Amazon Unbound

by Brad Stone

  As in the Doppler project, the deadlines Bezos gave were unrealistic, and to try to make them, the team hired more engineers. But throwing more engineers at failing technology projects only makes them fail more spectacularly. Kindle was strategically important to Amazon at the time, so instead of poaching employees internally, the Tyto group had to look outside, to hardware engineers at other companies like Motorola, Apple, and Sony. Naturally, they didn’t tell anyone what they’d be working on until their first day. “If you had a good reputation in the tech industry, they found you,” said a Fire Phone manager.

  The launch was perpetually six months away. The project dragged on as they tried to get the 3D display to work. The original top-of-the-line components quickly became outdated, so they decided to reboot the project with an upgraded processor and cameras. It was given a new owl-themed code name, Duke. The group started and then canceled another phone project, a basic low-cost handset to be made by HTC and code-named Otus, which would also use the Amazon flavor of the Android operating system that ran on Amazon’s new Fire tablets, which were showing promise as an economical alternative to Apple’s iPad.

  Employees on the project were disappointed when Otus was scrapped because they quietly believed Amazon’s opportunity was not in a fancy 3D display but in disrupting the market with a free or low-priced smartphone. Morale on the team started to sour. One group was so dubious about the entire project that they covertly bought a set of military dogtags that read “disagree and commit,” after Amazon leadership principle #13, which says employees who disagree with a decision must put aside their doubts and work to support it.

  In his annual letter to shareholders released in April 2014, Bezos wrote, “Inventing is messy, and over time, it’s certain that we’ll fail at some big bets too.” The observation was curiously prophetic. The team was preparing to launch the phone at a big event that summer. Bezos’s wife, MacKenzie, showed up for rehearsals to offer support and advice.

  On June 18, 2014, Bezos unveiled the Fire Phone at a Seattle event space called Fremont Studios, where he attempted to summon some of the charismatic magic of the late Steve Jobs and waxed enthusiastic about the device’s 3D display and gesture tracking. “I actually think he was a believer,” said Craig Berman, an Amazon PR vice president at the time. “I really do. If he wasn’t, he certainly wasn’t going to signal it to the team.”

  Reviews of the phone were scathing. The smartphone market had shifted and matured during the painful four years that the Fire Phone was in development, and what had started as an attempt at creating a novel product now seemed strangely out of touch with customer expectations. Because it did not run Google’s authorized version of Android, it did not have popular apps such as Gmail and YouTube. While it was cheaper than the forthcoming iPhone 6, it was more expensive than the multitude of low-cost, no-frills handsets made by Asian manufacturers, which were heavily subsidized at the time by the wireless carriers in exchange for two-year contracts.

  “There was a lot of differentiation, but in the end, customers didn’t care about it,” said Ian Freed, the vice president in charge of the project. “I made a mistake and Jeff made a mistake. We didn’t align the Fire Phone’s value proposition with the Amazon brand, which is great value.” Freed said that Bezos told him afterward, “You can’t, for one minute, feel bad about the Fire Phone. Promise me you won’t lose a minute of sleep.”

  Later that summer, workers in one of Amazon’s fulfillment centers in Phoenix noticed thousands of unsold Fire Phones sitting untouched on massive wooden pallets. In October, the company wrote down $170 million in inventory and canceled the project, acknowledging one of its most expensive failures. “It failed for all the reasons we all said it was going to fail—that’s the crazy thing about it,” said Isaac Noble, one of the early software engineers who had been dubious from the start.

  Ironically, the Fire Phone fiasco augured well for Doppler. Without a smartphone market share to protect, Amazon could pioneer the new category of smart speakers with unencumbered ambition. Many of the displaced engineers who weren’t immediately snapped up by Google and Apple were given a few weeks to find new jobs at Amazon; some went to Doppler, or to a new hit product, Fire TV. Most importantly, Bezos didn’t penalize Ian Freed and other Fire Phone managers, sending a strong message inside Amazon that taking risks was rewarded—especially if the entire debacle was primarily his own fault.

