Internal docs show Amazon is struggling to retain Alexa device users; in some years between 2018-2021, 15%-25% of new users were no longer active after one week
Each holiday season since 2015, Amazon.com Inc. has counted on selling a lot of its Alexa voice-controlled smart speakers.
BloombergPriya Anand
Context & Ripple Effects
Amazon had already built a large Echo installed base, but earlier reporting found that only 2% of roughly 50 million Alexa owners had made a purchase through the assistant, pointing to weak commercial engagement rather than a simple device-sales problem. Amazon was also improving Alexa’s capabilities, including lowering its error rate and retaining conversational context.
The newly reported early-user inactivity makes the gap between device shipments and an active service audience more concrete. It also helps explain why Amazon’s low-priced-hardware strategy was later reported to have produced more than $25 billion in devices losses without delivering the anticipated downstream returns.
First-order effects
Amazon’s Alexa business has fewer active new users than its device-sales totals imply, shrinking the audience available for Alexa services, commerce, and repeat engagement.
The retention data puts greater weight on Amazon’s product improvements: better recognition and conversational context must translate into sustained use, not merely smoother one-off interactions.
Second-order effects
Low early retention compounds Alexa’s already limited shopping conversion, reported in 2018 as only 2% of owners making an Alexa purchase, weakening the case for subsidizing speakers to generate commerce revenue.
Amazon must judge Alexa hardware economics by active-device value rather than units shipped, concentrating pressure on the experiences that bring users back after setup.
Third-order effects
If device makers cannot turn ambient assistants into habitual use, smart-speaker competition shifts from installed-base claims toward revenue and engagement per active device.
The pattern challenges the hardware-at-cost model for voice assistants: distribution is valuable only when the assistant retains users long enough to support services or transactions.
The trend: Ambient AI is moving from a device-distribution race toward proving durable, monetizable engagement per active device.
Each holiday season since 2015, Amazon has counted on selling a lot of its Alexa-enabled smart speakers. For almost as long, it's known that the devices have had trouble holding customers' attention even into January, internal documents show https://www.bloomberg.com/...
this is the primary reason I've been switching my home over to HomePods and Siri control, even if that's inferior in a number of ways. it wasn't privacy concerns, it wasn't because of speed, it wasn't because of lack of features. it's because Alexa has gotten too damn annoying.
Alexa costs Amazon a ton of $$$. But now we have a better sense as to how much. The company projected the fixed costs alone for the division would hit $4.2 billion in 2021
Amazon has tried many things — even an Alexa-powered microwave — to get users to engage with the voice assistant in new ways. Still: Alexa users discover half the features they'll ever use within their first 3 hours of using it, per a 2019 document
Lots of interesting revelations about Jeff Bezos's old pet project. Last year, Amazon projected the smart speaker market — which it pioneered with Alexa devices — would only grow about 1% over the next several years, according to the documents
news: Amazon sells lots of Alexa devices during the holidays. For customers, the novelty fades fast. There have been years when 15% to 25% of new Alexa users were no longer active in their second week with the device, according to internal documents https://www.bloomberg.com/...?…
this would explain why Alexa incessantly asks “By the way, did you know i can do this?” after every simple voice request for the past year or so https://www.bloomberg.com/...
Money & cool technology is not enough: Amazon spends as much per year on Alexa R&D as the total R&D spending of Hong Kong and Saudi Arabia combined. So far, though, there is no killer app for voice AI; everyone is just setting timers & listening to music. https://www.bloomberg.co…
Reminds me of Bezos's warning that his company was incubating several multibillion $$$ failures, 'cause that was the appropriate scale for a business its size https://twitter.com/...