What the article doesn't mention: Aldi optimizing the speed of their cashiers is nothing new. The northern Aldi ("Aldi Nord") for a long time had three-digit article codes for every product that cashiers had to know by heart--which also meant that they couldn't have more than 1000 different products. In fact, that system was so efficient that they completely switched to bardcodes only in 2002 as they didn't see any advantage before that.
Interesting - Part of me wonders about what their accuracy rate was with that kind of system. 1,000 different 3-digit numbers is still a lot to remember. It's possible, but it would seem to add a pretty big barrier for a lot of people who would otherwise be fine as a cashier, elsewhere. Though when I worked at CompUSA in the early 90s, we lacked barcode readers and each product had a 6-digit SKU which meant only being able to remember a handful of them for the most popular products[0]. It was also horrible for accuracy because each SKU was assigned in order, so fat fingering any of the numbers would result in the wrong product being rung up. This was often caught, but if it was the last or second to last digit that was missed, it was likely to end up being a different version of the same product (more RAM, different CPU, etc) and it would be missed[1] resulting in either the customer paying more (unlikely - those were not missed) or paying less (likely, because the cashier wouldn't notice and the customer was less likely to say anything if they happened to actually notice and the result is them saving a few hundred bucks).
[0] We did have one cashier that I'm fairly certain had well over 90% of the products in the store memorized (we'd grab random things trying to throw him off, but he was like a machine). He had brain surgery a decade prior or so and said the way his memory worked changed shortly afterwards.
[1] The worst case I recall was a $500-$600 memory upgrade kit for an Compaq laptop that was fat fingered, resulting in the generic being rung up. It had an identical description but was half the price. A few days after inventory was done, they traced it back to a single customer who bought every one we had in stock in one transaction. The customer probably knew the price was way too good and decided to buy them all as a result (it's unlikely even he knew it was a cashier error since the product looked right on the receipt).
Well, I am not that familiar with it, but given how much attention they gave to optimizing this, my best guess would be that they made sure that similar products didn't get similar codes. And I guess for a grocery store, mistakes aren't that big a problem, especially for ones that mostly don't sell anything but their own brands (and thus normally don't have all that many similar products at different prices anyway).
Also, yes, the rumor always was that Aldi paid well. Presumably the savings in additional staff and space and cash registers was still worth it.
I wonder if there is some information theory paper to be written here. On one hand you want the most common products sold to have low Komogorov complexity codes (111, 123, 147, …).
Then there is the question of the distance between similar products and similar codes: is it better to have similarly priced products have similar codes, so that errors don't amount to much in the final tally, or to space out similar goods so that confusion is avoided?
Maybe there is a whole phase diagram of optimal strategies that depends on the error rate. But then again one has to have a good error model. Do errors come from mistyping or from misremembering?
Aldi cashiers were like machines. They never had to stop and look-up a product code like modern cashiers do when the barcode gets rejected. They were just hammering the codes into the registers and they matched the speed of barcode scanners.
"What the article doesn't mention: Aldi optimizing the speed of their cashiers is nothing new. The northern Aldi ("Aldi Nord") for a long time had three-digit article codes for every product that cashiers had to know by heart--which also meant that they couldn't have more than 1000 different products. In fact, that system was so efficient that they completely switched to bardcodes only in 2002 as they didn't see any advantage before that."
Hmmm ... so you could populate an Oh By Code[1] with your list of grocery codes and just give them that single code, which they could look up and parse ...
Well, provided your grocery list was not longer than 4096 characters...