  On the other hand, it revealed a worrisome fact about life inside Amazon. Many employees who worked on the Fire Phone had serious doubts about it, but no one, it seemed, had been brave or clever enough to take a stand and win an argument with their obstinate leader.

  * * *

  After Jeff Bezos walked out on them, the Doppler executives working on the Alexa prototype retreated with their wounded pride to a nearby conference room and reconsidered their solution to the data paradox. Their boss was right. Internal testing with Amazon employees was too limited and they would need to massively expand the Alexa beta while somehow still keeping it a secret from the outside world.

  The resulting program, conceived by Rohit Prasad and speech scientist Janet Slifka over a few days in the spring of 2013, and quickly approved by Greg Hart, would put the Doppler program on steroids and answer a question that later vexed speech experts—how did Amazon come out of nowhere to leapfrog Google and Apple in the race to build a speech-enabled virtual assistant?

  Internally the program was called AMPED. Amazon contracted with an Australian data collection firm, Appen, and went on the road with Alexa, in disguise. Appen rented homes and apartments, initially in Boston, and then Amazon littered several rooms with all kinds of “decoy” devices: pedestal microphones, Xbox gaming consoles, televisions, and tablets. There were also some twenty Alexa devices planted around the rooms at different heights, each shrouded in an acoustic fabric that hid them from view but allowed sound to pass through. Appen then contracted with a temp agency, and a stream of contract workers filtered through the properties, eight hours a day, six days a week, reading scripts from an iPad with canned lines and open-ended requests like “ask to play your favorite tune” and “ask anything you’d like an assistant to do.”

  The speakers were turned off, so the Alexas didn’t make a peep, but the seven microphones on each device captured everything and streamed the audio to Amazon’s servers. Then another army of workers manually reviewed the recordings and annotated the transcripts, classifying queries that might stump a machine, like “turn on Hunger Games,” as a request to play the Jennifer Lawrence film, so that the next time, Alexa would know.

  The Boston test showed promise, so Amazon expanded the program, renting more homes and apartments in Seattle and ten other cities over the next six months to capture the voices and speech patterns of thousands more paid volunteers. It was a mushroom-cloud explosion of data about device placement, acoustic environments, background noise, regional accents, and all the gloriously random ways a human being might phrase a simple request to hear the weather, for example, or play a Justin Timberlake hit.

  The daylong flood of random people into homes and apartments repeatedly provoked suspicious neighbors to call the police. In one instance, a resident of a Boston condo complex suspected a drug-dealing or prostitution ring was next door and called the cops, who asked to enter the apartment. The nervous staff gave them an elusive explanation and a tour and afterward hastily shut down the site. Occasionally, temp workers would show up, consider the bizarre script and vagueness of the entire affair, and simply refuse to participate. One Amazon employee who was annotating transcripts later recalled hearing a temp worker interrupt a session and whisper to whoever he suspected was listening: “This is so dumb. The company behind this should be embarrassed!”

  But Amazon was anything but embarrassed. By 2014, it had increased its store of speech data by a factor of ten thousand and largely closed the data gap with rivals like Apple and Google. Bezos was giddy. Hart hadn’t asked for his approval of the AMPED p
roject, but a few weeks before the program began, he updated Bezos with a six-page document that described it and its multimillion-dollar cost. A huge grin spread over Bezos’s face as he read, and all signs of past peevishness were gone. “Now I know you are serious about it! What are we going to do next?”

  What came next was Doppler’s long-awaited launch. Working eighty to ninety hours a week, employees were missing whole chunks of their family’s lives, and Bezos wasn’t letting up. He wanted to see everything and made impetuous new demands. On an unusually clear Seattle day, with the setting sun streaming through his conference room window, for example, Bezos noticed that the ring’s light was not popping brightly enough, so he ordered a complete redo. Almost alone, he argued for a feature called Voice Cast, which linked an Alexa device to a nearby Fire tablet, so that queries showed up as placards on the tablet’s screen. When engineers tried to quietly drop the feature, he noticed and told the team they were not launching without it. (Few customers ended up using it.)

