Voice Assistants are Unprofitable

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.


Interesting. I never thought about the costs of running this kind of service before. I must admit that I’ve been turning off voice assistants since 2014. Amazon is somewhat unique in selling their voice assistant as its sole product, and even at cost. Ambitious even by Amazon’s standards. Google does this with Google Home, from memory, though I’m not sure whether that product is sold at cost.

This got me thinking about the cost of running the service on Amazon’s servers, and Ars links to a relevant article at the bottom of that one. I think Apple tries to do most of this stuff client-side, which should cut costs heavily? I’m surprised Google does anything differently. Why would you need to reach out to a Google Assistant server, anyway?

Not particularly relevant to the topic at hand, but I found it interesting how the first entity Google threw under the bus was their “business partner”, which is the kind of behavior I expect more from Amazon:

The goal here is to not spend more money, so Google is apparently sacrificing partner devices to focus on the Pixel division. (Making your business a Google partner just seems like you’re asking for trouble, doesn’t it?)

As for users of voice assistants, I have to wonder what able-bodied people get out of it. At best, it seems slightly more convenient than tapping your phone a few times to achieve the same result. Maybe that’s because I’m the kind of person that thinks saying, “Hey Google, can you dim the lights?” is tacky. I’ve known people who have talked their books and messages into their phone with a voice assistant under the impression that it saves time, but I have to wonder whether it really does.


The best uses cases seem to be for necessarily hands-free cases. For example when driving or washing-pots. If you think of it as voice search, then it’s also best suited to “requests” with one obvious (not ten or more) “action”. For example “Play the Economist News”, “Call John Doe”. There’ll be plenty of business applications, and more useful utility in VR/AR. Probably gaming, but I’m not a gamer. Sure there’s the novelty aspect and pleasure of tinkering with technology, but I think I’ll stick with the humble electromechnical light switch/dimmer. And not just because of the creepy surveillance aspects of “smart” devices. Would it be rude to use the phrase “smart devices for gullible users”?

1 Like

Ah, the “while driving” scenario completely slipped my mind. Most modern cars use Carplay or Android Auto or something similar, but I don’t think Amazon is going to find their way into this market easily. Not unless the driver is a fan of Echo Glasses® or Echo Earbuds®. Using a voice assistant would be somewhat safer than navigating the console while driving, so that’s one great use case for this technology.

I don’t have much interest in VR or AR and expect I never will, but I’m curious what kinds of business applications it might have. In all of those scenarios, I can only see it as an auxiliary service; not the main product. It’s just so…basic.

I think I’ll stick with the humble electromechnical light switch/dimmer. And not just because of the creepy surveillance aspects of “smart” devices. Would it be rude to use the phrase “smart devices for gullible users”?

That seems to fit with the messaging of the companies that create these “smart” devices. The implication seems, to me, that “it’s so easy that even an idiot can use it!” As far as I can tell, the only thing “smart” means is that it has a Wi-Fi chip (and can probably only be controlled by a proprietary mobile app), which seems like just about the dumbest feature you could give an embedded device that rarely receives any updates at all. Or perhaps the device does receive updates, and could download the latest parts list to make sure you’re using a Genuine® Phillips Lightbulb.

Not that you need an internet connection to do any of that.

I might find “smart” devices more interesting if they didn’t have a well-detserved reputation for treating customers poorly. And if that same exploitative attitude wasn’t creeping into every industry, including cars.

Well…I’m sure you’ve heard it all before. IoT devices bad. And reproducing at an alarming rate. Voice assistants are hardly the worst offender here.

1 Like

We can look back at a time when only priests could read or write or at a time when a book cost as much as a house. I take it for granted that reading and writing is nearly universal. And it doesn’t cost much to write a letter or send an e-mail.

But maybe the guy who invented the alphabet thought it was a good idea to charge a licensing fee. And his ghost doesn’t agree with me at all.

I think from the perspective of future people looking back, communicating with machines using your voice is a given. And the idea that voice infrastructure would cost ten billion dollars or someone would want to charge for talking to the machines which surround you is ridiculous.

I can’t account for the cost or complexity of today’s system. But I think writing provides a good example. Talking to machines should be low cost and decentralized. And the value to civilization should be in the innovation that builds on top of voice assistants instead of rent-seeking.


Decent discussion on HackerNews including someone with similar thoughts “Driving or in the kitchen when cooking seem the be the most successful.”

Well said. As ever the press is overly focussed on writing about the big companies; here Amazon and Google. The breakthrough innovations with net positives for society and the economy are not going to come from them. Hell there’s probably innovators out there who are addressing your points; or there will be. In other words I agree with:

If anyone knows of such companies I’d for one would be grateful to learn about them.

1 Like

If anyone knows of such companies I’d for one would be grateful to learn about them.

I’m not quite sure it’s what you were looking for, but have you heard of Mycroft?

They’ve explored selling hardware with the Mycroft software.

I admit I haven’t looked into them much, but what I’ve read seems to place them in the same market. Additionally, there’s Mozilla’s Common Voice, which is a fascinating method of creating a large set of high-quality, openly available voice data that Mycroft uses. Maybe voice assistants will finally be able to understand me.

One of the unfortunate things about machine learning (from my perspective, at least), is that it is weighed heavily in favor of entities with lots of resources. Anyone with enough knowledge and time can write a good and well-used program; most software tools are free software. GnuPG was created and maintained almost entirely by one person, who never had many resources.

On the other hand, machine learning software relies on lots and lots and lots of data to be reliable. Big businesses have the ability to collect that amount of data. Highly-skilled individuals who hack on free software overwhelmingly don’t.

I think Common Voice is a really important project for the world @mike speaks of with low cost and decentralized voice recognition.