Amazon Delays Opening of Cashierless Store to Work Out Kinks | Hacker News Reader

Amazon Delays Opening of Cashierless Store to Work Out Kinks

wsj.com

107 pointshbosch101 comments

All it takes is a loss factor of a few percent, and the entire model breaks down.

This is the same reason that long-range RFID (UHF RFID) never worked for cashierless checkout too. You couldn't ensure 100% coverage. Some tags just couldn't be read -- either due to adversarial tactics (eg. shielding with your body) or material properties (eg. metal) -- and the engineering countermeasures to overcome these challenges made the tags too expensive for low-margin goods.

i imagine any such store has a ridiculous amount of surveillance with probably a membership model like costco with scanning so they know who you are when you enter the store. more than do-able these days.

i honestly don't want amazon to continue to get bigger it's already big enough as far as i am concerned, but that's not how big business works. consolidation is the way and the government does little to discourage it and allow smaller players and more diversity.

RFID is expensive. Even buying the cheapest transponders in bulk, you're still looking at $0.01-$0.05 per label - that's enough to kill the concept entirely for high-volume, low-value goods such as supermarket/grocery store goods where the profit-margins are thin enough as it is, if not loss-making.

How much per item do you pay for cashier staff though. Might be the same as the tag cost. Also in regards to RAIN RFID tags vs dumb print tags yep you are talking about a 5 cent increase. But you also lift inventory accuracy from ~70% to 99.9%.

You still need some staff, sometimes items are damaged, the item doesn't scan, the payment won't go through, the customer changed their mind etc etc. You can't get rid of all staff.

You can just replace them with self-checkout. You still need some cashiers (as you would with RFID) and don't have the cost of extra tags.

A cashier can process tens of thousands of items per hour, at a wage of under $15 per hour. This means that the cost of a cashier can be around 0.1¢ per item sold for a busy store

Generously assuming a cashier can check one item a second for an hour, he can check only 3600 items. "Tens of thousands" seems off by at least an order of magnitude.

I was thinking of some costco cashiers I've seen, and they do far more than one item per second. That said, when you consider the downtime of processing the transaction, your estimate is probably not too far off. It makes the low end cost of transponders comparable, though just.

This seems easy to estimate. At my local large suburban supermarket doing weekly shopping for a hungry family I have a receipt with the last line listing "53 items". So at 5 cents per tag, that's $2.65 of tags ending up in the landfill per week. The cashier takes less than 5 minutes for the entire transaction including scanning and payment and small talk, so (5/60)*7.25 = about 60 cents. Now cashiers are not 100% busy and get breaks so I would round the ratio down to RFID is currently about 3 times more expensive than human.

Of course a cashier can supervise up to 16 self checkouts in a low crime demographic area, so add in a new factor of 16.

For people who don't care about cost and scoff at $2.65 per visit, for a mere $9.95 they deliver...

Tags used to be 60 cents. Now 5 cents. If they got used for masses of things e.g. groceries, they'd go down to a negligible cost. Folks like WalMart can actually drive this themselves.

Walmart already uses RFID in the backroom and for palettized loads, but not for stock that customers handle.

In my own work with RFID, the main problem we have is (after disregarding the cost of the transponders) is getting someone to put a transponder sticker on every single widget - stores already stopped using those "pricing guns" when they adopted UPCs/EANs, they don't want to regress backwards.

The solution is universal adoption of the EPC RFID system, but that requires all suppliers and logistics infrastructure to also adopt it. It's happening... just very slowly.

If they got really cheap, you could imagine 'spraying them on' a pallet and then wanding each item in. Combined with a simultaneous read of the RFID tags that 'stuck', and presto. Just brainstorming.

If they're using serialized EPCs (which track a unique item, not just SKUs) then this won't work because need to ensure each physical item is uniquely tagged, and that two tags with different serial numbers aren't on the same item (though duplicate tags with the same serial are fine) and that no two items have tags with the same serial.

