100 comments

[ 3.3 ms ] story [ 58.1 ms ] thread
Once their vision for "grab and go" vanished due to technological infeasibility [1] the entire premise for the stores vanished as well.

I suspect that they wanted to take a hail marry to see if somehow it was possible to get much greater efficiency compared to standard grocers, and it looks like that failed.

[1] it may come back. The technology is rapidly improving but they have bigger fish to fry ATM.

I thought they already did close them.

I know at some point they got caught basically paying people to watch cameras to figure out what products people we're grabbing. I'm sure were either at the point or very close to the point where AI can successfully do this basically 100% of the time.

So I doubt it's the tech aspect of this, more just the grossness a person feels walking into a store with Amazon's name on it. Compare this to whole foods.

Their fate seemed sealed when it was revealed a bit back that the “just walk out” technology was more hype than substance. Just lots of people watching what you’re doing on camera vs an actual AI that worked well at mass deployment scale. A good idea, poorly executed.

Reports said the “AI” was largely 1000+ people in India watching the cameras.

If Amazon actually managed to build AI that worked well at a decent cost point it would have been great since nobody likes those silly self checkout machines.

What’s amusing about all of this is that before it got leaked that it was basically a bunch of people in India watching cameras Amazon folks spoke about the tech like there was some super secret AI they developed. Since that story broke nobody there seems to want to talk about “just walk out” anymore.

I’m in an interesting place. Here in Seattle I am two blocks from one of the largest Amazon Fresh stores. It was built on the former location of a local grocer. The construction was almost complete before Covid hit, but Amazon shuttered the store during that time. As a result there was no groceries in my neighborhood from 2018-2023.

Now it seems Amazon is going to leave us a grocery desert yet again.

They were piloting smart carts at the location. The cart scans your items so checking out you just push the cart through a scanner that weighs it. But this invention was like a microcosm of Amazon’s whole fuckup with groceries. The problem with the store wasn’t that I couldn’t check out fast enough, it’s that it was a shit grocery store. They had popular products but they were missing all the unpopular, low margin products you need to actually cook (baking powder, shortening, tomato paste, soy sauce…). They only hire non-union jobs program people at the registers because Amazon believes that cashier is a sub-human role.

The previous store had an owner who would wander the aisles and chat with customers. The new store has Europeans with clipboards who watch you as you shop.

They never made sense to have but I’m sure someone made a huge career and got lots of bonuses for this initiative
> On April 4, 2024, it was revealed that Amazon's "Just Walk Out" technology was supported by approximately 1,000 Indian workers who manually reviewed transactions. Despite claims of being fully automated through computer vision, a significant portion of transactions required this manual verification. ( https://en.wikipedia.org/wiki/Amazon_Go )

Wonder how much of this is due to economics since computer vision tech never reached the expected performance + outsourced workers got (relatively) much more expensive after COVID.

I left the following comment some months ago, duplicating it here:

[Disclaimer: Former Amazon employee and not involved with Go since 2016.]

I worked on the first iteration of Amazon Go in 2015/16 and can provide some context on the human oversight aspects.

The system incorporated human review in two primary capacities:

1. Low-confidence event resolution: A subset of customer interactions resulted in low-confidence classifications that were routed to human reviewers for verification. These events typically involved edge cases that were challenging for the automated systems to resolve definitively. The proportion of these events was expected to decrease over time as the models improved. This was my experience during my time with Go.

2. Training data generation: Human annotators played a significant role in labeling interactions for model training-- particularly when introducing new store fixtures or customer behaviors. For instance, when new equipment like coffee machines were added, the system would initially flag all related interactions for human annotation to build training datasets for those specific use cases. Of course, that results in a surge of humans needed for annotation while the data is collected.

Scaling from smaller grab-and-go formats to larger retail environments (Fresh, Whole Foods) would require expanded annotation efforts due to the increased complexity and variety of customer interactions in those settings.

This approach represents a fairly standard machine learning deployment pattern where human oversight serves both quality assurance and continuous improvement.

The news story is entertaining but it implies there was no working tech behind Amazon Go which just isn't true.

I’m skeptical of this scoop.

It’s reasonable to expect a system like Amazon’s to use human feedback in training, and to quote the article linked on Wikipedia:

> Amazon said that the India-based team only assisted in training the model [and validating] a small minority of shopping visits.

I wonder if they were doing the same thing for palm recognition
I always found the "Amazon 4-Star" name funny. Presumably when it was first pitched internally it was called "Amazon 5-Star", then they realised that meant they basically couldn't sell anything, since nothing popular gets a full 5 stars. So they changed it to "4-Star", which just sounds awkward, and lacks the suggestion of top-quality that "5-Star" would. Instead, it's like the "Amazon Not-too-bad" store. I was amazed that they actually went ahead with it.
Presumably when it was first pitched internally it was called "Amazon 5-Star", then they realised that meant they basically couldn't sell anything, since nothing popular gets a full 5 stars. So they changed it to "4-Star"

This would not be the first off-by-one error at Amazon.

