Over Labor Day weekend 1995, Omidyar cobbled together freeware he scooped up from around the Internet with his own homemade brew of applications. The do-it-yourself site, with its clunky blue-black letters over a drab, gray canvas, “had all the graphic charm of a Usenet newsgroup,” observed The Perfect Store author Adam Cohen, who published a history of eBay’s early years. Having never attended an auction, Omidyar didn’t know what people might sell online, so he based categories on sheer guesswork: antiques, automotive, books and comics, computer hardware and software, consumer electronics, and the like. “The computer code Omidyar wrote let users do only three things: list items, view items, and place bids,” Cohen recounted. “The name he chose was as utilitarian as the site itself: AuctionWeb.”
Omidyar tacked it onto his existing hobby site, which also had a page devoted to the gruesome Ebola virus, he hosted on eBay.com, short for Echo Bay; he had plucked the name out of thin air because he liked the sound of it. When he learned Echobay.com was taken, he registered what he considered the next best thing. Over the course of several weeks, he publicized AuctionWeb on Usenet groups and in the what’s-new sections of websites (“The most fun buying and selling on the Web,” he boasted.). He listed a quirky array of items for sale and their respective bids, which included a pair of autographed Marky Mark underwear ($400), a used Superman metal lunchbox ($22), a Nintendo PowerGlove ($20), and a 1989 Toyota Tercel with 64,000 miles ($3,200). As traffic increased, so did the number of auctions, and word-of-mouth satisfaction spread. A week after posting his first list, Omidyar noted a 66 percent increase in the number of items for sale, including a 35,000-square-foot warehouse in Idaho (floor bid: $325,000).
It was so encouraging Omidyar ginned up an informal test. He had purchased a laser pointer for demonstrations that he ended up using to entertain his cat, which gleefully chased the red dot until it went kaput in a couple of weeks. Instead of tossing it in the trash, he posted it on AuctionWeb with the model number, a description of its failings, and an opening bid of $1. A little over a week later someone bid $3. Then $4. By the time the two-week auction ended, the broken laser pointer sold for $14.83 and Omidyar realized he might be on to something.
What he couldn’t have known was that his little free-market experiment would soon prove so popular it would outstrip technology’s ability to keep pace. Even if he had, there was little he could have done about it.
[ THE SCALING CONUNDRUM ]
Scaling is a massive challenge that can bedevil any viral loop business. Jeremy Liew of Lightspeed Venture Partners points out that if your viral coefficient is 1—not more than 1, exactly 1, the minimum needed to grow virally—and you have 100 users, the next day you’ll have twice as many. On the third day there will be 400 users and on day four, 800. On the fifth day you will have a higher order magnitude of users than you did when you started, and two days means another order of magnitude. Multiply that by days, months, and years and the result is millions of users doubling at an ever-increasing rate. To design and build a system that can scale to multiple orders of magnitude is no trivial engineering accomplishment. Transferred to the material world, what would the city of Los Angeles do if suddenly the number of cars doubled in the next six months, then doubled again six months after that?
An online business that is scaling fast is often faced with the classic chicken-and-egg dilemma. To implement a scalable system before you launch “is frankly a complete waste of time if you don’t know if anyone will show up at all,” Liew says. People are “doing these lightweight development cycles not planned to scale out that far and that’s what causes these scalability problems. Others are building for scalability, but it takes them so freaking long because it’s not an easy problem and a lot of times that effort is wasted because they never do get viral.” Simply put, it doesn’t make sense to plan for massive scaling unless you actually start to massively scale. For his part, Omidyar hadn’t given it a thought. He was too busy getting his business off the ground. But there were signs from the beginning.
Five months after he unveiled AuctionWeb, his hosting company, complaining the site was bogging down its entire network, raised his monthly fees from $30 to $250. Omidyar had no choice but to charge for the service, even if it meant an end to his free-market experiment. Wisely, he let sellers continue to post items for free. If, and only if, it concluded with a sale, AuctionWeb would skim 5 percent for items priced under $25 and half that for anything over. He had no way of enforcing payment, but was well aware of the concept of shareware, in which engineers posted software applications on the Web and requested users to send whatever they thought it worth. An inveterate optimist, Omidyar was fine depending on the kindness of strangers.
