimer that all or most of what follows in this blog is pure conjecture and imagination. So, I wanted to present some observations and make some educated guesses on how this works. The actual implementation might be completely different.

How Amazon Go works?

According to what has been published, a shopper can simply scan in into the store, buy stuff, and walk out. Yup, you are not pulling out the card, certainly not waiting for a 8-second EMV transaction to complete, and surely not thinking about how expensive that salad was. None of those hassles with Amazon Go - Amazon will automatically bill your card on file based on the items that it thinks you bought at the store. I believe the key to the success of this program will depend on how confident Amazon can tell whether you bought something, and secondly how the dispute management process works. Amazon.com was lauded for a seamless, customer friendly returns and dispute management process and you can expect that the customer friendly attitude will also translate to their retail efforts.

Invisible payments

Amazon identifies 3 key components that enables an invisible payment process at their stores:

  1. Computer Vision – for those not aware, Computer Vision is a field of study within image processing specifically intended to locate, track, identify and recognize objects in an image, and to enable actions/decisions based on that information. Plate pass for toll collection is a classic example where computer vision is used to read the number plate on a moving car and bill the customer for the toll amount, and is a much cheaper alternative than using UHF transponders (or other sensors) to accomplish the same goal. In Amazon Go, computer vision is presumably used to know (with a camera mounted on each shelf) when and how many articles was removed from the shelf (or if it was placed back on the shelf).
  2. Machine Learning and Artificial Intelligence – Amazon can (again presumably) build intelligence to know whether you’re the type of person to pick up certain food/grocery items as opposed to others. This can be based on past purchases (either at the store or online) or a shopping list.
  3. Location within the store – Presumably, the stores have beacons to build a data point on whether you are at a certain location within the store.
  4. Simply put, with Amazon Go, Amazon can score on whether you are the type of person that eats yoghurt, whether you are physically present in the yoghurt aisle, and whether couple of yoghurts was picked up from the shelf.

Fraud, Risk mitigation and Dispute Resolution

Amazon Go stores sell grocery and food items. For fraudsters, hacking of payment systems is their bread and butter, but in this case the rewards of such penetration might not be their cup of tea – pun fully intentional :). In other words, the risk of such a setup is lesser for a grocery/food store compared to stores that sell laptops.

This is the type of store (MCC) where programs like VEPS/QPS are already made available by the brands i.e. no card holder verification is necessary even at POS. The enrollment process is quite straight forward – in fact according to Amazon, all you need is an existing Amazon.com account and a compatible smart phone with the Amazon app to be able to make purchases at their store. Given that Amazon is not putting its users through an enrollment process at the store (e.g. register a photograph on their database), it is logical to think that Amazon will rely on on-device card holder verification (PIN or biometric) to use the app at the store. Perhaps in the long run, it is also conceivable that Amazon can build facial or gait verification to recognize the card holder.

How Amazon Go staff and systems handle people at the point of entry might be important – especially as they try to prevent people from simply walking into the store without scanning in. There would be folks that may try to benefit from the fact that there is no check out process. This can be mitigated through procedural controls – i.e. a consumer cannot walk in to the store, unless they have scanned in at the entry. The trick however is in balancing convenience for the average customer with establishing a mechanism where looters don’t take advantage of.

The system relies on the use of barcode in the Amazon app. Fraudsters can potentially try to replay the barcode (replay attack) from a past transaction or perhaps try to conduct a relay attack. To mitigate this risk (and presumably Amazon is already using these features) the barcode can be setup for single use and/or bound to the Device (phone) ID. So, simply relaying the barcode at the entry without the phone being present in the store may not work. That is the physical presence of an Amazon registered phone might be the key to prevent such hacks.

Disputes could potentially arise from a consumer who claims that Amazon billed for an item that they did not buy. Building up the Artificial intelligence to know who bought what will take time, so such type of dispute management activity can be expected in the initial phases. For dispute management /resolution purposes, it is conceivable that a photograph of the card holder picking up the items from the aisle and/or walking out of the store with the goods could be used in this situation.

Some questions to ponder?

  • Can other merchants or processors implement this model?
    Sure, why not. This will require substantial investment to make it happen, so the business case needs to be refined. There are certain merchants (Walmart, Best Buy etc...) however where their risk acceptance profiles might not allow them to allow card holders to walk away with goods. These merchants also benefit from spot sales – e.g. buy a chewing gum or magazine while idling at the cash register.
  • Is a hybrid model (some customers walk out and some go to a cashier) possible?
    Again, sure. It is conceivable that merchants allow their preferred and trusted consumers to walk through, whereas other clients need to see a cashier.
  • Interchange models?
    I suspect Amazon is paying a CNP interchange rate, although the card holder/device is physically present in the store. It will be interesting to see how Amazon has sorted out the interchange models for their store.
  • If this has legs to crawl beyond Amazon, who might it displace?
    If this type of system starts to take roots, it will have all sorts of ramifications to displace a number of payment ecosystem members. There are two large ecosystem members that I feel will be following this announcement from Amazon with bated breath.
    • Payment networks: Given their size, Amazon could conceivably become third major payment network in the US and effectively go straight from merchants/acquirers. This is assuming that merchants buy the Go system from Amazon including its processing capabilities.
    • Terminal vendors: This announcement will obviously worry the terminal vendors, but astute terminal vendors will realize that they will have a role to play (in fact much larger than the role that they play at the point of sale) in developing the infrastructure to make this happen.

Disclaimer

These are the personal opinions of UL’s employees and its guests and should not be misunderstood as representing the opinion of UL's clients, suppliers or other relations.