If they're using non-serialized EPCs (so just logging SKU) then they need to ensure exactly 1 tag is on each item - no more and no less - otherwise inventory counts will be inaccurate.

Unique tag serial numbers would work fine. No duplicates (don't even understand how that would happen). More than one tag not a problem (enter one or all; whatever, as long as scanning later identifies the one item).

I have seen supermarket chains do away with automated checkout (Jewel which parent company is Albertsons) systems because they increased loss factors.

Here in Australia our two major chains just convinced the police to station publicly-funded officers in the checkout area: http://www.news.com.au/lifestyle/home/crackdown-of-self-serv...

Cut staff numbers without having to pay for increased security! That's just good business sense.

I'm the US we have the same thing... It's just informal, and being done by Walmart:

https://www.bloomberg.com/features/2016-walmart-crime/

> without having to pay for increased security!

They don't pay for the police?

No, the police are paid for via taxes.

Oh. I assumed Albertson's dropped these because they were slow and inefficient, not because of loss/shoplifting. The stories I can find about it suggest the same.

Jewel/Albertsons also tried portable hand-scanners to speed up the checkout process. That quickly was pulled as well.

I guess it depends on how you set them up. In the UK you'll find self-checkout scanners in nearly every supermarket that contain scales. Stealing with those isn't easier than just stealing in the first way. You can't just "forget" to scan an item.

Scanning while you shop, however, doesn't really seem to work. Tesco tried that and basically stopped it again. You either need to check really often (high staff costs) or will have losses.

> Scanning while you shop, however, doesn't really seem to work.

Carrefour does that in France, and it works like a charm (I've used that for years when I used to live there). They run semi-random manual inspection of your grocery bag at checkout (happened like once every 1-2 months).

Amazon can realistically do something run-of-the-mill shops cannot: choose its customers. In any case, this is bleeding-edge territory. The point of such experiments is to measure and modify. The margins are too close for armchair philosophising.

It is not the same, but our local Sam's Club has an app that you can scan all items as you go, and pay inside the app. Then as you leave an employee checks the receipt like they do for everyone else anyway. It works great.

I've seen that at Sam's Club, and at a couple grocery store chains, but nobody seems to use them at the grocery store. I did once.

I wish Costco would do this too. The lines at my Costco are huge no matter what day or time I visit. Sometimes I'd like to go in just for one or two things, but I decide to get it somewhere else at a higher price just to avoid standing in line behind a dozen full carts. Self-scan would fix that.

I'm guessing since they just got rid of self checkout they're not gonna be doing that.

(I heard they got rid of it because of too many people sharing cards, and no cashier to check the picture. Seems like an easy thing to fix, but oh well.)

Funny, BJs warehouse club still has self checkout lanes.

That seems rather silly. Why not check the reciept + card at exit?

How is that any better though? The receipt checking probably takes just as long as scanning the items through the till.

Unless they just take a cursory glance and just say "well whatever you're probably trustworthy"

At a shop I visit often, the receipt checking is random, triggered by some prediction algorithm based on your history and current cart contents when you swipe your membership card when you exit the store. You get selected once out of perhaps 20 to 50 visits.

As a bonus, you get a small candy bar or similar "sorry we bothered you" gift if the check does not find any errors.

And if there is an error?

In the UK with Tesco's:

You have to scan every item in your shopping. There is no immediate penalty, but who knows what the system does in the background. I guess you could get banned from the system.

Do you get something better if they find they have made an error?

Mainly because Sam's already had receipt checking at the exit, so it's nothing outside of their normal operating process.

At my Sam's, they just eyeball it. A lot of times they just seem to be counting to make sure you don't have extra items, not actually checking the contents. It only takes a few seconds.

I wonder if a "Trusted Traveller"-style program would work. You go through a small vetting process, and you build a solid history with receipt checking through their app, and then you're assumed trusted, maybe with more occasional checking, and higher trust levels?