Soon after, envelopes brimming with cash and checks arrived at his Bay area home. In his first month he took in enough in dimes, quarters, and dollars to cover the $250 hosting fee. “That put his fledgling little website in a category almost all by itself,” Cohen noted. “It was one of the very few Internet companies to be profitable from its first month of operation.” A month later Omidyar collected $1,000; in April he received $2,500. May’s tally was $5,000, and Omidyar, who didn’t have time to open all the envelopes, made his first part-time hire. May’s revenues doubled again to $10,000, and Omidyar quit his day job to work full-time on his fledgling business.
Throughout history cities sprang up wherever there was trade, people from all over gravitating to wherever buying and selling took place. The more who traveled to these marketplaces, the more who would learn of its existence through word of mouth. AuctionWeb was simply replicating this phenomenon in a virtual space, attracting future power users like Jim Griffith, a middle-aged veteran auction attendee living in Vermont who, in May 1996, built inexpensive computers out of discarded parts that he sold for a small profit. He was hunting a hard-to-find memory chip from IBM when a friend emailed him a link to AuctionWeb. There, among a thousand other doodads and doohickeys for sale, Griffith found what he was looking for. The bid was up to $8. He got it for $10. Right then and there he became an addict.
He spent much of his time over the next few months on the site, bidding for computer parts, offering his own computers for sale and answering questions on the bulletin board, which Omidyar had set up as a way for the growing community to handle the thousands of niggling details that arose daily, everything from settling disputes to fielding technical queries and passing on tips. As a former rock and blues musician with what he describes as an “unquenchable need to perform in front of people,” Uncle Griff, with his oddball online persona, quickly developed a cult-like following, dispatching homespun wisdom under an endlessly patient online demeanor, “a cross-dressing bachelor dairy farmer” with a talent and desire to answer questions. When, battling depression, Griffith disappeared off the site for two weeks in the fall, Jeffrey Skoll, Omidyar’s first full-time employee, called to check in and offer Uncle Griff a job in customer support.
The site’s traffic continued to swell 20 to 30 percent a month, with mailbags stuffed with money arriving at Omidyar’s doorstep each morning. In October 1996, AuctionWeb held 28,000 auctions, and over the next four and a half months the number of items posted grew 350 percent, fueled by Beanie Babies, the era’s must-have collectible. By year’s end the site counted forty-one thousand registered users. Every time Omidyar ventured a projection he would find that his forecast had drastically low-balled it. Meanwhile other auction sites, hoping to ape eBay’s success, sprang up, including onSale, which touted millions in venture capital, a far glossier website, and $30 million in transactions. While an auction site, it pursued a significantly different business model. It took possession of the goods it sold and profited on the actual sale, as well as levying a transaction’s fee. This meant it had to budget for warehousing and shipping, unlike Omidyar’s virtual model, which simply took a sliver from bringing buyers and sellers together. His strategy also offered incentive to sellers to attract buyers, while onSale, though capitalizing on the au
ction mechanism, was more like an online retailer with a liquid price scheme.
OnSale never did manage to dent AuctionWeb’s torrential growth. Later neither could AOL, Yahoo, or more than a hundred other auction businesses. Omidyar had unleashed a viral loop. He had first-mover status, a large and growing pool of enthusiastic users and an insurmountable lead. Soon AuctionWeb’s market share hit 80 percent.
[ FASTER, PUSSYCAT. SCALE! SCALE! SCALE! ]
In January 1997, AuctionWeb held 200,000 online auctions compared to the 250,000 it had hosted the entire previous year. With the spike in traffic and usage, the site, which still chugged along on Omidyar’s original mishmash of code, reeled. New listings sometimes didn’t post for a full day; it could take a minute between the time a potential buyer clicked on an item and the description appearing. Asking users to list items for sale during off-peak hours did little to ameliorate the situation. Omidyar hired an engineer to re-architect the entire infrastructure and gave him a deadline of Labor Day. Then, desperate to drive away users before the whole site came crashing down, he instituted several measures that included requiring credit approval for people with less than stellar payment histories and limiting the number of items that could be offered for sale in a day.
