Hacker News Reader

Comment by somenameforme | Hacker News Reader

Perhaps a silly question, but can somebody describe how they might lose $10 billion on Alexa in one year? I don't understand the math. That's enough to pay 33,000 employees $330k a year, or some recombination of such. I don't think the real numbers come even remotely close to that.

So it presumably has to be hardware costs, but Alexa is software, and they have vertical integration of all servers which presumably would drive voice processing costs to negligible levels as well (and that's assuming 0 on-device caching/learning ability). And this is all assuming that the gross revenue generated by Alexa is $0, which also certainly isn't true. So I don't see where the numbers are coming from?

View on HN

2020 - It took Amazon four years to populate the world with 100 million Alexa-powered devices. It took the company just one more year to more than double that number. The e-commerce titan announced Monday that there are now "hundreds of millions of Alexa-enabled devices" in customers' hands worldwide, a massive increase from the 100 million it announced last January. https://www.cnet.com/home/smart-home/amazon-sees-alexa-devic...

Annual sales ~100 million device, then losing $50/device would lead to a figure of $5 billion loss in a year. Loss of $50/device would include cost for running R&D, running factories, giving minimum guarantee to factories, and all other associated cost - marketing, customer acquisition, support etc..

So $10B loss in a year sounds in the realm of possibility.

Alexa-powered does not necessarily mean Alexa devices made by Amazon though.

For example: I own a multiroom speaker setup not made by Amazon that is Alexa-enabled, all speakers could run Alexa if I had set them up for it (which I won't, ever), and they're very likely counted on this metric. I can remember a multitude of other speakers and devices with an Alexa-enabled badge on their packaging that aren't made by Amazon.

It'd be good to have hardware figures for sales of Amazon's own devices.

Good point.

However, supporting all the 'partner devices' comes with associated cost in terms of engg' development and support (somewhat like what Android must be dealing with. Android division in Google would not be a low cost division I'm sure).

It surely does, I work at a place where we also have to deal with multiple partner integrations across multiple vendors and types of devices for our products, there's some overhead for it but not nearly as much as one would expect.

One would not be able to imagine when things operate at mega scale (FAANG level), with world conquering ambitions at Exec level, and particularly in consumer domain.

I was at MS when Windows was a 'thing'. Windows was purely about software, the complexity and support requirements scales in logarithmic scale as more and more OEMs gets involved with their unique hardware, business constraints and cultures.

I work for a pretty sizeable corporation, a household name you have heard of or use products from. I don't want to dox myself too much but I can say we are not MS level of support of OEMs (because that's a quite ridiculous bar to cross, haha) but still we do have quite a few integrations across multiple consumer device types.

Assuming you're referring to Sonos -- while they were once the best, they seem to just not work well anymore. I'm lucky if I can get it to function correctly. I've observed my brother's Chromecast whole-home audio which cost him a fraction of the price and has a ton more variety - and I'm thinking I may ditch my Sonos setup.

Building your own is also ridiculously simple and as cheap or expensive as you’re willing to spend on speakers and/or drivers.

My setup is from Audio Pro, a Swedish brand. Never really thought Sonos were good value for the price and sound quality to be honest... Not an audiophile but I always look for the best bang for the buck I'm spending on audio.

Strange. I recently purchased some Sonos products and haven’t had any issues whatsoever. They’ve been absolutely fantastic for me.

What issues are you experiencing? Wonder if it’s something I can keep a look out for.

My Sonos speakers has been kind of buggy but I also have a brand new google home and it is not any better. I am thinking the blame is on google assistant. Google assistant seems to do a much better job understanding me but is inconsistent with nearly identical input. Alexa seems to be worse but is at least consistent.

I switched my Sonos speaker over from Alexa to Google and it’s working well enough now.

Nice numbers but don't those Alexa enabled device bring in revenue offsetting the cost and limiting the loss? Or doesn't anyone have an Amazon premium account or buy anything with Alexa on Amazon?

You'd really want to show that people bought more stuff due to Alexa than they would have otherwise. Tapping out an order on a smart phone is already pretty damn convenient so that's actually a high bar to clear.

We have I think… 5? Echo devices in my house, including an echo dot Amazon just sold me for $1 because it was my “Primeaversary". I haven’t purchased a single thing through them.

"Hey Alexa, I need a new pan, mine is old and broken" - "I have the Pan Frying Pan Cooking Pan Cooking Children Family Happy Best Pan from Fhrgwdsst for $14.99, would you like me to order it for you?"

I think my favorite notification was a few months ago—my echo speaker turned yellow because there was a notification. My kids find this incredibly exciting, and told Alexa to play it. And then Alexa announced to us, in breathless terms, that bananas are a good source of potassium. It even said "this is B A N A N A S", as only a robot can.

