5 ms·

Chain grocery stores, at least, must have a running total in their internal systems. They know how much product was delivered from the supplier. They know how

Chain grocery stores, at least, must have a running total in their internal systems. They know how much product was delivered from the supplier. They know how much product has been sold (at least for countable items). It does seem plausible that they do not have an API for Instacart to interface with, though. A 20% no-fill rate (which seems about right in my experience in NYC) is far too high for most use cases, and so either they'll have to figure out something to get accurate counts, or it's a forced win for someone with an integrated warehouse.

confidantlake 8y ago
I work on exactly this problem, it is actually pretty complicated. A grocery store does know how much is delivered but it does not know how much is in stock. A simplified version is like this. 20 boxes of cheerios come in, there are room for 5 on the shelf, there are room for 5 on top of the shelf, then 10 go to the backroom. Someone steals a box from the shelf, another takes a box and puts it back behind the frosted flakes. Now the inventory is showing 20 boxes, but there are only 19 in the store, and one of those 19 is somewhere hiding somewhere. These are the easy cases. Consider shirts. A shirt can come in different sizes or colors but they all have the same upc number. We have 20 shirts according to inventory, no need to order more right? The only problem is that all the mediums and larges sold out and we only have xxls and xsms. Also items can come in packs, but are sold individually. There can be special displays, so some of the items are moved from their old location to a new location. On and on with edge cases and exceptions. The parent comment on most stores have no idea is closer to the reality than stores know because they know what came in on the trucks.
- arkades 8y ago
  > I work on exactly this problem What are some potential solutions?
  - confidantlake 8y ago
    My solutions would go in two broad categories. A. Worker methods. 1. Tell workers to scan (count) the inventory more often. 2. Do some gamification to actually get workers to scan more often. Not really sure about this, could easily have perverse incentives or turn it into some dystopia. B. Non worker methods. 1. estimate inventory based on sales data and restocking data. 2. Rank the sections by profit to prioritize high profit sections. I don't think my ideas are particularly great. Besides, nobody listens to me, I'm just a lowly programmer. I just do what they tell me ;)
panarky 8y ago
You would think so but perpetual inventory is very rare, even for the large chains. Grocery stores get less than half their assortment from their own warehouses, with the majority delivered direct to store from hundreds of vendors. The large chains have some integration with their largest vendors like Coca Cola and Frito Lay, but usually this is just for item authorization and invoicing, not for perpetual inventory. Spoiled and damaged product is often thrown away without recording it. Cashiers scan 12 units of one UPC instead of scanning each flavor of baby food or Gatorade separately. Scanning accuracy is even worse with self-checkout. Reordering is typically done by looking for empty spaces on the shelf, since they don't have the perpetual inventory to do automated replenishment. It's a real shitshow from a data quality standpoint, which is why the no-fill rate is so high with online orders. > it's a forced win for someone with an integrated warehouse Webvan thought so too. https://en.wikipedia.org/wiki/Webvan https://en.wikipedia.org/wiki/Webvan

confidantlake 8y ago

I work on exactly this problem, it is actually pretty complicated. A grocery store does know how much is delivered but it does not know how much is in stock. A simplified version is like this. 20 boxes of cheerios come in, there are room for 5 on the shelf, there are room for 5 on top of the shelf, then 10 go to the backroom. Someone steals a box from the shelf, another takes a box and puts it back behind the frosted flakes. Now the inventory is showing 20 boxes, but there are only 19 in the store, and one of those 19 is somewhere hiding somewhere. These are the easy cases. Consider shirts. A shirt can come in different sizes or colors but they all have the same upc number. We have 20 shirts according to inventory, no need to order more right? The only problem is that all the mediums and larges sold out and we only have xxls and xsms. Also items can come in packs, but are sold individually. There can be special displays, so some of the items are moved from their old location to a new location. On and on with edge cases and exceptions. The parent comment on most stores have no idea is closer to the reality than stores know because they know what came in on the trucks.

arkades 8y ago
> I work on exactly this problem What are some potential solutions?
- confidantlake 8y ago
  My solutions would go in two broad categories. A. Worker methods. 1. Tell workers to scan (count) the inventory more often. 2. Do some gamification to actually get workers to scan more often. Not really sure about this, could easily have perverse incentives or turn it into some dystopia. B. Non worker methods. 1. estimate inventory based on sales data and restocking data. 2. Rank the sections by profit to prioritize high profit sections. I don't think my ideas are particularly great. Besides, nobody listens to me, I'm just a lowly programmer. I just do what they tell me ;)

panarky 8y ago

You would think so but perpetual inventory is very rare, even for the large chains. Grocery stores get less than half their assortment from their own warehouses, with the majority delivered direct to store from hundreds of vendors. The large chains have some integration with their largest vendors like Coca Cola and Frito Lay, but usually this is just for item authorization and invoicing, not for perpetual inventory. Spoiled and damaged product is often thrown away without recording it. Cashiers scan 12 units of one UPC instead of scanning each flavor of baby food or Gatorade separately. Scanning accuracy is even worse with self-checkout. Reordering is typically done by looking for empty spaces on the shelf, since they don't have the perpetual inventory to do automated replenishment. It's a real shitshow from a data quality standpoint, which is why the no-fill rate is so high with online orders. > it's a forced win for someone with an integrated warehouse Webvan thought so too. https://en.wikipedia.org/wiki/Webvan https://en.wikipedia.org/wiki/Webvan