Amazon ditches 'just walk out' checkouts at its grocery stores

According to The Information, 700 out of 1,000 Just Walk Out sales required human reviewers as of 2022. This widely missed Amazon’s internal goals of reaching less than 50 reviews per 1,000 sales. Amazon called this characterization inaccurate, and disputes how many purchases require reviews.

I can’t imagine there’s that much untapped profit in the grocery business that Amazon could turn a profit with such an expensive and unreliable mess like Just Walk Out and I find it’s so typical Amazon to find out that the man behind the curtain is actually a bunch of offshore workers in India.

Wanted to comment on this part -- Prime itself was nothing more than faster shipping, though yes the difference between 4-6 weeks standard vs 2-days was massive (not just slightly faster). But the point is, Amazon excels at identifying friction points that others have just accepted as industry norms, but which, if unblocked, could actually meaningfully shift consumer behaviors.

Go might be a failed experiment, but "slightly faster shopping" is probably an unfair trivialization of what the experiment hypothesis was really about. A core thesis of Amazon in general is basically, to fanatically remove any unnecessary extra steps/friction/bureaucracy/etc., between a consumer and their act of purchasing.

For another example, think back to why Amazon cared so much about the 1-click patent -- legal validity aside, the idea that you can have 1-click checkout, was pretty revolutionarily customer-obsessed, compared to the average online shopping experience of the early 2000s.

And in fact, the natural progression from 1-click, is of course going down to "0 clicks", which is what things like, subscribe-and-save, memberships/subscriptions, or Alexa/Echo styling clothes for you, are meant to do -- they are meant to shift the consumer's mental model of shopping away from emphasizing the build-up from browsing into the climax moment of then clicking to purchase, to instead make the actual purchase decision more of a hidden-in-the-background/automatic thing, instead of a foreground conscious choice.

Theft is an obvious one. It happens in self checkouts in many ways. We had people come into our store to just leave with baskets of stuff, load their pockets, take stuff from one box to put in a cheaper one they bought, or even put raw eggs and meat in their pockets. That was with a ton of workers in the store looking out for thieves.

Errors could result in us being charged for what we didn’t buy. The amount could be lower or higher. Some people go through their whole receipt double checking it. Many of us just watch it as they ring it up, deal with it there, and stop thinking about it once we leave. Could this impact peace of mind?

Cost to operate seemed like it might be high. Aldi and Costco show a traditional self-checkout can be cheap to operate on top of serving high-end customers. (See theft/shrink, though.) I figured the machine learning alternatives would cost thousands to tens of thousands a unit for combinations of cameras, servers, and licensed software. They’d also need to be monitored and/or trained by humans which this article just confirmed.

I was also worried about fraud. My company bought a solution that monitored customers with infrared cameras, monitored checkouts, and predicted how many registers we needed. That was a terrible system that hurt us more than it helped. We were required to act like it worked to please those who spent hundreds of millions on it. I wondered how often such dishonest referrals helped the supplier sell them. Made me skeptical about more AI.

Finally, the human touch that comes with good cashiers increases customer satisfaction. Our customers often looked forward to seeing us. For some, we were the only or best social contact they got that week. Others liked getting out of the house for a long period of time. Sometimes they’d have us push products and services which, normally a negative for all, is largely how they drove up credit card use and feedback on the customer tracker.

For those reasons, I advised against using AI for anything other than helping employees do their jobs more easily. If for that, that it would be a support tool instead of trying to replace skilled, decision makers. Then, simply invest those millions to hundreds of millions in the employees. Worked for Costco and Publix. :)