That's how Tesco's do it in the UK. No vetting other than getting a loyalty card of some kind.

In a shop in Italy if you are a member you can take at the entrance a scanner with you and then just pay in the automatic checkouts. Periodically (there is a statistical model behind I guess) you can be asked to be checked if the stuff you scanned are the same you have. So if you do mistakes, you will be checked more often. It seems to work pretty well, even with the old ladies.

We have similar systems in stores here in Switzerland as well [1]. Very easy and quick. Fraud check is random.

[1] https://subito.migros.ch/de/self-scanning.html

Non-Paywalled direct link - https://m.facebook.com/l.php?u=https%3A%2F%2Fwww.wsj.com%2Fa...

I wish more companies would delay their products due to quality concerns. It shouldn't be viewed as a bad thing.

To be fair, this quality concern mostly affects Amazon. I don't imagine too many shoppers would complain about being undercharged. Though it is possible the reverse could be a problem as well (and that would not be something Amazon would highlight).

Anecdotally I've heard that the store oftentimes works fine in high-traffic situations. But if my memory of my time at Amazon serves right, the "P99 performance" of this store should be as good as the "mean performance", which is probably where they're being cautious.

Funny how high-availability constraints can make for ominous headlines like this.

I don't think the headline is ominous. It is not like Amazon is delivering a regulatory project that has real deadlines. They set their own deadline and made a decision to adjust it. They are also creating the first of its kind so kinds are to be expected right?

I don't find the headline positive or negative, I find it to the point and succinct.

You're right. I think I was thinking of The Verge's less tactful headline:

Headline: "Amazon’s cashier-free store reportedly breaks if more than 20 people are in it" Subtitle: "So it’s being delayed" Lead-in: "Amazon’s first cashier-free convenience store was supposed to open to the public early this year, but apparently some big technology issues have the launch on hold."

The Verge's article is a hot-take derivative of the WSJ article. My criticism was misplaced.

I wonder if these stores will suffer any more from shoplifting than regular stores. I assume they will function roughly the same, i.e. an alarm goes off as an unpaid high ticket item leaves the store, but what's the psychological impact of having a human near the front door? will people feel they are stealing from a robot rather than a person (in fact they are stealing from neither), and will that lower the cognitive barrier to theft?

it comes with built-in plausible deniability for the guilty, too -- "sorry, I guess your robo-store is broken, let me try signing in and out of my amazon app"

It doesn't seem to function the same at all. If an item is detected leaving the store, then the person who took the item (or the person the store thinks took the item) is charged. And if they aren't detected, there's no way to have an alarm.

how exactly is someone charged when they don't have an amazon account / connected phone / whatever method of payment is being used here? are you saying you need to swipe your card on the way in?

Yes. You need to have the Amazon Go app installed on your phone, and you scan your phone at what looks like a turnstile to enter the store, from the store's PR videos. Then the store keeps track of what you pick up using computer vision, your phone and various sensors, and charges everything you walk out with to your account.

Unless you can avoid the requirement to load an app and scan it to get into the store, walking out of the store with goods stuffed in your jacket or bag isn't shoplifting, because the store knows you picked those things up and is charging them to your Amazon account.

ah, interesting - thanks for the explanation (and sorry if that was mentioned in the article, I couldn't get past the paywall).

So children are not allowed in the store? What's stopping someone from just stuffing their kid full of high end goods and asking them to wait in the car?

The same thing that stops people from doing that in a regular store? Mostly good morals backed up with cameras and the occasional security guard.

Theft rates can get pretty high, look at the Tide theft phenomenon.

Everyone entering the store has to be scanned in. A parent could bring a child in, but they are scanned in on the parent's account. Similar to how at some offices, visitors are scanned in using the employee's badge. At least, that is my understanding of how it works.

I don't know how they designed it, but if they use anything like a phone-unlocked turnstile or revolving door to let people in, they might just count every person that comes in until the gate closes again as being attached to the account of the person that opened the gate. So when any of them leave with something, that's a charge to you.

