Amazon Fresh’s AI retreat – practical AI applications will win out in the grocery sector

Over the past 12-18 months, no retailer has been able to send a senior executive out to talk without prepping them with a party line on AI – and generative AI in particular. As with every other sector, the ‘silver bullet’ potential of the tech has been played up in varying degrees as the gen AI hype cycle continued to expand. 

The grocery sector is no exception. At last month’s Morgan Stanley Technology, Media and Telecom Conference, Walmart’s Global CTO and Chief Digital Officer Suresk Kumar enthused: 

I am particularly excited about how we are using the latest gen AI to help drive search, to help drive a more conversational experience with our customers. So part of adaptive retail means that we need to understand what the customer wants first and then to be able to adapt to it. And this is where things like gen AI really come into picture, because that gives you a natural way to be able to interact and to be able to understand.

Kumar added: 

Gen AI is already in use. We announced our gen AI-based search. We use Large Language Models that are trained from on world knowledge. We have built that, we have taken that and we have trained that in-house in our own platform and it’s now powering increasingly significant fraction of our searches that happen on our app. So it’s already real. 

Amazon

That said, recent events might be read as indicating that a lot more realism is creeping into the assumption of the potential of AI, at least for the moment. As far back as 2016, Amazon was talking about a revolution incoming: 

What if we could weave the most advanced Machine Learning, computer vision, and AI into the very fabric of a store so you never had to wait in line?

This accompanied the launch of the first Amazon Go, “a new kind of store featuring the world’s most advanced shopping technology. No lines, no checkout – just grab and go!”.

This evolved into the wider spread use of Amazon’s Just Walk Out technology across 47 of its 64 Amazon Fresh stores. That experiment came to an abrupt end earlier this month as The Information broke the news that the tech is being dropped in the US in favor of self-checkout shopping carts. (It will remain in use in some US Amazon Go convenience stores as well as smaller Fresh locations in the UK).

The news followed an earlier report that the checkout free strategy was dependent on a network of cameras operated by 1000 outsourced workers in India. Amazon admits that there are Indian workers who annotate video images to help evolve the machine learning models. 

Others have been here before. Back in 2019, UK grocery chain Sainsbury’s abandoned a checkout free trial in one of its London stores, although this had been positioned as an experiment at the time of launch rather than “a new format for us” as its Chief Digital Officer put it. But the conclusion from that experiment was stark: 

It’s clear that not all our customers are ready for a totally till-free store.

But…

Of course, that was before the generative AI hype bomb exploded. Tesco, the UK’s largest grocery store operator, opened its first GetGo checkout free store back in 2021. Flash forward to 2024 and CEO Ken Murphy makes it clear that AI tech will play a major role in the firm’s strategy moving on: 

We’re continually looking for new and improved ways to serve customers better. Recently introducing a new AI-based tool to improve our retail proposition. This will enable more bespoke product ranging by store location and demographic. Meaning, we increasingly stock the products that customers want to buy in the stores, they want to buy them.

We’re also now using data-driven software to drive more efficient transport scheduling and stock assembly processes in our supply chain. We expect AI-driven solutions such as these to become a growing part of how we do business going forward, supporting our colleagues to focus on the most value added parts of their roles.

The emphasis here falls on operational efficiencies and back end optimization, perhaps a more pragmatic proof point of AI’s role in the grocery sector. Back to Walmart’s Kumar: 

It’s not just about gen AI by itself. The real unlock comes when you use these breakthroughs, like gen AI and Large Language Models in conjunction with a whole bunch of other technologies. I’ll give you one example in terms of how we do demand forecasting. We have been building out our own ML models, deep learning models to really help understand what customer demand is going to be like, near term, long term, by area, actually even at the household level. So, [these are] traditional models. Infuse that with the latest breakthroughs from gen AI where we can really understand not just at a product level, but really at a basket level, at a household level. Combine these pieces together then that’s where a lot of the magic starts happening, right.

Others are taking a similar approach with their own AI initiatives – pragmatic, practical, not flashy ‘up front’ stuff. UK firm Morrisons is installing thousands of AI-enabled cameras from US AI company Focal Systems across its store network to assist store workers to fill shelves quicker. The cameras scan the aisles and flag up inventory using a series of categories – out of stock; planogram non-compliance (ie something’s in the wrong place); low stock; and re-stocked. 

Others are turning to AI to tackle the very real issue of theft at self checkouts. The UK’s Co-op reckons to have lost £70 million to retail theft last year alone and is now rolling out AI-enabled CCTV to track what shoppers are putting in their bags at the checkout. In the US, Target has faced a huge uptick in thefts, so it’s hardly surprising then that this week the retailer is reported to be rolling out AI camera tech at its own self checkout stations. 

My take

Those kind of operational uses of AI tech make pragmatic sense in the retail sector and despite the somewhat gleeful tone of media coverage of the Amazon Fresh decision, that’s where the benefits are going to be realised in the near term, regardless of the distractions generated by gen AI hype. One last comment from Walmart’s Kumar to close: 

Where I see this going is that gen AI is going to become just one more tool, one more technology. It will become more and more available underneath the covers everywhere. But it won’t stand out all by itself. It will get seamlessly integrated with everything else that we are doing, whether it’s around customer or whether internal, just like any other piece of technology really has almost become invisible, everybody uses Excel now. So like that, I think gen AI is just going to become just one more part of the toolbox, and people are just going to adapt to it.