Humor probably helps a lot with making people feel better around these devices. Are you using it for anything non-trivial?

Maybe, but it wasn't funny so much as dumb and pointless. And: no, I just use it as a speaker and for setting timers.

Very curious why you would wiretap your home with 5 always-on internet-connected speakers running opaque software for a tech giant.

I dunno, it's the sort of thing that would be fit for a robocop-like dystopia just a couple decades ago but is seemingly... "normal" now.

Because they're convenient. In my house, you'll hear - Alexa, when is the Bills game? Alexa, when is daylight savings. Alexa, play Christmas music. Alexa, what is the etymology of wanker? Alexa, is Chipotle still open? Alexa, how old is Patrick Maholmes? Alexa, what time is it in Kenya? Alexa, ad nauseum.

If Amazon wants to datamine the dickens out of that to sell me more socks and USB cables, I really don't care.

Actually Alexa gets most of those same requests in my house, too ;-) Anyway, it's at 12:30.

Isn't this the most trivial and minor of conveniences? Wouldn't it take roughly the same amount of time to pick up your phone and type that? Don't you feel the least bit uncomfortable that the trade-off for that is that a tech giant gets a 24/7 audio feed into your and your family's home?

Boggles the mind.

The expected value of harm caused by tech giant listening to my house = 0. Trivial and minor > 0.

Not everyone's paranoid about that kind of thing and not everyone actually values privacy that much

We leave most of them on "mute" most of the time. The one we don't, we use to play music, mostly.

There is no such thing as mute

Maybe. The cameras have a physical shutter that you can keep closed, but the mute buttons are always just an electronic button that you have to trust Amazon to respect. Still, I don't think I've seen anybody claim that Amazon is lying about it, and there's enough people on the internet—especially here—that monitor their traffic, and would have noticed if echo devices were sending recordings that they shouldn't have taken.

From the article you’re commenting on:

One internal document described the business model by saying, "We want to make money when people use our devices, not when they buy our devices."

That plan never really materialized, though. It's not like Alexa plays ad breaks after you use it, so the hope was that people would buy things on Amazon via their voice. Not many people want to trust an AI with spending their money or buying an item without seeing a picture or reading reviews. The report says that by year four of the Alexa experiment, "Alexa was getting a billion interactions a week, but most of those conversations were trivial commands to play music or ask about the weather." Those questions aren't monetizable.

Why would they need to spend billions on validating that model though? Wouldn't it be enough to have just a 1000 or even 100 devices to test it?

The article says that the hardware was generally sold at cost, though.

Doesn't that mean cost of materials/COGS though?

Which doesn't include the overheads (engineering and staff, development and maintenance of back-end services, legal and compliance and all the rest) that you would normally plan to cover with the gross profit on the device.

It's a good question what exactly they meant by 'at cost', but it's at least going to be the plain hardware cost (plus maybe shipping), I suppose. And the point of the parent and GP was that the overheads shouldn't be that significant either, so it is still a bit of a mystery where exactly they spent that much money.

Even if the hardware is sold at cost, you still have to pay for running the service.

article said sell "at cost" not "at a loss"

I think the Echo devices are included, so it means hardware sold. Not sure they were being sold at a profit.

Also, you need to count the server costs, including machine learning training and inference.

Also, people are expensive, and it takes them to design and program this stuff.

For a while Amazon loved selling the smaller units for a little as a dollar (and often including several months of Amazon music for free or almost free as well).

Sell enough of these 30 dollar gadgets for a dollar and it will add up quickly.

The Echo devices were sold for around BOM, but IIRC they were able to get some significant price reductions on that on a few popular devices, though I'm not sure if that ever got released. I think some of the newer devices might have turned a profit (not including R&D costs though).

> This year, an employee familiar with the hardware team said, the company is on pace to lose about $10 billion on Alexa and other devices.

Creative journalism or creative accounting.

Or both ;)

Creative journalism of creative accounting is more apt.

To me this looks like a deliberate leak from Amazon. Why? I don't know. Perhaps a signal to wall street that more layoffs are coming to Alexa division?

> To me this looks like a deliberate leak from Amazon.

That was what I thought as well. Whether to sugarcoat propr or upcoming layoffs, this niewspiece sounds like a justification in the public's eye of why everyone at Amazon is getting the axe.

I don't understand the point about vertical integration. Amazon use their own hardware and services internally and AWS makes a ton of money of course.

But if you were accounting for the profit or loss of Alexa specifically, of course you would account for all the AWS services used at the price they are sold for externally.

Amazon is not a philanthropy. The company needs to make money and as much of it to satisfy shareholders (who, BTW, are never satisfied).

