Hacker News Reader

Comment by laurentoget | Hacker News Reader

This kind of system is notoriously difficult to switch to. In theory a centralized system with a good algorithm should easily do better than the local employee who only has local information, however before you switch the system on, you only have the data which has been collected before and this is probably incomplete and inaccurate data, so the algorithm will miss some peaks and suffer out of stock issues, which will generate more garbage data as you cannot measure demand of an out of stock item.

View on HN

This was my thoughts exactly. More often than not, these systems are still implemented but a manual "exception" process is later tacked onto it.

It sounds like they haven't hit the normal Stage 2 phase: Implement a manual override process with the goal of allowing employees to temporarily bump levels for holidays and special sales and the like. The key here is temporary. The computer resets the levels to whatever it thinks they should be after a normally (short) period of time.

Stage 3 is to act surprised (as a corporate culture) when it is discovered that salaried managers are spending a 10 hour day every week/month adjusting all the levels. Effectively, store-level employees have returned to the original process but in a very inefficient and undesirable way.

Stage 4 varies. Sometimes you loop back to Stage 1: Zealous enforcement and the removal of any way for store-level employees to provide input into the ordering system.

Sometimes, you scrap the whole thing and go back to manual ordering. This usually keeps the algorithmic system but instead as a sort of assistant/recommendation engine. Usually it picks the top X overstocked/spoiled items in the store and gently nudges management to reduce inventory levels. This usually turns into a consistent win because management feels empowered and eventually comes to trust the recommendations, often rubber stamping most things and only correcting egregious errors: "I.E. Christmas in 2 weeks, only order 10 hams."

Unfortunately, the source of this info is directly working with a retailer trying to integrate one of these systems. After we hit Stage 3, I really did try to build a consensus around repurposing it into a weekly recommendation engine of 10-20 items to adjust in the store.

Instead we just removed the override capabilities and went right back into trouble. Apparently, there was a view I had overstepped my boundaries trying to tell the business how to treat the stores. So I left before I could get placed at the bottom of the Annual Stack and Rack. :-/

Surely they have enough data at a long lived supermarket to predict stock demand using weather factors and seasonality automatically.

I'd expect them to be factoring in customer onboarding rates, including anticipated accelerations, accounting for ongoing ad campaigns and such too.

Whole Foods whittled down the list of items that are "never" out-of-stock. Now that tier of redundancy is reserved for a select group of core items explicitly called "never-outs." In doing so, HQ implicitly reclassified the remainder of the inventory as "sometimes-outs."

Having a lot of "sometimes-out" items predictably will lead to "sometimes-all-out-at-the-same-time" categories and "some-stores-are-always-out" items. The outages are clumping, in terms of categories, timing, and location. Which is a bad and apparently unanticipated mode of failure for the procurement system.

If the problem is too much spoiled stock, wouldn't it be simpler and easier to implement a dynamic-pricing system, where the price of a product starts going down automatically (say, every 6 hours) starting two weeks before the product is marked to be spoiled? I've seen shops do this manually, but it usually only happens when they have a lot of stock of the same product that's about to go bad, so they sell 2-for-1 or something like that.

Yeah stores must spend a lot of time on manually marking things down. Plenty of room for innovation, would just need to move to digital price displays.

Actually, as a customer, I can confirm that all the Whole Foods Markets I’ve seen do have digital pricing displays for most things - they are based on e-paper technology.

But you don’t want prices changing between the time a customer has picked up a product and the time they get to the checkout. And then there’s the whole problem of the in-store prices not matching the published circular prices.

So, lots of issues to work out there before dynamic pricing can work well in a physical store.

If the price drops between picking up an item and paying for it, I don't think anyone would complain.

And you can handle the other direction with a lag. Charge the old lower price for 30 minutes.

There's no centralized replenishment system yet at Whole Foods, OTS is just a collection of guidelines enforced by inspections. 99% of ordering is done manually at the store level for now.

Reply on news.ycombinator.com