(comment deleted)
Shame, shopping there felt like magic. I hope the technology is developed in future without having to rely on remote workers validating transactions. Definitely felt like the future of shopping
I quite liked the ones in London. I loved the experience of just walking in and out really quick.

I knew someone who worked at Amazon UK and they told me years ago they were doing very baldy and there was talk of them closing.

So I'm not surprised to hear this at all.

The headline in their corporate press release says "Amazon doubles down on online grocery delivery and Whole Foods Market expansion to reach more customers"

That's one way to spin things I guess.

Has anyone used their go stores? I'm curious how the experience felt from a consumer standpoint. Do you feel welcomed or more like a thief?

I remember WAY back in the day when Arby's implemented touch screen ordering (on CRTs!) and it was a very quirky process. An Arby's employee would sit behind the counter and stare at you while you spent 5 minutes poking a CRT display. Very slow and very impersonal. They discontinued them in a short period of time.

i didnt use their system, but the experience wasnt that great, it felt like a target grocery store in terms of product quality and selection. its a grocery store, but the regular grocery store is better.
The Amazon Go stores in San Francisco were weird. They always had no people shopping in them, which would make sense given the increased efficiency, but it amplified the "am I stealing?" vibe. And the cost of goods wasn't made any cheaper than comparable stores in SF despite the touted increased efficiency.
I don't live around any Amazon Fresh stores so I never saw them though I did see the technology in use at several airports (though I've never personally used it). IMO I think places like airports are the best place for something like this, people are usually in a rush so not having to wait in line to checkout is nice and you don't have to worry about security as much because everyone there is a ticketed passenger (only saw them post-security) and even if someone did try stealing they wouldn't get very far.
Hmm Amazon fresh was useless anyway. It was this weird niche of grocery delivery but for small urgent orders. I just don't have that need like ever, if I need a bottle of shampoo or a head of lettuce urgently I'll just go to the corner shop.

Edit: oh oops I see this is about physical fresh stores, we never had those in the first place. Here in Europe Amazon fresh is a weird service for quick small grocery orders. For the bigger ones they partner with a local supermarket ("dia" here in Spain). But I never do grocery delivery because I never make any plans, I just make my life up as I go along :)

But Amazon fresh here is expensive and still slow (2hrs) so really not good for anything.

Amazon go I'm not even sure what that is.

Amazon fresh in the Bay Area was equivalent to Safeway but for better price/quality
Fortunately my Amazon branded subcutaneous chip still works at Wholefoods.
These bastards drove out some nice stores near me (supposedly the lease ended and did not renew) and rebuilt the buildings in order to open an Amazon Fresh location. That Amazon Fresh store never opened. Now we have a giant empty storefront nobody uses.
The "1000 people in India watching cameras" reveal was the moment the magic died. Once you know the wizard is just a guy behind a curtain, you can't unsee it.

The interesting question isn't whether the tech was ready. It wasn't. The question is whether Amazon learned anything useful from the attempt.

Computer vision for retail checkout is a legitimate hard problem. Occlusion, similar-looking products, people changing their minds. I've worked on CV pipelines and the gap between "works in the demo" and "works at scale" is brutal.

My guess: they collected a ton of training data from those human reviewers. Whether they'll use it for a v2 or just write it off, who knows.

My wife will be heartbroken. We moved recently and she loves shopping at Amazon Fresh. (Though part of the reason was that it was never busy :)
Well, that article made me nervous for a second! I love my Amazon Fresh grocery delivery. I started using it during Covid, but could never go back. It's so nice having groceries feel automated instead of a semi-daily chore. I eat much healthier and the rationale for using DoorDash evaporated.

Absolutely zero interest in a physical version that lets me check-out easier, though. So, I can see why they're making this switch.

I really liked the local Amazon Fresh, until they discontinued "just walk out" and replaced it with those hellish smart carts. I scanned one item successfully with the cart, got completely stuck trying to get it to scan a second one, handed the cart back to the employee, and never went back.
I’m not surprised about Amazon Go but I’m surprised about Amazon Fresh.

They almost seemed like an extension of Whole Foods to a more mainstream suburban market, and I thought they had solid foot traffic.

I like Whole Foods because it feels warm and the food looks good. The Amazon stores felt like walking inside a vending machine and that is not how people want to buy dinner.
Curious thought - will they be shutting down other “just walk out” powered stuff like Hudson Nonstops in airports?

I also know some Amazon warehouses had an entire Just Walk Out powered concessions area in their breakroom for purchasing snacks in partnership with one of their canteen vendors.