A number of media articles this week cited that Amazon was removing its Just Walk Out solution from Amazon Fresh grocery stores. It turned out to be a bad game of telephone tag stemming from an interview with Tony Hoggett, head of Amazon’s physical grocery store endeavors, conducted by The Information. Even The Information’s paywall didn’t stop industry insiders from riffing on the supposed demise of JWO and its expensive cost.
I talked to Jon “JJ” Jenkins, a senior executive who worked for years with “Jeff B.” and is on the team at Amazon Web Services that runs Just Walk Out (aka “JWO,” which they also affectionately call J-Wo, like J-Lo). He clarified several points:
- Amazon is still investing in JWO and, in fact, is “pushing hard,” particularly in the smaller formats where the solution has proven viable. Jenkins told me that there will be more stores launching JWO in 2024 than ever before and that four opened just last week (one at a stadium in Ireland, as an example, and another at Wrigley Field in Chicago). Today, more than 130 stores globally have JWO technology. Why the smaller formats? Amazon evidently looks at “net margin per square foot,” which is essentially the profitability of a store. JWO makes sense if a store is particularly profitable, which airport convenience stores and sports stadiums are (think about the markup you pay for water bottles or snacks at those locations). Chris Walton of Omni Talk has a great interview with another AWS executive on the four characteristics that work for JWO.
- The claims that JWO requires a lot of human intervention (including estimates of 1,000 employees in India manually labeling transactions) had “no factual basis.” Jenkins said that small-format stores, especially where planograms were largely identical, required little to no human intervention. He did mention that at launch for completely new store formats, there is sometimes some manual work to improve the machine learning algorithms but nothing significant over time (e.g., properly identifying two similar-looking cans of soft drinks located together on the same shelf). In the words of Amazon, “The underlying machine learning model is continuously improved by generating synthetic data and annotating actual video data.” I’d argue that the recent stories about human intervention in driverless cars helped this part of the narrative pick up traction.
- The true story that Hoggett was trying to communicate to The Information was that the automated Dash Carts have a lot of promise. Jenkins said that 16% of Fresh store sales come from Dash Carts (which have checkout integrated) and that customer satisfaction is high among Dash Cart users: Those shoppers spend much more, and shoppers evidently drive long distances just to shop with a Dash Cart.
For what it’s worth, Jenkins said that the AWS vision for retail is to provide a “next-generation experience in stores,” which is what the palm-based payment is intended to achieve, as well. While people have speculated about autonomous and cashierless checkout (i.e., the category where JWO lies) since its inception, this recent news confirms the concern people have had all along with making the technology cost-effective in large, low-margin environments like grocery stores. It seems to have better success in high-margin, small formats. Time will tell if this type of solution in this type of store format is a flash in the pan or built to last.
