←back to thread

Amazon Go

(amazon.com)
1247 points mangoman | 5 comments | 05 Dec 16 14:57 UTC | HN request time: 1.177s | source
Show context
elicash ◴[05 Dec 16 15:31 UTC] No.13105963[source]▶
I worked at a grocery store for several years, and one thing I recall is customers CONSTANTLY putting items back in a random aisle, rather than where they found it.

I wonder how this tech deals with that? Maybe they figured that out, too. But I was amused in the video when I saw the customer putting it back where it belonged, because that's not how I remember that going...

All that said, this is fantastic and exciting.

Edit: I also hope they're already thinking about EBT cards and WIC.

replies(26): >>13105994 #>>13106026 #>>13106046 #>>13106095 #>>13106097 #>>13106098 #>>13106177 #>>13106252 #>>13106276 #>>13106292 #>>13106365 #>>13106391 #>>13106456 #>>13106541 #>>13106638 #>>13106641 #>>13107002 #>>13107318 #>>13107752 #>>13108231 #>>13108233 #>>13108570 #>>13110608 #>>13110959 #>>13111172 #>>13170269 #
makecheck ◴[05 Dec 16 16:28 UTC] No.13106541[source]▶
One interesting benefit of the “detect removal from shelf” concept is that there might finally be a time stamp associated with the removal of the item.

Right now, if you see a random perishable item sitting on a shelf, you HAVE to throw it away because it could have been there for a long time. On the other hand, if you can see that some Frozen Peas were only taken off the shelf 2 minutes ago, you can just put them back and they’ll be fine.

Although, what I’d really want is not only the time stamp but the customer. I’m sorry but if you cost the store $25 by leaving a damned ROAST in the cereal aisle, I would be perfectly happy to never let you in the store again.

replies(5): >>13106589 #>>13106649 #>>13106687 #>>13107061 #>>13107485 #
saosebastiao ◴[05 Dec 16 16:34 UTC] No.13106589[source]▶
UPC codes are not uniquely identifiable to the unit, only to the product. You could narrow it down to all of the people who took that item off the shelf, but you'd probably be into surveillance reviews after that.

An option is to label all the items with a unique identifier, but that is typically seen as too costly, which is also the reason Amazon hasn't fully fixed the FBA counterfeit problem.

replies(1): >>13106694 #
jonknee ◴[05 Dec 16 16:44 UTC] No.13106694[source]▶
Amazon is boasting it involves "computer vision, sensor fusion, and deep learning" so there is definitely quite a bit more going on than the UPC code. It also doesn't appear any UPCs are actually being scanned in the first place.
replies(2): >>13107760 #>>13108602 #
karambahh ◴[05 Dec 16 19:40 UTC] No.13108602[source]▶
I think it might be purely rfid based. All major rfid vendors have theft prevention devices which can detect in real-time products crossing a given line.

Decathlon, the world largest sport goods retailer already use that technology.

They still have checkout lines but technically speaking they would be able to charge you when you leave the store.

Last time I spoke with them they were using Embisphere hardware, but any vendor, Checkpoint for instance, could be used to similar effects.

If all products are rfid tagged (which is entirely possible given the current price of metallic ink) then this store is at most state of the art.

If they actually use vision techniques then it is actually quite a feat. Current vision techniques used in retail are either too crude (when based on the store cctv cameras) or too costly (another French company, IVS, has demonstrated a self service buffet style automated checkout but AFAIK it is still prototype).

Disclaimer: I have links with investors in both Embisphere and IVS.

replies(1): >>13108992 #
politician ◴[05 Dec 16 20:15 UTC] No.13108992[source]▶
The last time I checked, the price of a printable RFID tag was around 10 cents. That's far too expensive for a 15oz can of beans.

What sort of prices are you seeing?

replies(1): >>13109099 #
1. karambahh ◴[05 Dec 16 20:25 UTC] No.13109099[source]▶
The ballpark of 10 cents per printed RFID is right.

I have reasons to believe several ink manufacturers/printers are working on an order of magnitude less per tag.

But are you so sure that 10 cents is too much?

What do you think the gross margin on that can is ? ;-) What do you think the net margin on that can, once the checkout lines (and their personnel) are removed, would be?

If a stitched RFID chip on a 2€ thsirt is currently cost-effective for Decathlon, I see no reason why printed RFID tags would not be cost effective in the very near future, if it is not yet the case (and once again, I have reasons to believe it is almost already the case).

replies(1): >>13109858 #
2. politician ◴[05 Dec 16 21:32 UTC] No.13109858[source]▶
I share your optimism - the future of retail belongs to self-service with the checkout process integrated into the shopping cart. Grab and Go.

However, I don't think that end game is the first use case for an RFID solution - inventory monitoring/LP/loyalty can demonstrate ROI long before the checkout lines are removed. For those use cases, the unit price needs to be much cheaper, but an order of magnitude improvement might be right.

replies(1): >>13109991 #
3. karambahh ◴[05 Dec 16 21:46 UTC] No.13109991[source]▶
You are absolutely correct. Inventory monitoring and LP are where money is to be made (or rather "not lost" :-) ).

I have skin in the game on the loyalty part so I will abstain from speaking on that.

Coming back on your thoughts about inventory, the birth of Embisphere, the RFID company I spoke about earlier (and the reason why Decathlon started to use them), was solely the invention of a "racket" for fast instore inventory

http://www.embisphere.com/en/rfid-products/embiventory-power

Other uses (checkout, LP) were almost an afterthought. The sole gain on speed and accuracy of in-store inventory was enough to decide Decathlon to add RFID chips on all its inventory and Decathlon inventory is massive ! They are in the range of 30-70k active SKUs, with sometimes hundreds of thousands units of stock. Sell that in hundreds of stores worldwide, add the warehouses in every country and the manufacturing facilities in China, and you end up with millions of euros of investment just for that damn inventory ;-)

replies(1): >>13110298 #
4. politician ◴[05 Dec 16 22:20 UTC] No.13110298{3}[source]▶
Nice! It looks like Nuukik is well positioned to take advantage of that new infrastructure!

It would be lovely if you developed an in-store navigation capability. It's so frustrating to run into a store to pickup something, and not be able to immediately find it.

The results of periodic scans should provide a decent point cloud that could be used to determine shelf/aisle geometry without a blueprint. Foursquare uses this sort of approach for its interior mapping process, but they can't tell me where to find the bean dip. There are multiple obvious ways to monetize that dataset including simply selling it to Foursquare.

replies(1): >>13110611 #
5. karambahh ◴[05 Dec 16 22:55 UTC] No.13110611{4}[source]▶
We may or may not already have developed proofs of concept of in-store navigation systems for european retailers ;-)

In-store location of products is deeply linked with very complicated discussions between retailers and product manufacturers.

Moreover, facing is a very strategic part of retail and I doubt retailers would be happy to release their facing strategies to outsiders or competitors.

Even inside a retailer's organization, several opposing views exist, between maximizing breadth of product range, giving prime exposure to the private label, etc....the equation they have to solve is very complicated and I don't think there's an ideal solution to this. A retailer facing strategy is linked to its core values. It has a direct impact on its bottom line and an indirect one: the consumer's unconscious perception of facing "strategies" is probably very significant.