Having spent tens of billions on something that doesn't add up to the bottom line doesn't go well with shareholders, especially when there's no prospect that it ever will.

Hardware is very expensive. Alexa dot is $14 on Amazon.com right now, i bet they are selling it at a loss for that price.

I bet that’s at cost tbh. You’d be surprised how cheap things have gotten. However, r&d and every single Alexa request costs money. So from day one, it’s losing money.

How much are two hundred million microphones in earshot of private conversations worth?

This

One Zuckerberg.

Demonstrably very little, at least with regard to $€¥.

Maybe right now it's value is $0.

Think bigger picture. Imagine a panopticon spanning Ring.com camera networks and voice recordings telling you what's going on inside, with AI providing automated summaries of recent conversations. All delivered end-to-end by Amazon!

Apparently -$10,000,000,000.00.

While I believe that this is actually a simmering threat to society, I have a lot of trust that the current slew of devices are not eavesdropping on you. I believe smartphones to be a much bigger threat.

I’m more worried about smart phones acclimating people go carrying tracking and microphones. But, we find those tools indispensable now, so instead, we seem to want to focus on smart speakers

It’s got to be at least $5-7 to take my payment, pick, pack, and deliver it.

I can get a wifi-enabled microcontroller delivered from China for $3.40 [1] including payment, picking, packing and delivery. And the product is at least $1 of that price.

I find it hard to believe Amazon is spending more than $2 on payment and delivery.

[1] https://www.aliexpress.com/item/1005004563128378.html

Can you get it the next day/day after? Speed has a value and locality has a cost.

Amazon puts good components in their Echos.

https://asia.nikkei.com/Business/Technology/Amazon-Echo-tear...

China subsidizes mail to the US. Amazon has to pay for all their logistics without government support.

Perhaps a silly question, but can somebody describe how they might lose $10 billion on Alexa in one year?

Amazon seems to be a very poorly run company.

So many issues with search, with spam, with listings, and here's another data point...

Just bought a laptop. Waited until it was on sale, after doing Linux compatible research.

It shows up. 3 days later, it is even more on sale. This is a product sold and shipped by amazon.com, in both cases.

I contact support, and ask for a price drop refund, as there is a $150 price difference now. Nope, guy keeps linking me to "Price match" docs. I explain, I am not comparing vs another supplier, but amazon itself.

I am told, just return the laptop and order it again. That there is no other option.

For me, there is no return cost, other than a 5 minute detour to drop it off. Free return shipping.

For them, they lose the original shipping cost, the return cost, the processing of the return, and will sell it via amazon warehouse, for a $250 price drop.

I imagine that to sell a laptop via amazon warehouse, a check that it boots, a drive wipe is done, etc.

This is indicative of a company incapable of handling scale correctly.

Yet they blather on about loss on returns! And about return scams!

And yet, I still get the laptop at the price I want.

Only an insane person would act this way, and this insane act is a weird emegent behaviour at amazon.

It's telling, to me.

Amazon === borked

Amazon used to be happy to price match themselves within a month or so of a purchase, or do things like let keep a cheap item you wanted to return.

My guess is that they did some A/B testing and found that the "irrational" behavior actually saves them money. In your example, if 50% of people don't bother going through the return process they still come out ahead.

I bet this is actually it. I totally would ask for a price difference refund for a few bucks. But I’m not bothering with a return/repurchase for a couple dollars.

My guess is that they also might do market segmentation analysis to understand what type of customer you are, and this affects the return policy.

For example, you consistently buy small ticket household items and rarely return anything: when you do return something they will be helpful and generous to keep you happy.

You buy large amounts of fairly expensive clothes and shoes but return around 50% after trying on. They will accept the returns without issues but won't let you keep anything.

You're a bargain hunter who buys expensive laptops or hard drives whenever the discounts look really good. They will make returning stuff a little harder for you since they know you don't have loyalty, and they don't want to facilitate arbitrage.

Doing this is computationally no more sophisticated than the work they are already doing to detect professional scammers. They aren't supposed to show different customers different prices, but they can freely use things like this to do stealth price differentiation, since many details of how the return policy is implemented are at their discretion.

Somewhat famously, Best Buy did this and differentiated customers into "Angels and Devils"[0].

I've always assumed there's some sort of credit score system built into every retailer account, and am somewhat conscious to try to keep it positive.

[0]https://www.hbs.edu/faculty/Pages/item.aspx?num=32709

Hah, this is hilarious. I just went through this 30 or so minutes ago. I explained to the CS person that I would in fact be saving money and carbon. “Nope”, he said, just go ahead and repurchase and return.

Insanity. It’s tripping over dollars to save pennies.

It's a lack of democracy and autonomy in the workplace. This is what it looks like -- needless waste and bureaucracy.