I could see it being a case of scan, pass the phone back to child/whoever's with you (or scan it their side for them), then they're all registered on that account. A bit like plane digital boarding cards. That does remove a potential fraud signal though so maybe they have a better idea.

What about using worthless phones and masks? For under $5 I can get an Android or a dozen of them, load them up with this app, and be good to go. Need a credit card? No trouble, get a prepaid card for $1.95 at the corner store, load 5 dollars on it, auth it with the Amazon account, and pull $4.99 off it. Then go get some steaks on the house!

There's nobody guarding the doors at the average grocery store. You can walk in and walk out with a steak without anyone stopping you, and without having to buy prepaid debit cards and make online accounts in advance. Far fewer tracks to cover. They also don't have every square inch of the store covered in face-detecting cameras, like Amazon's store, making pretty much anywhere else a better target once you're set on shoplifting. A grocery store probably won't even notice the missing steak, where the Amazon store will have timestamped logs of every unpaid product leaving the store and video of it happening to pass on to authorities.

Wow, you are unfamiliar with retail. Sure, you can probably steal small things, but repeatedly stealing larger objects (book size and above) is generally a pain in the ass. Hence why Costco & Sams Club make better targets than average stores, its much easier to throw something in the cart of high value and slip it by.

Booster bags will be a major issue for Amazon's stores if they ever launch this.

> Wow, you are unfamiliar with retail.

That was unnecessary.

> Booster bags will be a major issue for Amazon's stores

None of the products in Amazon's stores carry any kind of tag, so there's nothing for a booster bag to block.

They're using computer vision to identify when a customer takes a product off a shelf, or puts a product back on the shelf. That customer's virtual cart is updated in turn, and charged after the customer leaves the store. The people are what's being tracked going in and out, not the products.

I don't understand what this gets you. You'd be on camera, accessing the store using your own amazon account, wearing a mask. Do you think you wouldn't be caught?

Yes, that's how it's working for now, anyway. The app has a QR code you scan at the entrance to gain access to the store.

I think they will still have people working in loss prevention / security. I'm not sure though.

I guess there will be security that will stop non-Amazon customers, and probably help setting up accounts.

People interested might also enjoy the first, automated store: SmartMart. Made in the Windows NT days. We got one out here.

http://www.smartmartinc.com

How is Smart Mart? Their website is quite sparse.

I haven't been there in a long time. The founder struggled to convince people to deploy them as the upfront cost with robotics and tech was many times a regular, convenience store. Basically what it is. The majority of store owners would rather pay less money to exploit people with dead end jobs and deal with the shrink from theft. Definitely not a growth business but he's been opening more and more over time.

It was an old interface that wasn't as good as modern self-checkouts at grocery stores. Predated them, though, automating the whole process. You drive up to a fuel pump looking thing, tell it what you want, pay it, the equipment gets the right stuff, and it brings it to you somehow. The only people necessary are those that stock it. Probably just one or two doing a full shift on busy days with a part time one on slower ones. Pretty amazing given big companies were bragging about rolling out self-checkouts that just automated a register when he had people automated a whole store many years before.

Note: Guy's name is Michael Rivalto. He's a VC that my brothers told me. Family friend that used to occasionally take them to school in his supercars. Lasting impression there haha.

I'm surprised that Amazon didn't go this route. With their warehouse logistics expertise they could operate everything pretty much the same except that instead of putting a bunch of stuff into a shipping van it goes in a customer's car.

It blows my mind, too, that Mike is about the only person I've seen do it. It's the kind of thing I'd think Amazon would find obvious. Dollar General and Aldi each make a fortune with DG having several times the margin of much of the industry. Convenience stores are all over the place with all kinds of personnel-driven issues. I'd have thought they'd prove themselves out serving the low end before aiming for the high.

I think maybe it's because they're a technology company aimed at solving big problems for large numbers of consumers at once. They might have not thought of the cumulative effect of a ton of small deployments that their big thinking merely orchestrates and supplies. A weakness, perhaps?