  But he was also right about many things. As launch neared, one faction of employees worried that the device wasn’t good enough at hearing commands amid loud music or cross talk and lobbied to include a remote control, like the one the company made for the Fire TV. Bezos was opposed to it but agreed to ship remotes with the first batch of speakers to see if customers would use them. (They didn’t, and the remote disappeared.)

  He also headed off a near disaster when it came to what to actually call the device. For four years there had been no consensus on that topic. The team debated endlessly whether there should be one name or two for the virtual assistant and the hardware. After opting for separate names, they cycled through various options for the speaker and settled on… the Amazon Flash. The news updates would be called “Flash briefings,” and packaging with the Flash brand printed on it was ready to ship.

  But then less than a month before the introduction, Bezos said in a meeting, “I think we can do better.” Searching for a replacement, they opted to pilfer the name of an Alexa feature, Echo, which allowed a customer to ask Alexa to repeat a word or phrase. (The command was then changed to “Simon says.”) There wasn’t enough time to print new boxes or user manuals, so the Echo’s earliest buyers ended up receiving plain black boxes. Toni Reid, a director Hart hired to launch the product, had to write the user manual without ever actually naming the product. “That’s a skill everyone should have,” she said.

  The introduction of the Amazon Echo on November 6, 2014, was molded by the failure of the Fire Phone only months before it. There was no press conference or visionary speech by Bezos—he was seemingly done forever with his halfhearted impression of the late Steve Jobs, who had unveiled new products with such verve. Instead, Bezos appeared more comfortable with a new, understated approach: the team announced the Echo with a press release and two-minute explanatory video on YouTube that showed a family cheerfully talking to Alexa. Amazon execs did not tout the new device as a fully conversational computer, but carefully highlighted several domains where they were confident it was useful, such as delivering the news and weather, setting timers, creating shopping lists, and playing music.

  Then they asked customers to join a waiting list to buy an Echo and reviewed the list carefully, considering factors like whether applicants were users of Amazon Music and owned a Kindle. Recognizing that it was an untested market, they also ordered an initial batch of only eighty thousand devices, compared to a preliminary order of more than three hundred thousand Fire Phones, and distributed them gradually over the next few months. “The Fire Phone certainly made folks a little cautious,” said Greg Hart. “It led us to revisit everything.”

  After four years of development, more than one Doppler veteran suspected that the Amazon Echo might leave another smoking crater in the consumer technology landscape, right next to the Fire Phone’s. On launch day, they huddled over their laptops in a “war room” from their new offices in the Prime building, a few minutes’ walk from Fiona, to watch as the waiting list swelled past even their most hyperbolic projections.

  In the midst of the vigil, someone realized they were letting a significant accomplishment slide by unappreciated. “It was our launch moment and we weren’t ready for it,” said Al Lindsay. So, a hundred or so employees headed to a nearby bar for a long-awaited celebration, and a few of the weary executives and engineers on the project closed it down that night.

  * * *

  Over the next few weeks, a hundred and nine thousand customers registered for the waiting list to receive an Echo. Along with some natural skepticism, positive reviews rolled in, with quotes like “I just spoke to the future and it listened,” and “it’s the most innovative device Amazon’s made in years.” Employees emailed Alexa executives Toni Reid and Greg Hart, pleading for devices for family members and friends.

  After the Echo shipped, the team could see when the devices were turned on and that people were actually using them. Bezos’s intuition had been right: there was something vaguely magical in summoning a computer in your home without touching the glass of a smartphone, something valuable in having a responsive speaker that could play music, respond to practical requests (“how many cups are there in a quart?”), and even banter with playful ones (“Alexa, are you married?”).