Why do you assume that Amazon don't know what they're doing here?

Although you returned the laptop, the annoying process might have discouraged many others from doing so. The shipping cost and the cost of processing the return are likely much smaller than you expect when accounted for on Amazon's scale. And by making you return and re-order, they are putting the risk on you if the laptop is not the identical product, rather than taking on the costs of administering a price drop refund themselves.

Of all the things that Amazon do, analysing the game theory of customer behavior so that they can do as well as possible out of interactions like this seems like one of their biggest strengths. They certainly have thousands of skilled people working on it.

As for return scams, I would not take what they say at face value. My opinion is that they have quite sophisticated models for analysing returns, both for detection (to distinguish between the behavior professional scammers and regular customers) and economically (to decide that it's cheaper to give the latter the benefit of the doubt, let them keep returned items, etc). The only flaw is if otherwise legitimate customers start to think they can get away with falsely returning 1 in 20 items or whatever. So they 'blather on' about return scams, not because they are material to their business, but to make it clear that if you cross an invisible line, they will go from not caring about your $120 air fryer that you ended up with two of, to pursuing serious legal and criminal consequences.

Not sure I’m buying the argument that Amazon doesn’t know how to scale. I’ve had the exact same thing happen. Price drops after I’ve bought the product and have had to purchase the new one.

I’d say that Amazon has the data to determine if it’s more profitable to give the customer a credit or have them ship and return it.

So while annoying, I’d say that scale is probably why they are doing this seemingly unintuitive thing.

They probably have evidence that enough people just drop it instead of returning that this policy saves money. Amazon employs more economics Phds than anyone but the government so I would be surprised if they hadn't studied something as fundamental as the return policy.

How common of a use case do you think this is? How common do you think it needs to be for Amazon to invest in special logistics for this pattern (I.e. for it to be a net positive)?

Over billed from AWS to make that part of the balance sheet look good?

(entirely speculation)

From what I've heard, AWS prices for large customers like Netflix are a lot lower than for Alexa or Amazon Retail. Netflix has the option to switch to GCP or Azure; Alexa doesn't so they don't get those discounts.

What does it mean for Alexa to "pay" for AWS though? Is it recorder as revenue somewhere or does it just get accounted for internally?

Probably means Alexa loses $10 billion instead of $9.5 or whatever and has to fire some more people

MMAesawy gave the correct answer. It makes it clear what Alexa is really costing Amazon in an format that's easy to interpret. This is how most companies handle "internal purchases" from another business unit that also sells that product externally.

Consider it this way: if Alexa doesn't buy that capacity, someone else could, so it's important to capture this opportunity cost.

It's not perfect, however. I've seen examples where the repairs division of a company had to buy parts from its distribution branch at retail price. All well and good, so far. Except they then had to mark-up that part's price in the repair cost. Combine that with the fact distribution would give discounted rates to other repair companies to secure business and what you have is a repair division that's being outbid on price by competitors using the same parts!

Sure I was talking about what it means for alexa to not get the discount netflix gets. Obviously they need to account for the resources they use.

There are tax laws around this to you can’t cheat.

I remember when I toured a large local forklift company (Brand you’ve probably heard of - big plant) and on the main production floor they had just about every OTHER brand of forklift.

Someone asked and was told they had to pay full retail (including all taxes) on their own units, and they have to buy their competitors products for research… so, two birds one stone…

Even if it's Amazon paying Amazon, there is an opportunity cost to using all those cloud resources. Every resource used internally is a resource not available for sale externally.

One of the main selling points of the cloud is dynamic scaling which necessitates that Amazon have enough servers for some multiple of their customers’ base load. As long as internal Amazon resources are given a lower priority and booted in favor of customers during leak load, the opportunity cost is basically zero leaving just the marginal cost of electricity and hardware maintenance.

I am fairly certain this does not happen. An internal AWS customer account can have numerous flags associated with it but "boot me out first" isn't one of them (aside from spot instances that everyone has access to).

Source: work at AWS

That honestly surprises me. Having worked at a small subsidiary of a different big tech giant our costs for the big tech's cloud were charged at cost price at the behest of the accountants.

No actual knowledge of the details, but the story I’ve heard at Microsoft is similar to the AWS one: internal users of Azure pay the same (based on volume and service level) as an external customer would. Supposedly their capacity constrained so any extra compute being used internally can’t be sold externally.

I work in Retail and I’m pretty sure this isn’t true. I don’t know what AWS charges Netflix but there is an internal rate card and I’m pretty sure it’s just based on what it costs to run the machines.

It is a surprising number. I assume that it also includes the development and manufacturing of the devices that they sell that use Alexa. They clearly sell those devices at close to cost. Or below?

Reply on news.ycombinator.com