As noted in my newsletter this week, Amazon is phasing out its checkout-less grocery stores with "Just Walk Out" technology (I'm making this separate post about it for future reference...).
The technology, which relied on cameras and sensors to track purchases, is being replaced due to its reliance on human intervention. Despite appearing fully automated, it actually required over 1,000 people in India to monitor and label videos to ensure accurate checkouts. This revelation explains why these decade-old concept stores have had their deployment stalled until now.
It also illustrates in hindsight how Amazon was probably extremely strategic about the kind of Moore's law behind customers' behavior monitoring and real-time analysis. A decade ago, investing in an unperfected technology while understanding the social aspect and the possible market adoption or rejection of cashier-less stores was an uber-smart move. Expecting the technology to finally catch up with your needs as a global retail company already expert in IT was quite easy to do (do not forget AWS powerhouse!). Even if you'd run the risk of being 2-3 years off in your timing, Amazon could afford that.
However, the problem with any type of Moore's law is that it's never a law. It's only a glorified empirical rationalization of complex (and sometimes chaotic) phenomena. Even if more than a decade or so of the "law" is respected, it can end up stalling and not delivering the last % of efficiency that would be required for proper industrial applications. When self-shopping in a fully automated store works 98% of the time, the remaining 2% might still be an unbearable loss for a possibly expanding network of +500 physical stores.
And in case these AI-powered systems are still outperforming the losses incurred in regular stores where human cashiers make mistakes, where merchandise is spilled, stolen, and wasted, they might amount to killing a fly with a bazooka.
This difficult representation of how fast technology goes when it's in exponential mode, until it's not anymore is absolutely fascinating. Sure "we wanted flying cars, instead we got 140 characters" but ask yourself, what if you had the flying cars already? What would you have done with it anyway?
After all, we have a gazillion times the computing power of what NASA used to land people on the moon in our very pockets, yet we still mostly send cat pictures with it 🤗
