Comment by bckygldstn | Hacker News Reader

Comment by bckygldstn | Hacker News Reader

There's a few comments asking why this can't be done by subtracting sales from ordered inventory. All large retailers do this, I've worked on this problem, but there are lots of things that reduce the accuracy of the data-driven approach:

• Theft

• Spoiled/broken products

• Customers taking a product then browsing for another hour before checking out

• Customers buying multiple flavours of a product with the same price but different UPCs, which the cashier rings up by scanning the first product then hitting "times 10"

• There is stock on the shelf, but it's hidden behind something else or out of reach

• There is stock in the store, but not on the shelf (many retailers don't track backroom → shelf stocking)

• Product is stocked in multiple locations in the store, one of which is empty

• Plus all the usual data issues caused by wrong barcode scanned, data entry error, out-of-date information about units per box, etc

Those issues are all small, but still significant:

• For slow-moving products, being a few units off can mean the difference between out of stock and 1 month stock remaining.

• Even for fast-moving products these errors add up over time, you never really get a ground-truth event to reset the error drift. Stocktake and out-of-stock help but are infrequent and not a perfect reflection of reality.

There's a big unsexy industry working on this problem with data and human checks. The economics of robots, drones, smart shelves, and smart carts don't seem competitive right now. When you have 10k stores in the US, the expense is massive.

90% of these problems are fixable with RFID tags, which are relatively cheap to roll out and maintain. Lululemon has been doing it for years[1], any insights about why it's not more popular?

https://www.globenewswire.com/news-release/2017/05/16/125712...

This is a great question...When I was in college in the 2000s RFID was going to be a game changer. The ability to track everything...the ability to just push your cart out and pay without scanning your items. Technology is there, but my only guess is somehow it's cheaper to have stolen items and less accurate counts.

I believe Walmart uses RFID extensively in their logistics/distribution pipeline, and they have a lot of control over their suppliers, so I'd assume they would use it in the store if it was viable.

https://www.atlasrfidstore.com/rfid-insider/walmart-and-rfid...

I'd be interested to know the precise reasons for not using RFID on items though, I suspect it might be something like "tagging $1 items is cost-prohibitive, and the system isn't useful if you don't tag every item".

I'd be more inclined to suspect that the sheer square footage of their stores makes it cost-prohibitive to outfit the space with the sensor infrastructure (comparing to Lululemon, who has much smaller stores). The tags themselves are fractions of a penny (I'm seeing $0.002 on alibaba) and are just applied as stickers, so I don't think that's too much of an issue -- not to mention Walmart probably has the buying power to insist their suppliers pre-tag all items.

> not to mention Walmart probably has the buying power to insist their suppliers pre-tag all items.

That wouldn't stop them from paying for the tagging. If it happens at all, everyone later in the supply chain has to pay for it.

Thieves could just pull off the tag or swap tags with a cheaper item.

If you want it integrated then your store must convince every manufacturerto modify their production line to put a tag inside the product packaging. Reading is still a problem also, RFID works great alone but what about reading a shelf full of products at once?

IMO computer vision (like Amazon Go) is a more likely solution to the problem.

They already can with barcodes. It's harder to forge an RFID then a UPC 128 barcode (print as sticker, apply at store)

Reading a shelf full of products isn't an issue, how do you think they can wave a wand over clothes and know the exact sizes they have and quantity?

They should double down on RFID but many of the manufacturers don't want to incorporate it in their process.

The processing time of applying in the store would be enormous there's a reason very few stores still use the price sticker gun to label items, it's time consuming when there's already a tag on each item.

Also in a number of places walmart and other stores literally just open a box provided by their sources and place it on shelves.

Does anyone have a (well-founded) idea of the amount of extra waste this would create? Or how much this would increase manufacturing costs?

This is a genuine question by the way, I honestly have no idea of the scale of this when applied to a store chain as huge as Walmart.

Bulk raw wafers can be obtained for a couple cents in 120k quantities but the tooling to apply them is where the cost is mostly. Also Walmart is a huge customer but not the sole customer for the brands they stock so they'd have to convince a lot of different companies to include it to make it worth while.

I think the second is a big part of why RFID tags haven't taken hold, stores need enough of their stock to use them to be useful while manufacturers need enough clients to want the tags embedded to go through the trouble of including it in the packaging.

https://www.mouser.com/ProductDetail/NXP-Semiconductors/SL3S...

The tags are still much too expensive for grocery and low-end retail. The ~US$0.10 to buy and apply the tag really eats into profit for those segments, where average price is a few dollars and margins are a few percent.