* * *
EBAY’S EARLY YEARS
1996
41,000 registered users $7.2 million in gross merchandise value $350,000 in revenue Hosts 28,000 auctions Receives 100 emails a day Half a dozen employees
1997
341,000 registered users 115,000 listings a day 138,054 items for sale $4.3 million in revenue 371 product categories Receives 1,200 emails a day 41 employees
1998
2.1 million registered users $700 million in gross merchandise value $48 million in revenue 6.58 million items for sale 10-millionth item listed Receives 2,500 emails a day 138 employees
1999
10 million registered users $2.8 billion in gross merchandise value 1,628 product categories Receives 4,000 emails a day 641 employees
Source: eBay; Forbes; Adam Cohen, The Perfect Store: Inside eBay (Boston: Back Bay, 2003).
* * *
Yet nothing stanched the flow of users, although AuctionWeb’s business didn’t seem to be suffering. In fact, in June, when Omidyar tried to pull up the site during a meeting with venture capitalists, it was down. Nevertheless Benchmark Capital invested $3 million, which would sit forever in a bank gaining interest, since Omidyar never saw reason to spend it. Famously frugal, he was adamant about keeping expenses low. The business had only one phone line, which staff were instructed to ignore, and as part of new worker orientation, each employee picked out cheap, often used office furniture. AuctionWeb’s headquarters was a study in disorganization, with no receptionist and chairs and desks arrayed in no discernible pattern.
On September 1, 1997, AuctionWeb unveiled its revamped site, renamed eBay, running on an entirely new scalable architecture. Omidyar then turned his attention to hiring a chief executive officer to run eBay with the intent of taking it public. He settled on Meg Whitman, head of Hasbro’s Playskool division, who, he believed, with her down-to-earth manner, embodied the eBay ethos. (He didn’t hold it against her that when she was first contacted she had never heard of Omidyar’s company.) Whitman took the reins in March 1998, and six months later eBay went public, its shares nearly tripling in its first full day of trading, resulting in a market cap of $1.9 billion. By year’s end eBay’s $48 million in revenue was ten times greater than the previous year’s, which itself was thirteen times more than all of 1996.
All of this came at a price. EBay’s rebuilt architecture proved no match for the rapidly growing onslaught of users and the computing power that millions of auctions required. For the next year and a half brownouts and outages were frequent. Security was, at best, an afterthought, which was starkly illustrated on March 13, 1999, when a twenty-two-year-old college student who went by the handle MagicFX hacked eBay. He achieved “root” access on its computers, the same kind that the site’s own administrators enjoyed, which meant he could change prices or place fake ads, divert traffic to other sites, or even take down the entire network. When a reporter told him to back up his boasts with evidence, the hacker took down eBay’s home page for two minutes and replaced it with the message:
Proof by MagicFX that you can’t always trust people…not even huge companies. (who woulda known that?) It’s 9:30 PM…do you know who has YOUR credit card information?
After figuring out a simple password (eBay had not followed standard password protection schemes), the hacker copied Omidyar’s source code governing auctions and took over a Solaris server with a well-known exploit—eBay’s technicians had not kept up with the latest patches. From there, MagicFX modified the system’s software so he could intercept passwords and log in names. He bragged that he could actually watch everyone’s keystrokes and monitored email from users alerting the company to the hacked page.*
Embarrassing as it was, it wasn’t nearly as disruptive to eBay as the outages that struck the company on June 10, 1999, following two less severe ones in May lasting several hours each and a half dozen others running from ninety minutes to several hours in the months preceding. While an outage for any company is serious, it was far worse for eBay, which depended on its computer network to conduct business. Twelve hours after eBay’s site went dark, engineers still couldn’t figure out what had gone wrong and weren’t sure they could bring it back.
If they couldn’t, all of eBay’s data, the lifeblood of its business—its millions of registrations, credit card numbers, feedback, auctions that had been in progress when the machines went down—would be lost forever. The company might never recover.
