Amazon sold more than 500 million Alexa-enabled devices in eleven years. Voice landed in every kitchen, every bedside table, every car. And then Alexa stopped. The product won adoption at a scale almost nothing else has matched, and never figured out the second act. The story isn’t a failure of distribution. It’s a failure of imagination about what Alexa was actually for.
Section 01 What Alexa got right.
The first Echo, in 2014, created the smart speaker category. The pitch was simple: control your home and get help just by speaking. The product delivered on it immediately. Plug it in, say “Alexa,” get a response in two seconds. The “wow” moment was real, and it scaled to half a billion devices.
But the job underneath the wow wasn’t voice. Users didn’t show up because they wanted to talk to an AI. They showed up because they wanted less friction in daily life, and Alexa, more than any product before it, made the home respond without effort.
The real job was always reducing interruption.
Section 02 Where it stopped.
The arc of an Alexa device is consistent across households. The first month is exciting and magical. By the third month, usage settles into a small handful of habits: timers, music, lights, weather, shopping lists. By year two, it’s furniture.
Alexa won activation. It lost retention depth.
Magic, then utility, then limitation. Three phases, every device.
Usage stayed high; revenue didn’t. Users were comfortable completing low-value tasks and showed no path into higher-value ones. Meanwhile, the real competition wasn’t Google Assistant or Siri. It was the smartphone in everyone’s pocket, and behind it, ChatGPT, Claude, and Gemini, quietly eating the high-value cognitive work that Alexa was never positioned to do.
Alexa won the room. Others won the brain.
Section 03 The home intelligence opening.
The pivot isn’t getting smarter on commands. It’s not competing on commands at all. Alexa shouldn’t be measured by how fast it can set a timer. It should be measured by whether it can quietly solve a problem before the user notices it.
That means moving from a command surface to an intelligence layer. From “one action per request” to “end-to-end outcome.”
The PM lesson generalizes past Alexa. Adoption is not value. The most-used product in a category can still lose if the jobs it serves are the cheap ones.
A product doesn’t win because people use it often. It wins because people trust it with something that actually matters.