10 cents is likely closer to being the raw cost of the tag itself; the actual costs including tooling and tracking would likely make this closer to 15-20 cents per unit at scale. It's still too high for a low-margin retail environment, where you're competing with services like Amazon Go's camera systems (including the removal of overhead associated with cashiers) which have a fixed upfront cost regardless of unit volume.

Lululemon sells a small number of luxury items at a margin of 90%. (They don't capture that entire margin for themselves, but the presence of this margin also affects their relationship with their suppliers.)

Walmart sells bulk poverty goods with a margin of 1%. It's often the only store in town it's customers can afford to shop in.

Back in the late 1980s I worked (indirectly) for a clothing rental place and looked at RFID tags to replace the barcodes that were iused to track individual garments.

Barcodes are not ideal on something that is flexible and gets cleaned a lot, but RFID was discarded as a solution then.

My point is that RFID tags have had over 30 years in the market. If they were a better general solution than barcodes, they would be everywhere.

RFID tags are problematic in a lot of environments. Warehousing in particular, especially with containers of fluid or other dense goods that can block radio signals. I can see a lot of problems at a place like Walmart. A shelf full of cleaning products sounds like a nightmare scenario.

Source: I worked at a company selling RFID-based inventory tracking hardware.

RFID is definitely not cheap, especially at scale compared to barcoding.

One unmentioned issue is that RFID is also not universal. In that, each country has it's own part of the EM spectrum regulated for RFID. And most of these bands don't overlap. So you need one tag for the US, one for the EU, one for Japan, etc.

Do you have any firsthand knowledge of how well lululemon's implementation actually works? that seems like the sort of press release where corporate and the implementation partner are very proud of their thing, but the frontline staff who have to actually manage the inventory roll their eyes and fix the computer manually every time it's wrong.

A few other HN comments on what is holding back RFID tags:

https://news.ycombinator.com/item?id=22529512

Yay, more e-waste!

I have some experience of RFID from working for a UK supermarket in their supply strategy team, who did a very limited partial rollout.

The reason is that your original premise is not true - they are not relatively cheap to roll out or maintain (relative to the cost of goods).

They are not cheap to roll out:

- Readers in all stores, including handheld devices, tills, doors e.t.c.

- If you are tagging 100% of products, it affects every single supplier production line, of which there are tens of thousands. A cost to your supplier is a cost to you, because it will inevitably affect the cost of goods sold.

- IT cost implications both internally and at your suppliers.

- If you don't get 100% of suppliers adhering, you don't get as much benefit. Good luck getting Coca-Cola to change their production lines for no benefit to themselves!

They are not cheap to maintain:

- At an estimated 2p per item, if you are shipping 10 billion items that is £200,000,000 a year just in tag costs. Alternatively put, if you sell items for £1 on average and make 4% net profit, that cost of tag cuts deep.

- There is an incremental cost for suppliers to maintain all that equipment to apply tags or build them into the packaging (which comes back in COGS).

- Maintenance of all RF readers

And inventory routines are cheap relative to this cost: - Assuming you have 1000 stores, and spend 5 hours per day on inventory routines @ £12 total cost to employ, you are talking a £20m cost (5 hours assumes gap scanning & code checking). I assume a manual financial account is still required once a year to account for failed tags.

So we have to come up with other benefits, like top-line growth driven by a better customer experience, but customers will have to be willing to pay for this with higher prices.

> Customers buying multiple flavours of a product with the same price but different UPCs, which the cashier rings up by scanning the first product then hitting "times 10"

I worked at retail (Uniqlo to be specific) and it was definetely a rule to not do this. They wanted us to scan everything one by one even if they were the same color and size. This allowed for a pretty accurate information on stock, e.g. we could see how many of a given sweater in given size and color is still in storage etc.

Probably this is too tedius for some retail (groceries with very small unit price) but very smart rule in general.

The rule is probably on the books at most retail stores, getting everyone to follow it all the time though is another matter. Cashiers and their direct supervisors could easily not understand or care about the rule and ignore it to save a few seconds on big orders.

Self-checkouts I'm sure now screw with the accuracy too..

At least with self-check out, every single items needs to be scanned, as there is no "times 10" button, and they get picky about sensing weight after each item. This can be quibbled of course (skip bagging, theft, etc).

The weight checking drives me up a wall. I use this primarily at hardware stores and I frequently can't fit large/heavy items on the output scale. Or the opposite problem, I'm buying a shim or a screw and it simply doesn't register.

For sure, after having discussed with someone working there I can tell you results are awful.

Reply on news.ycombinator.com