Hacker News Reader

Comment by LaundroMat | Hacker News Reader

I once had a retail client wishing for an "in-store Google Maps" that would guide customers to the products they are looking for. "So if I'm looking for pasta, the app should show me on what aisle to find it. That shouldn't be too difficult, should it?", he asked.

I asked him how the location of products is stored now. "Is there a system, a database that keeps track of where in the store, on what shelves, product is stocked? Something the app could use so it knows where your pasta can be found?"

He didn't understand the question. So I asked him how workers who stock the shelves know where to put what products. There must be a system that gives the workers that information, right?

"Oh no", he said, "there is no system. They know through experience and gut-feeling."

View on HN

Most modern supermarket chains are organized in such a way that location data should be available (or at least calculable). To assert that there is no "system" to a store's layout and that employees know through "experience and gut-feeling" means that this retailer was not conforming to industry norms or was just not a very large retailer. All supermarkets have some sort of product taxonomy, categories and subcategories, that influence the macro-placement of a product within the store. This discipline is known as "space planning", where large categories are placed within the foot print of a store's layout. Within the category placement, stores use "planograms", which are shelf-by-shelf layouts of which products go where, how many facings they have, how much footage is designated to the sub-category, etc. While its true that workers for the most part know the locations of items by experience, this experience is informed by their ever growing knowledge/intuition of this taxonomy. (Usually, products are organized by aisle in the backroom when the delivery comes in; workers are simply taking groups of products that have already been organized out to the floor). The planograms themselves are usually stored in a database, and this data could be cross-referenced with other common data sources (product, pricing, etc.) to determine store location.

You forgot about the manufacturers who pay for specific product placement. Sometimes the store workers will only have a vague idea because the supplier has been given full responsibility for shelf layout.

This usually only applies to certain categories and depends on the pricing model of the supermarket (and the supplier/purchaser power dynamics that depend mostly on the size of the retailer). Soda, chips, crackers and cookies are the most prominent categories that have any significant amount of placement power, and that's mostly because the big players use a direct store delivery model that helps alleviate labor costs at the store level.

Apparently, you are not familiar with Trader Joes. Some (all?) of their stores regularly rearrange the entire store. My assumption has been that they are hoping for you to serendipitously discover new products/categories and purchase them. The produce section generally seems to remain fixed.

Rearrangements like this come from the top down. They are systematic, and done using the same planogramming techniques as most large scale retailers.

Slightly offtopic of this discussion but I am wondering if you (or anyone) knows why Home Depot does not set up every single store the same, identical way? Sometimes stores are mirror images of each other and sometimes electrical is in the front and sometimes its in the back.

It seems so inefficient. I can sort of understand having the garden center on either side of the building depending on the lot but the inside of the store being different from location to location is baffling to me.

Wouldn't the entire ecosystem from people who stock shelves, to customers to store managers benefit from an identical layout in every store nationwide?

I'm picking on home depot here but the same question could be asked of any big box store.

They could have a minimap at the start of the store to alleviate it too.

Perhaps they simply don't mind customers having to wander or ask though.

I feel your frustration with Home Depot in particular - I live right in between two stores with wildly different layouts, and it drives me bonkers. One store seems to have the "traditional" layout, that I can successfully navigate off of general muscle memory from stores past. The other has departments placed so differently that I end up walking to where something "should" be, register I'm in the weird store, and open up the minimap in the app to find where I actually need to go.

From what I can tell, it's based mostly on optimizing to the local foot traffic. The one that throws me off seems to be more consumer focused. It's in a suburban area and has a _lot_ more internal space dedicated to home decor (lighting, curtains, flooring, etc). Then every other department seems to be rejiggered like Tetris blocks to fit within the remaining space.

The store with the more intuitive/familiar layout has mostly the same product assortment, but far less space dedicated to displays and "consultation" style areas (the consumer oriented store has an entire section carved out for custom curtains, complete with an associate consultation counter).

Make an app that texts a contractor those questions and lets them drop a pin on a map. Then bill 250,000 a year for the contractor and call it software as a service.

As as art project, someone made a camera that, when you snap the shutter, prints out a text description instead of an image. Under the hood, the camera actually submitted the image to Mechanical Turk.

That sounds like a super fundable MVP.