  Many Doppler employees had expected they could now catch their breath and enjoy all their accrued vacation time. But that is not what happened. Instead of stumbling ashore from choppy seas to rest, another giant wave crashed over their heads. Bezos deployed his playbook for experiments that produced promising sparks: he poured gasoline on them. “We had a running success on our hands and that’s where my life changed,” said Rohit Prasad, who would be promoted to vice president and eventually join the vaunted Amazon leadership committee, the S-team. “I knew the playbook to the launch of Alexa and Echo. The playbook for the next five years, I didn’t have.”

  Over the next few months, Amazon would roll out the Alexa Skills Kit, which allowed other companies to build voice-enabled apps for the Echo, and Alexa Voice Service, which let the makers of products like lightbulbs and alarm clocks integrate Alexa into their own devices. Bezos also told Greg Hart that the team needed to release new features with a weekly cadence, and that since there was no way to signal the updates, Amazon should email customers every week to alert them to the new features their devices offered.

  Bezos’s wish list became the product plan—he wanted Alexa to be everywhere, doing everything, all at once. Services that had originally been pushed to the wayside in the scramble to launch, like shopping on Alexa, now became urgent priorities. Bezos ordered up a smaller, cheaper version of the Echo, the hockey puck–sized Echo Dot, as well as a portable version with batteries, the Amazon Tap. Commenting on the race to build a virtual assistant and smart speaker, Bezos said, “Amazon’s going to be fine if someone comes along and overtakes us,” as part of the annual late-summer OP1 series of planning meetings, the year after Alexa’s introduction. “But wouldn’t it be incredibly annoying if we can’t be the leader in creating this?”

  Life inside the Prime building, and in the gradually increasing number of offices around South Lake Union inhabited by the Alexa team, became even busier. Many of the new features would be rushed out the door, so that Amazon could start gathering customer feedback. Silicon Valley startups call this style of product development “minimum viable product,” or MVP. At Amazon, Jeff Wilke had popularized the idea of calling it “minimum lovable product,” or MLP, asking, “What would we be proud to take to the market?” It didn’t seem to matter that many Alexa features, such as the voice calling, were initially half-baked and rarely used. Over the course of the 2015 holiday season, Amazon sold a million Echo devices.

  Alexa’s division-wide motto became “Get Big Fast,” the same slogan used in the early years for Amazon. History was repeating itself. An organization of a few hundred employees swelled to a thousand in the first year after the launch, and then, incredibly, to ten thousand over the
next five years. Through it all, like a crazed pyromaniac, Bezos kept spraying lighter fluid on the fire, promoting Alexa by paying an estimated $10 million for Amazon’s first ever Super Bowl ad in January 2016, starring Alec Baldwin, Missy Elliott, and former Dolphins quarterback Dan Marino.

  Despite all this attention, there was a sense inside Amazon that the Alexa organization was not moving fast enough. Greg Hart, who had produced the device out of nothing more than a Bezos email and whiteboard drawing, left the division and moved over to help run Prime Video. “The thing that I got up every day loving doing was the creation of Alexa,” he said wistfully years later. But with the Alexa group growing fast, “it was probably a better fit for another leader.”

  In his place came a longtime Bezos favorite, Mike George, a bald, charismatic, cowboy boot–wearing Amazonian with a penchant for face paint, who liked to walk into meetings with an Amazon Tap under his arm, blasting music.

  Mike George had what Bezos called a “fungible” energy. Over the years, Bezos deployed him like a firefighter to douse the flames of chaos and instill order in divisions like human resources, marketplace, payments, and later, Bezos’s private philanthropy, Day 1 Academies Fund. Various colleagues endearingly referred to him as a “brute,” a “high school jock who never unjocked,” and “totally cleaved from Jeff’s rib.”

  Mike George ran Alexa for a year, but the impact is still broadly felt. The Alexa division couldn’t recruit fast enough to fulfill its hiring needs, so Amazon instituted a sort of company-wide draft, giving every new hire to other parts of Amazon—like AWS and retail—an alternate job offer to join the Alexa division instead. Unhappy Amazon managers suddenly lost sought-after engineers they thought they had recruited.