The eleventh-hour evo­lu­tion of Ama­zon’s voice-con­trolled sur­prise hit, the Echo

▶ How Ama­zon’s voice-con­trolled speaker be­came a sur­prise hit ▶ “We want to be a large com­pany that’s also an in­ven­tion ma­chine”

Bloomberg Businessweek (North America) - - Con­tents - The bot­tom line Af­ter flop­ping with the Fire Phone, Ama­zon has sold more than 3 mil­lion of its voice-con­trolled speaker, the Echo.

In the fall of 2014, some of the men and women build­ing Ama­zon.com’s new voice- con­trolled smart speaker felt they needed to con­front Chief Ex­ec­u­tive Of­fi­cer Jeff Be­zos. The re­lease of the speaker was loom­ing, and things were fall­ing into place. The de­vice looked good, its voice recog­ni­tion soft­ware was im­prov­ing quickly, and the boxes it would ship in had been de­signed and as­sem­bled. But there was a lin­ger­ing is­sue with the de­vice’s name: the Ama­zon Flash. Many peo­ple who worked at Lab126, the hard­ware divi­sion, hated it, ac­cord­ing to two former em­ploy­ees. Be­zos was strongly in fa­vor.

There was an­other worry. A core fea­ture of the de­vice is a “wake word” that, when spo­ken, cues it to take voice com­mands. One of the two words be­ing con­sid­ered was “Alexa.” Be­zos thought the best word would be “Ama­zon.” This pre­sented a chal­lenge: The speak­ers would wake upon hear­ing Ama­zon ads on tele­vi­sion and, be­cause it con­nects to a Wi-fi net­work, could start buy­ing stuff from the In­ter­net.

Gen­er­ally, the en­gi­neers and prod­uct man­agers at Lab126 sti­fled their mis­giv­ings, in­stead con­cen­trat­ing on giv­ing the boss what they thought he wanted. But weeks be­fore the speaker was set to ship, the dis­si­dents con­fronted Be­zos. He agreed to the changes: The de­vice would be called the Echo, and its wake word would be “Alexa.”

In a gad­get land­scape dom­i­nated by rec­tan­gu­lar touch­screens, the Echo stands out—a screen­less cylin­der, just over 9 inches tall and 3¼ inches in di­am­e­ter. It plays mu­sic and an­swers ba­sic house­hold ques­tions like, “how many tea­spoons are in a cup?” (about 48). The only way to in­ter­act with the Echo is to talk to it. It’s al­ways lis­ten­ing for its wake word, which users can change to “Ama­zon” or “Echo” if they want.

When the first batch shipped to buy­ers in Novem­ber, Ama­zon’s crit­ics mocked it. Some called the Echo a use­less gim­mick; oth­ers pointed to it as ev­i­dence of Ama­zon’s Or­wellian ten­den­cies. Then some­thing weird hap­pened: Peo­ple de­cided they loved it. Ama­zon never re­leases data about how its prod­ucts sell, but Con­sumer In­tel­li­gence Re­search Part­ners is­sued a re­port on April 6 say­ing that the

com­pany has al­ready sold more than 3 mil­lion of the de­vices, 1 mil­lion of them dur­ing the 2015 hol­i­day sea­son. About 35,000 peo­ple have re­viewed the speaker on Ama­zon.com, giv­ing it an av­er­age 4.5 stars out of 5.

Dozens of in­de­pen­dent de­vel­op­ers are writ­ing apps that work with the speaker’s voice con­trols to do things like turn off lights or or­der a pizza. The Echo, which seemed like a su­per­flu­ous toy at first, now looks like a way for Ama­zon to be­come the de­fault choice for voice-con­trolled de­vices.

“We want to be a large com­pany that’s also an in­ven­tion ma­chine,” Be­zos wrote in a let­ter to in­vestors in April. The Echo is what hap­pens when Ama­zon achieves that goal. Be­zos de­clined an in­ter­view for this story. Ten cur­rent and former Ama­zon em­ploy­ees agreed to talk, mostly on the con­di­tion of anonymity, be­cause they weren’t au­tho­rized to do so by the com­pany.

Echo comes out of Lab126, which was cre­ated in 2004 to build the Kin­dle e-reader. The lab’s name refers to the al­pha­bet, with 1 rep­re­sent­ing the let­ter A and 26 rep­re­sent­ing Z. The Kin­dle was Project A. Project B was Ama­zon’s smart­phone. Lit­tle is known about Project C, but a re­view of Ama­zon pa­tent ap­pli­ca­tions sug­gests a de­vice that would dis­play aug­mented-re­al­ity im­ages that peo­ple could in­ter­act with. The Echo—project D—was an off­shoot of Project C. It started in 2011, and at one point sev­eral hun­dred em­ploy­ees from Seattle to Cam­bridge worked on it.

The Fire Phone, which was un­veiled in June 2014, was a colos­sal flop. After it failed, Ama­zon moved some of the peo­ple who worked on it to the Echo team. They brought with them dif­fer­ent ideas and vary­ing lev­els of en­thu­si­asm about the speaker, which grated on some who had been on the Echo since the be­gin­ning. It didn’t help that the stakes had got­ten higher, since the speaker had to re­deem Ama­zon’s rep­u­ta­tion and dis­pel doubts—in­side

and out­side—about whether Ama­zon could make de­sir­able high­end con­sumer gad­gets at all.