In that case the app can display the same helpful message: “Please follow your experience and gut feeling.”

I can tell you that most major retailers (target, walmart, etc) most definitely have a system. It not only specifies which aisle to put things on, but exactly which shelf, and how much space is allocated for that product. I can't remember what they're called, but they're required for every product sold in those stores. When I worked at Walmart in college, we were issued these plans weekly. Obviously this is probably not the case with smaller retailers.

So the best way to solve this is to have an in-store map to find workers of the store so you can ask them.

Your comment is funny. Have an upvote.

Seems like a perfectly good request. Stores have an incentive in providing this information, as they currently have to pay employees to provide this information directly to customers. When I was a retail employee (Kmart, in the 80s) it took a long time to learn this information, and it wasted a lot of my time. (as often as not, I had to ask someone else for the first few months)

I wasted a 3-4 minutes of a Safeway employee's time in this way the last time I was there a few days ago (I couldn't find birthday candles).

Theoretically, the solution you described would probably save a ton of money overall. I have no doubt it will exist in the not-so-distant future.

It’s wasteful if you assume it could be automated, but helping customers find things is just “service” most of the time.

Yes when using the word "waste" I am comparing it to being automated, which it most certainly can. Just as we might consider it a waste to have an elevator operator, when that task has been automated for nearly a century. At one time it was just accepted as a cost of running a tall building.

You're still missing the point. There is a point where the systemic rigidity enforced to make automation feasible is far more expensive than the value it actually produces by existing.

I.e. In order to replace those store employees, you have to build all your buildings the same way, stock them in a certain way, hire maintainers to input tracking data, hire auditors to ensure customers aren't tampering with the store.... So on and so forth.

It'd be a great way to protest though. Every customer goes out of their way to disorganize the store until they hire locals.

I don't think I'm missing any point.

Of course the economics has to justify it. And they will. Everything you say could be applied to just about everything that has been automated in the past. I mean, it sounds like someone arguing "but librarians can do this or that that Google can't do." And yet, most of the time, when we need information, we can find it a whole lot more quickly than going to a library and asking someone for help getting us what we need.

In this case, there will still be store employees if the system fails to identify what a customer wants. But that will be the exception, and asking for help finding things will be like, today, someone asking for directions at a gas station. Most of the time, you can just say "where are the birthday candles?" to your phone and it will tell you, and it will have been far less effort for both the store and the customer vs. having to go find and employee. (and yes, I'm counting development time, entry time, and all that, averaged out)

For a typical Safeway or Home Depot or Target, there are already tons of benefits to keeping track of their inventory and where it goes on shelves. It's already almost certainly in a database somewhere. They need this so they don't have to spend so much time training employees, for one thing.

I guess all I can say is, come back in 5 years. I'd bet big money you will see this, whether in Google maps or in the various apps by retailers. (I mean, even 10 years ago their was an app that did this on some level: https://mashable.com/2010/11/22/aisle411/ )

Home Depot, IKEA, and Microcenter already have apps/websites to tell you which shelf has your product.

Yes. It seems like a fairly obvious idea and I have no doubt it will become more and more widespread and built into general purpose apps (such as maps)

Maybe the workflow could be reversed. First let the workers put the product wherever they feel like; then get the information into the application. From user perspective, the easiest way would be to take a photo of the shelf using their smartphone. (Either the worker is required to take the photo after putting a product on the shelf, or you would have another worker periodically walking through the shop, making photos. I assume the products do not change their places often.) The picture would be analyzed, possibly additional questions asked. The image recognition could be helped by a database that would know which products are typically placed next to which ones.

Okay, I assume the client's next question would be: "Why does it take 10 years and 10 million dollars?"

Aside from an interactive map being overkill, are you sure you weren't complicating things here too? It seems like you were expecting some sort of system integration, but the customer wanted to just email an excel spreadsheet when there were updates, if they'd even thought of that.

I don't know how long ago "once" is, or how big this client was, but every modern large grocery company is pretty data-driven, especially about where products are placed. Lots of places have robots crawling the aisles in real time updating stocking information and taking video/images of product presentation on the shelves. They are presented as "cleanup robots" and they put googly eyes on them and a name tag or whatever. I even once asked several employees about them, and they just shrugged and said "yeah it's the cleaning robot." But it's a column 6 ft high it's not a roomba. It has a clearly visible array of cameras on each side, and who knows what other sensors.

