Amazon pulled Just Walk Out technology from its grocery stores earlier this week. Amazon says 130 convenience stores, and some Amazon Go locations, will keep using the technology, but a source close to the team told Gizmodo that Amazon has killed off much of the internal development. “Nearly all” of the engineers dedicated to building Just Walk Out were laid off on Wednesday, the senior team member who was let go said, leaving what they described as a skeleton crew to sustain the technology’s future. Amazon disputed that characterization as a “fabrication.”
A handful of engineers and managers are left to maintain Just Walk Out, the former employee claims. Amazon’s checkout-less shopping tech, first introduced in 2016, uses cameras and sensors to track which items customers are buying. The e-commerce giant laid off several hundred employees from its AWS unit and Physical Stores Team on Wednesday, first reported by Geekwire. Our source says dozens of Just Walk Out team members were impacted. The layoffs could spell an uncertain future for hundreds of Amazon Go convenience stores, sports stadiums, and third-party retailers that rely on the technology.
“Gizmodo’s story is wrong,” said an Amazon spokesperson. “We continue to invest heavily in our Just Walk Out team and technology, and have hundreds of employees working to keep up with the rapid demand.” Amazon disputes that this engineering team is now a skeleton crew, but did not clarify how many Just Walk Out engineers were laid off.
Amazon had been planning internally to pull Just Walk Out from Fresh grocery stores for roughly a year, according to the former Amazon employee. They say the technology was too expensive and too complicated to run for large retail locations but seemed good enough for small convenience stores. These small shops are less complicated as they involve fewer products and fewer customers, so they’re easier to automate.
However, the Just Walk Out team was blindsided when their unit was hit with layoffs, according to our source. Amazon says it will open more Just Walk Out stores in 2024 than any year prior, but the laid-off team member says the team that was responsible for developing the technology is largely gone.
Amazon employees who were laid off Wednesday took to LinkedIn after the news. Gizmodo counted at least 15 Just Walk Out employees who are now “open to work.” One of the posts claims that half of their Just Walk Out team was laid off. A few of the team’s managers appeared to still be on staff, based on their LinkedIn profiles.
Just Walk Out repeatedly failed at several internal metrics, the former senior team member claims. Most glaring, it never became truly autonomous. The checkout technology relied heavily on a team of human reviewers to label and verify purchases that its computers can’t, our source says. An India-based team labeled videos for AI training, a common practice in machine learning, but also verified videos of purchases flagged as “low-confidence events.” Amazon confirmed this but disputed what percentage of videos needed to be verified.
At the launch of Just Walk Out, almost all the purchases required human review, according to our source. By mid-2022, The Information reported, 70% of purchases required review. Today, somewhere between 20% and 50% of purchases require human review, the former senior team member told Gizmodo; according to The Information’s reporting, Amazon’s internal goals wanted that number to be close to 5%.
But there were other problems. Just Walk Out developed sensors that were more reliable than cameras for tracking purchases, however, they were way too expensive, according to our source. The former employee said Amazon aimed to bring the price of one sensor down to $100, but the engineering team could only bring it down to $350.
“We’ve identified a few targeted areas of the organization we need to streamline in order to continue focusing our efforts on the key strategic areas that we believe will deliver maximum impact,” an Amazon spokesperson said in an email to Gizmodo this week.
A Reckoning for Automated Retail
Amazon pioneered the automated retail space when it launched Just Walk Out in 2016. It was the biggest player then, and it still is today. Back then, this technology in grocery stores seemed too good to be true, and that still appears to be the case. Amazon now appears to be scaling back its aspirations.
“Fully unstaffed grocery is probably not the right path,” said Steve Liguori, who co-founded another automated retail company, Juxta. “At least, not with the technology that’s out there today. That’s not the application for this tech.”
Juxta focuses on small convenience stores, like most automated retailers in this tiny, but growing industry. The company claims to have a high degree of accuracy, but that’s partially because they ask customers to review their own receipts on a screen before they leave. It’s a more practical approach to automated retail, where the AI still has a low degree of confidence in many purchases.
“When you’re shopping in a small convenience store, the maximum amount of products you’re picking up is around five or six,” said Ankur Sharma, CEO of another automated retail company, Brysk. “In a supermarket, you could pick up over 100 items,” continued Sharma, noting that this amount of data can overwhelm early machine-learning tools.
Amazon’s venture into large supermarkets with an automated retail system was always seen as bold, according to Sharma and Liguori. Most competitors play in small convenience stores, but even there, the tech does not always have a high degree of confidence. However, Amazon is big enough that it can afford to take those kinds of risks. Internally, Just Walk Out was seen as a startup that could reap huge rewards if successful, according to our source.
Ultimately, Amazon’s automated retail systems never fully removed humans from the loop, albeit, many more popular AI systems today don’t either. While the company doesn’t see this as a failure, it would be hard to see the alleged dissolution of its internal engineering team as anything else.
Trending Products