[ IF YOU CAN’T SCALE, YOU FAIL ]
All viral loop businesses face quandaries as they spread, at times forced to invent whole new technologies and practices. But not every company is successful. Before MySpace and Facebook there was Friendster, founded in 2002 by a self-proclaimed “Jewish tech-geek poster boy” in his early thirties named Jonathan Abrams, who started the site because his girlfriend had “dumped” him and he wanted to “get laid.” Trolling dating sites felt creepy, and he wondered if there was a way to bring “real-life context” to cybercourting. His solution was to lift the user profile page idea from Match.com and add links to friends’ profiles, which he believed would encourage people to be as authentic online as they were off. Abrams coded the entire site himself in his apartment, plopped on his queen-size Posturepedic mattress while watching TV. The result was a site that would allow people to conceptualize their social relationships in a completely new way.
After raising $400,000 in start-up capital from a dozen investors, Abrams launched the site in March 2003 by simply inviting twenty friends, who invited others, and so on, until the site took on a life of its own. Every time a profile page displayed, the user’s connection to other users was mapped out within four degrees of separation. In some cases this skein of connections could result in a web of hundreds of thousands of people. “The effect was to give users a vivid sense of how they fit into their social groups as well as into the larger world,” wrote Max Chafkin in Inc. “Abrams, it seemed, had created a piece of software that could tell us who we were.”
Within three months Friendster had logged 532,000 visitors, according to Alexa, the rating service. The next month the number rose to 675,000 and Abrams raised $1 million in venture capital. By August, there were 962,000 visitors. Seven months after launch Friendster was one of the top 100 most popular English-language websites and college kids from around the globe were flocking to it. But this frenzied growth led to serious technical snafus. Mapping users’ social connections within four degrees of separation was doable when there were ten members, or a hundred, or a thousand, or even a million. But Abrams had in essence created two viral loops and his servers simply couldn’t keep pace. The first was the growing number of members, which was growing virally and the second was the four degrees of connections
, which were increasing even faster. “Because that feature is inherently not scalable, the scale gets exponentially worse as the number of users grows,” says Matt Cohler, a former Facebook executive who is a general partner at Benchmark Capital.
Abrams’s servers were required to perform trillions of calculations every time a user clicked, which required a terabyte of pricey RAM. As a result the site became buggy, with pages taking minutes to load at peak times. Frustrated users would email the company to complain but never received a response because Friendster had only a few overworked employees. Others left Friendster because they objected to policies Abrams had instituted governing acceptable language and the kinds of photos that could be posted. In August 2003, Abrams’s Friendster profile listed interests including wine, parties, friends, and painting. By late September his profile reported that he was interested in only one thing: sleep. As Friendster sank and rivals like MySpace rose, his investors ousted him. Today Friendster barely cracks the top 15 in social networks in the United States, although it is popular in Asia, especially Indonesia, Malaysia, the Philippines, and Singapore. “All they had to do was keep the damned servers up and running” and Friendster would have been unstoppable, Cohler says. “But there was this one feature they refused to take down,” the one that calculated distant connections, which showed how you were socially tied to everybody else. “If they had just made that one change, [Facebook] might not be here today.”
A more modernday example of a company that has had major hiccups along the way is Twitter, which at its core confronts a similar computation challenge to the one that brought Friendster to its knees. While a heavily trafficked site like Yahoo burns bandwidth by serving up millions of webpages and images, it faces a known and predictable scaling equation. Social networks engage in more complex issues of interactivity, but they need only to route messages to one user at a time, or at most to a defined group. But Twitter text messages simultaneously go out to hundreds of “followers” (Twitterese for contacts), while each of these hundreds of users are also transmitting messages to hundreds of others. This means that every additional Twitter user and every new connection yields an exponentially greater computational requirement. During peak times—during one of Steve Jobs’s keynote presentations at MacWorld, for example—Twitter suffered embarrassing outages. Some weeks it seemed Twitter was down as often as it was up. While the patience and loyalty of its legions of users has been sorely tested, its viral growth has continued unabated.
Viral Loop Page 16