Check out this image search, for "grocery data cleanup robot": https://www.google.com/search?q=grocery+data+cleanup+robot&r...

This is true but generally only for grocery and a handful of the big box retailers.

The merchandising information is typically well known along with product taxonomy, but the mapping of this relative information to a physical layout isn't always cleanly implemented or available.

A typical scenario is sku 85111 is grouped in a product hierarchy (family > sub category > subdepartment > department or the retailer equivalent) and that sku is assigned to a planogram or set of planograms associated with a display fixture (which sometimes have a hierarchy of their own (e.g. peg > shelf > section > aisle) with mixed relative distancing. There's generally a departmental divide between who manages product and merchandising information.

Many of these robots to my knowledge use the products or special tags as unique identifiers as opposed to an indoor-positioning system that ties back to an exact product location. So it's frequently easy to say that a product is in aisle 20 section 2 shelf 1 it's not easy to say from your current location go down 8 aisle and up half an aisle, look on your left.

That'd be pretty fun to make assuming they could bankroll it.

Your replies are a mixture of "This sounds false - large supermarkets would have this data" and "This is just an ML use case" and i think both miss the point.

Work backwards from the customer need. If they truly need this system, and are willing to pay for it, then it sounds like what you need to build is:

* A database for product taxonimies/locations * An internal tool to make it easy and quick for staff to go through and log the locations of products to build out the data model * The actual maps UI and display tech.

The thing is, this might not even be that much work. It's all a factor of how many "categories" the store expects customers to be able to search for. If the granularity level is "pasta", you can probably get away with 100 entries, and 1 days of grunt work for a staff member to log the location in the store where "pasta" and "coffee" and "condiments" etc are.

This story is not a good example of a difficult customer or a difficult technical problem.

and what's wrong with that? the client is asking for an ML solution. or, more basic, a stock person still has to stock the shelves, so they can scan as they stock. or customers themselves can report inventory locations. just tell the customer a random (at first) location for pasta. when you sense they've gotten there, prompt them whether it was there or not. amortized over time, this is an immeasurable fraction of customers that would have to do the inventory work.

i get it, you're trying to give an example of an impossible request, of which there are many, but this just isn't one of those.

What you're describing would be a nonstarter in retail in my experience. They care too much about customer experience to amortize the cost of a bad experience. Typically, step 1 is capturing data via added company process about the thing you're gonna automate. A lot more boring than what you describe.

no, that's exactly what i described. a paid existing employee collecting location data while stocking.

i actually do think that customer-finding is a good approach, but it's up to the client to decide isn't it. the comment i'm replying to was about impossible requests by clueless clients, not about bad but possible implementations.

even if the idea is in fact bad, which again i dispute, it's always good to give a client an alternative that stinks, so that the idea you want them to accept seems "obviously" good. i forget what this technique is called.

EDIT: after an hour of searching, I believe it's called a "duck". based on the legend of Battle Chess, but of course the technique has existed long, long before Battle Chess.

That sounds incredibly customer hostile. "Let's send you to the wrong location and waste your time, oh and while you're there, do our job for us!"

>just tell the customer a random (at first) location for pasta. when you sense they've gotten there, prompt them whether it was there or not.

Pissing off an enormous number of customers in the process. You will run out of customers at the store willing to use such a useless app before the thing learns where the pasta is.

Your stock person idea makes sense if they keep the shelves stable, but "gut feeling" organizations tells me that things can shift around over time.

enormous??? once you've located an item (say a low single digits of customers per item), the next <uncountable> number of customers now get to use that data. the number of annoyed customers will be in the "epsilon" fraction -- uncountably low.

and you can give an incentive to customers. give them a 10% discount on their total order when they checkout, if they've reported the correct location of an item. don't tell them about it until they checkout, to avoid gaming.

there's many more ideas here, i don't need to iterate through them all. i'm impressed by the lack of creativity on this subject.

> things can shift around over time.

not just that, but products come and go. it should be obvious that this is an ongoing task when stocking, not a one time inventory location task.

> the client is asking for an ML solution

In other words, nobody has an idea how it will work. ML doesn't generate essential data for you, or solve problems which you cannot reasonably describe a solution for.

Reply on news.ycombinator.com