As orig­i­nally con­ceived, the Echo was sim­pler and cheaper. One per­son who worked on the project re­mem­bers that the com­pany ex­pected to man­u­fac­ture the de­vices for about $17 and sell them for $50. They now re­tail for $180. Be­zos had lots of ideas about the speaker’s main pur­pose. “Jeff had a vi­sion of full in­te­gra­tion into ev­ery part of the shop­ping ex­pe­ri­ence,” says one per­son who was at Lab126 at the time.

Ama­zon hired a hand­ful of peo­ple from the speech recog­ni­tion com­pany Nu­ance Com­mu­ni­ca­tions. It bought two voice re­sponse star­tups, Yap and Evi, to build a speech recog­ni­tion sys­tem that could match those cre­ated by Google and Ap­ple for their smart­phones.

Once Ama­zon’s en­gi­neers started build­ing the Echo, they re­al­ized it would need more pro­cess­ing power than they’d an­tic­i­pated. They swapped out the mi­cro­con­troller, the kind of

sim­ple com­puter used to run such de­vices as re­mote con­trols, with a mi­cro­pro­ces­sor, which could han­dle more com­plex tasks.

The Echo went through sev­eral key changes at the eleventh hour. The speaker had to emit and lis­ten for sound si­mul­ta­ne­ously, a chal­lenge that had pre­oc­cu­pied en­gi­neers through­out de­vel­op­ment. What if the mu­sic on the Echo was so loud it drowned out peo­ple’s voices? Early on, en­gi­neers cre­ated pro­to­types for smaller de­vices that looked like hockey pucks, to be placed around the house to lis­ten for com­mands when peo­ple strayed too far from the main speaker. That idea was later pushed aside to fo­cus on the main de­vice. It re­cently reemerged as the Echo Dot, which Ama­zon is sell­ing on a lim­ited ba­sis.

By the fall of 2014, there was still dis­agree­ment over whether the Echo’s hear­ing was good enough on its own. Be­zos and his top deputies were adamantly op­posed to re­ly­ing on any form of in­put other than voice con­trol. They saw it as cheat­ing. Some en­gi­neers dis­agreed, push­ing for a re­mote that peo­ple could speak into from any­where in the house. As it hap­pened, the com­pany had just made such a re­mote for the Fire TV, its stream­ing me­dia player. The two sides agreed to send the first batch of speak­ers with a re­mote in­cluded. They’d gather in­for­ma­tion about how of­ten peo­ple used it and tweak the prod­uct ac­cord­ingly. Ap­par­ently the fears were overblown. The peo­ple us­ing the Echo in their homes al­most never touched the re­mote, and it was qui­etly re­moved from the box.

Con­nect­ing the Echo to In­ter­neten­abled light­bulbs and ther­mostats made by other com­pa­nies wasn’t a fo­cus within Lab126 un­til late 2014. On a lark, an engi­neer had rigged the speaker to work as a voice con­troller for a stream­ing TV de­vice. “It was some­thing he grew to em­brace, ag­gres­sively,” one em­ployee says of Be­zos’s re­ac­tion. Ama­zon’s vi­sion for the Echo now re­lies heav­ily on the speaker serv­ing as a hub for the so-called smart home. Dave Limp, Ama­zon’s se­nior vice pres­i­dent for de­vices, jokes that it’s only a mat­ter of time be­fore some en­ter­pris­ing de­vel­oper writes a pro­gram to use the Echo’s voice con­trols to flush the toi­let.

With the Echo, Ama­zon has found a way to in­sert it­self into cus­tomer in­ter­ac­tions with other de­vices and ser­vices. Part of this is good tim­ing. The tech in­dus­try has been search­ing for the next big com­put­ing plat­form after mo­bile. In­vest­ing in some com­bi­na­tion of voice con­trol and ar­ti­fi­cial in­tel­li­gence was pru­dent, and no one else has quite fig­ured it out yet. Ap­ple, Google, and Mi­crosoft all have vir­tual as­sis­tants, but they were made to make smart­phones work bet­ter.

The Echo has more than 500 “skills,” or voice- con­trolled apps, that can check your bank bal­ance or make your child’s fa­vorite an­i­mal noises. The com­pany keeps an in­ter­nal list of cus­tomer sug­ges­tions for new con­trols, rank­ing each one ac­cord­ing to pop­u­lar­ity to de­cide the or­der in which it will pur­sue them.

The next big chal­lenge for Ama­zon is to be­gin ty­ing ser­vices to­gether in new ways, says Julie Ask, an an­a­lyst at For­rester Re­search. She says be­ing able to tell the Echo to call an Uber is fun, but in­cre­men­tal. “In five years, my Echo will say, ‘Hey, it’s about time to go to the air­port. Should I get you a car?’ And I’ll just say yes,” she says. “That’s the dif­fer­ence be­tween where we are to­day and where we want to be.”

�Joshua Brustein, with Spencer Soper

“Jeff had a vi­sion of full in­te­gra­tion into ev­ery part of the shop­ping ex­pe­ri­ence.” ——Former Lab126 em­ployee

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.