Hacker News Reader

Comment by roland35 | Hacker News Reader

Thanks for the insight, that seems plausible to me! A moving robot doesn't make too much sense in this application since the products and shelving is stationary. It just adds a ton of complexity.

What's interesting is the Amazon store concept which just like this post is heavily based on cameras watching the shelves.

View on HN

The idea was for the shelf scanning robots to send messages when they found a low stock item or a mess, and then a human could just load up a cart with stock from the back room and head directly to the problem areas.

This would be in contrast to having people regularly walk every aisle and examine everything personally, going back and forth to get stock repeatedly. Much time was wasted.

As a former Walmart employee I was actually excited to see these proposed.

But now with so many people walking the aisles collecting items for pickup, I guess they’re just going to have them send “problem detected” messages instead of the robots. Which is fine I guess. Shopping there has become an even greater hassle with all the employees and their pickup carts, but that’s a whole other issue.

All items are scanned on the way out anyway - you should be able to get an idea when an item is low on stock from that alone, no? Sure there will still be things like people in the store that haven't checked out yet, but...

What am I missing?

Maybe that data is not sufficiently relayed to people on the floor? It also misses theft, although that can eventually become an assumed rate.

That's what I mean, though. They have decades of data at this point, you can likely guess what the theft rate is.

How much do these margins really matter? Probably not enough to have a robot that's very pricy going up and down isles.

When an item comes in from a truck it either goes directly onto the sales floor, into overstock (a high shelf above the other items), or into the backroom. People put things back in random places, stuff get destroyed (think of dropping a can on the ground), stuff get stollen, cashiers make a mistake, people make a mistake in self checkout, probably other stuff I am missing. It is a store, things are chaotic.

Let's take the journey of some cans of Goya black beans. 12 come in on a truck. The shelf capacity for them is 4. The person unloading the truck takes beans to the sales-floor, puts 4 on the shelf, puts 4 into overstock puts 4 into the backroom. Joe walks in takes a can, drops it and it bends. Jane comes along takes a can, continues shopping, the changes her mind and leaves it somewhere in the cereal aisle. Janet comes along and takes a can, takes it to self checkout but forgets to scan it. Jacob comes along takes a can, takes it to the checkout, and 1 can is subtracted from inventory. Inventory shows 11 cans, but there are actually 0 cans out on the sales floor. And I haven't even gotten into stuff being misplaced in the backroom ect.

I work (transitively) for a very large grocery retailer that rhymes with "Broger". Live inventory data is mostly junk -- it's sorta reliable in broad strokes, but you can't rely on it for determining out of stock events. There'd be all sorts of false positives and false negatives.

As someone who has had to clean and analyze inventory data, I can't agree with you enough.

In every company I've worked or consulted for, inventory reconciliation is always an ongoing headache for every business with a retail component. It is unbelievable how much waste and loss comes about from this.

Are their any companies achieving some level of automation? I would be very interested in learning more.

Like 20 years ago I worked at a grocery store when they were introducing "live inventory". Goods got scanned when unloaded from truck, and of course the register.

One day a customer complained we had no more washing powder of the most popular brand. Boss came, said they just got some the previous day, checked the inventory system and saw there should be three "trays" left, each with 12x packs of 1.2kg. He went to check the back, nothing there.

After a bit more searching he had a hunch, and checked the cameras. Sure enough, someone had loaded the three trays into a duffel bag earlier that morning and walked right out.

So yeah, even with all the normal human elements in the logistical chain, kinda hard to keep track of blatant theft.

Theft, misplaced items, etc. Employee grabs it to use for the store, damaged items, etc etc.

With only using in-out the inventory system thinks there are plenty to sell and won't reorder, but if it's wrong the shelf is empty and can't be reordered. There are posts from Walmart people online complaining about this exact problem.

Affects me as a customer, hit their website says they have 10 in stock for pickup today, drive over and the shelf is bare. Ask, they say the inventory is probably wrong, check another store because they can't reorder.

Now they are missing x items + y lost sales. That's why place do inventory.

> That's why place do inventory.

Obviously, we know why places do inventory.

We are talking about paying an expensive robot to go up and down isles to track inventory and if that is worth it.

This doesn't make sense: WalMart doesn't really have a backroom. It comes off the truck and goes onto the shelves. The backroom is too small (compared to the size of the store) to have spares of everything on pallets that are moved to the shelves as needed. This is by design - spares in the backroom are excess inventory which needs to be paid for. I'm sure there exceptions for a few things that sell in large numbers and take up a lot of space (bulk packs of toilet paper for example, keep them on a high shelf and move a few down as needed), but for the most part WalMart is against inventory.

WalMart already knows how much of something they have because they get everything scanned at the register. Sure there is a hour delay between it leaving the shelf and it being subtracted from inventory, but they need more than an hours worth on inventory on the shelf anyway because it takes time to load a truck with whatever they need more of. Shoplifting is a problem that mucks with this (I could actually see this being a useful help to that problem, but I'm not sure if it is useful enough)

Doing inventory isn't just how counting how much of item A exists in a store for accounting purposes. More importantly it is making sure that the there enough items on a shelf for customers to purchase. This is known as "on shelf inventory".

An item in the backroom does not count towards the on shelf count. An item that is on a high shelf to be moved down as needed (overstock) does not count. And there isn't just a bit of overstock, most aisle the top shelf will be overstock. Also Items can get damaged, meat can expire ect. A customer can pick up an item decide he doesn't want to buy it, and throw it back on a random shelf.

The job involves going to an aisle, scanning the items on the shelf, and if the shelf count is not full either moving items down from overstock or from the back room.

But once again you get that data from the register sales. there is a delay, but for the most part there are enough items on the shelf to last through the delay, or you would have sent someone to move items to the shelf anyway. Only when someone buys all of something is there are problem.

I'm extremely curious about the loss rate at the Amazon stores. They're relatively new, but I'm going to be a bit surprised if they don't wind up having half a store walk off on them at some point. I think we've all seen enough examples of what happens when you release algorithmic systems into the wild.

Reply on news.ycombinator.com