You thought Amazon’s Just Walk Out experiment was a failure? Just think of their data goldmine…

Although Amazon’s in-store test failed to scale, it has created a uniquely rich data set to teach new AI models and to test the next iteration of in-store technology at vastly lower costs than other physical retailers. 

I have just spent a few days with a group of amazing product folk from across the online payments market. We spent the time talking about everything from Wero and Pix to “painted doors” and A/B testing. The focus of the conversation was making evidence-based improvements to consumers’ payments UX using their revealed preferences in online payments, not just survey feedback. This made me reflect on the recent news from Amazon that they are scaling back their in-store no-checkout ambitions. Their experience demonstrates the difficulties and costs of testing payments technology, at any scale, in-store rather than online. It also made me think about their unrivalled consumer experience data set and the opportunity to create lower cost digital test environments for the next generation of in-store technology. This should cause other physical retailers significant pause for thought as they experiment with their own no-checkout solutions. 

Amazon entered the UK grocery market back in 2021 with its Amazon Fresh stores. They differentiated themselves with innovative checkout technology that delivers a ‘just walk out’ experience with frictionless checkout. The format was going to be better than the more traditional self-scan and self-check-out by completely removing the checkout pinch point that causes us all so much frustration. Amazon promoted their use of tracking sensors and facial-recognition technology that enable customers to walk out with their products and receive a bill the next day. These sensors gather billions of data points around how consumers shop, and the odd things we all do when buying in-store.  

Unfortunately, this vision for next-gen grocery shopping hasn’t been able to scale or deal with complex set of edge cases which are generated while shopping. As the technology struggled, Amazon had to employ a large back-office staff to fill the gaps. Even with this additional human support, customers still had a significiant number of errors such as being mischarged, charged for items they didn’t pay for and items which didn’t scan through at all. Alongside the technology challenges, consumers don’t seem quite ready to trust AI to get it right, with reports of modest sales, and less social/traditional media attention than required to ‘break through’ to customer’s general habits.  

Utimately, despite the millions spent on creating new store formats, and new consumer shopping experiences, the test has been a failure. Amazon has recognised this and discontinued the Just Walk Out feature in all the US stores.  Amazon has already responded to direct consumer feedback on what they liked and disliked. For example getting alerts from the app when items are picked up or placed back on shelves or immediately getting a checkout receipt upon walking out. Amazon will, I am sure, also be embedding their learning into their more traditional Dash Cart technology.  

Unlike the live testing environment we have got use to in the eCommerce world, in-store payments technology is MUCH more expensive to iterate and test. We have to applaud the confidence shown by Amazon as it launched and tested this technology, and for recognising its constraints when it would not scale. Most importantly the data Amazon has collected will be vital in creating a digital test environment for new iterations of in-store technology which may well become the envy of the rest of the retail world. I am sure this will be used to train the next generating of GenAI models on how to respond to a myriad of edge cases. It can also be used to digitally test new iterations of the solution at vastly lower costs than those of its competitors. We may have to wait a few years before another version comes to market, but I look forward to mystery shopping the experience when it is ready to deploy…! 

To find out more, get in touch