In a nutshell: Their tech can't track items if there's a lot of people or fast movement.

And once they hear Amazon can't track a lot of people moving quickly... A lot of people will decide to start moving quickly.

I can already foresee a dozen people showing up in zebra-striped hoodies to see what happens when they shop together.

I'm sure that some people will but there will almost certainly be security in the store that will politely ask them to stop.

They realize that amazon now measures the height of each customer, also has a miniature drone following them around and each floor tile measure the weight of the people for that extra data point, oh and some have employees disguised as customers whom help the system when in doubt.

Weight belts, got it.

and platform shoes.

'Who help'. 'Whom' is the objective case, parallel to 'them'; a good way to test the correctness of a particular use is to substitute the latter for the former and see if the sentence is still grammatical.

Time for some "identical twin" experiments.

I am an identical twin and on my brothers surface book I'm able to fool windows hello for some reason.

> for some reason

Hmm, maybe it's because he's your identical twin?

Forgot to mention that Microsoft claims that I shouldn't be able to do that.

Really? Microsoft says a webcam should be able to instantly tell the difference between you and someone who looks literally exactly identical to you?

Amazon will counter by convincing third parties to resell their goods on the shelf. Amazon takes their cut of sales, but doesn't chip in on losses or fraud. Oh wait...

Does Amazon already do this?

Previously on HN: https://news.ycombinator.com/item?id=13955981

We had a discussion about Amazon.com comingling inventory with seller inventory.

More discussion at https://news.ycombinator.com/item?id=13924546

It's a nod to their online model. You sell android phones as a 3rd party on Amazon.com. Buyer claims "not as described". Keeps phone, puts potato in Android box and returns it. Amazon forces you to refund. Probably distressing to Amazon that they can't foist off losses of physical in-store shoplifting onto 3rd parties.

A lot of people will turn up to play basketball, while someone in a bear suit steals things.

It was a gorilla suit, not a bear suit ;):

https://m.youtube.com/watch?v=vJG698U2Mvo

There's a variant with a moonwalking bear which is rather well known: https://youtu.be/Ahg6qcgoay4

Interesting. Your post got me to finally look up the history of this [1]. Apparently it was a woman with an umbrella in a 1975 study, but made extreme by Simons in 1992 with the gorilla. The moonwalking bear is apparently from a 2008 Transport for London advertisement [2] to remind people to see cyclists.

[1] https://en.m.wikipedia.org/wiki/Inattentional_blindness

[2] https://www.theguardian.com/uk/2008/nov/16/transport-invisib...

You know you can already get things for free at current stores.

It's called shoplifting and it's REALLY EASY.

The truth is in Western Society most people don't do this.

And after an initial novelty factor (which I'm sure Amazon will account for), people won't bother any more than they shoplift now.

P.S. Nor will they shoot Amazon Prime AIr drones out of the sky like other HN threads fret.

Shoplifting by outpacing a tracking algorithm has high levels of plausible deniability.

It's not illegal to hurry, and if Amazon fails to notice you picking up a product, that's their problem due to the system they thought up.

That would entice a lot of people.

and goods misplaced on wrong shelves. Kinda hard to track.

That's what happens when you use mturk instead of computer vision.

And "a lot" here is 20 customers. For any decent shop they'd probably need to track at least 50, rather 100 during busy hours. Not sure if they can fix that within a few weeks.

Has anyone tried fooling the system as an experiment? Considering that computer vision remains relatively non-robust, I imagine this shouldn't be too hard.

On the lighter side,

https://www.youtube.com/watch?v=b-RXDCEErWQ

Want that the point of the beta version of the store?

Should beta test this where you can keep what ever you manage to get out of the store without paying.

http://www.marketwatch.com/story/amazons-cashier-less-conven...

Here's a link to a non paywalled site.

Reply on news.ycombinator.com