December 2015, a Friday, eight in the evening, and the main floor of Uber’s headquarters on Market Street in San Francisco is still full. The floor is two football fields of light wood, concrete and steel. Flags of every country hang from the ceiling, and the meeting rooms are named for cities, from Abu Dhabi by the entrance to Zurich at the far end. One room breaks the pattern. The War Room stands in the middle of the floor, glass and metal outside, dark wood within. It can’t be booked, it is sometimes guarded, and it never closes. Red clocks on its wall keep the time in Singapore, Dubai, London, New York and San Francisco, because somewhere in the world there is almost always an emergency. Tonight the emergency is at home. A dozen executives have gathered for the review of Uber’s biggest North American cities. Andrew Chen, in his first year at the company, is there for the team that recruits drivers, the scarcest thing in the business.
Travis Kalanick, the chief executive, sits nearest the screen in a grey T-shirt, jeans and red sneakers. When the first slide comes up, a grid of two dozen cities, he springs out of his chair and squints at the numbers. Why has surge pricing jumped in San Francisco, and even more in Los Angeles? Chen and a few others in the room have a theory. Lyft has stepped up its driver referrals, and drivers are switching over in droves. Holiday demand has left the West Coast short of cars. Riders wait, sometimes twenty minutes, then cancel and check the rival’s price. Drivers who have driven to a pickup for nothing grow annoyed and log off, and every driver who leaves makes the next wait longer. Kalanick floats a referral bonus of seven hundred and fifty dollars for each side, far beyond anything the room has discussed. Others hesitate. Uber is already burning about a billion dollars a year fighting Didi in China. He paces, and chops one hand into the other. “Our network is collapsing,” Kalanick says. Chen’s team promises to ship the new bonus before Monday. At almost ten o’clock, Chen walks the few blocks home to Hayes Valley.
Those meetings become his education. He watches Uber grow to a hundred million riders in more than eight hundred markets, a rocketship and a roller coaster at once. In 2018 he leaves to invest at Andreessen Horowitz, in calm offices near Stanford. The firm sees thousands of startups a year, and one phrase answers every hard question in their pitches. What stops a competitor? Network effects. Why will growth continue? Network effects. Chen has heard it countless times, and he realizes he can’t properly explain it either. So he spends three years on the question. He conducts more than a hundred interviews with the people who built products like Dropbox, Slack and Zoom, and digs back through chain letters and the telegraph. The name he gives the first and hardest stage comes from a car that won’t start on a freezing morning. An empty social network, a chat app without your colleagues, a marketplace whose listings sit for months: unless a new product gets past its cold start quickly, it dies.
What is a network effect, anyway? The classic answer is that a product gets more valuable as more people use it, and the classic example is the telephone. In 1908 America has fewer than five million phones for nearly ninety million people. In one of AT&T’s annual reports, Theodore Vail writes that a telephone with no connection at the other end “is not even a toy or a scientific instrument.” Chen draws a duality from that line. Every networked product is a product and a network together, and today the product is software and the network is people. Airbnb owns no rooms, and YouTube owns neither its creators nor their videos. The value lives in the connections. That matters more now than ever, Chen argues. When the iPhone’s app platform opened in 2008 with five hundred apps, home screens were nearly empty, and a new app’s only rival was boredom on the subway. A decade later there are millions, and the top charts barely move. Software is now easy to build and easy to copy. Bigger rivals, Chen writes, “are often able to copy the product, but find it difficult to capture the network.”
The popular idea of network effects took shape in the dot-com boom. In 1996 only twenty million people have internet access, mostly over dial-up, yet investors talk of winner-take-all and first-mover advantage. Behind the talk is Metcalfe’s Law. Robert Metcalfe, a pioneer of Ethernet, held that a network’s value grows as the square of the devices on it: double the nodes and you quadruple the value. It justified sky-high valuations. Chen calls it “a simple, academic model that fails the test of real-life messiness.” It says nothing about how a network starts with no one on it, whether its users are active, whether it has buyers and sellers, or what happens when it gets crowded. He finds a better model in a college ecology class at the University of Washington. Meerkats in southern Africa live in mobs of thirty or fifty. When one spots a predator, it stands on its hind legs and barks or whistles a warning that tells the mob what kind of danger is coming and how urgent it is. In the 1930s Warder Clyde Allee, at the University of Chicago, found that goldfish grow faster in groups. Social animals have a tipping point. Above it, a group protects itself and grows. Below it, members get picked off one by one, and the group dwindles toward nothing.
Collapse has a place, about an hour south of San Francisco. In the early 1900s industrial fishing comes to Monterey, and a whole street of canneries rises, called Cannery Row. At the peak the boats bring in hundreds of thousands of tons of sardines a year, and the town grows to tens of thousands of people. Chen tells what happens next.
One year in the 1950s, the sardines mysteriously disappeared. The town waited patiently until the next year for the fish to come back—but they didn’t. And the year after that, it was the same. And the same again, the year after. The sardines were gone.
— Andrew Chen, The Cold Start Problem, Harper Business, 2021, pp. 37–38
Chen goes on: the sardines were fished below their tipping point, and an app can fall the same way. An app that is empty every time you open it soon gets deleted. Years later at Uber, standing at a whiteboard, he sketches how the number of drivers changes a rider’s experience, and a familiar curve appears. With only a few dozen drivers in a city, the wait is thirty minutes and almost nobody books. Past the tipping point, a fifteen-minute wait becomes usable, then ten, then five. Then the curve flattens. Four minutes or two make no difference, and a car that arrives instantly is almost a nuisance, because you still need to find your keys. Allee’s threshold is a network’s tipping point, and his carrying capacity is its saturation. Humans are social animals too. As Chen writes, “Rather than hunting and mating, our networks help us with groceries and dating.”
From that curve Chen builds a road map in five stages, an S that droops at the end. It begins with the Cold Start Problem, where most new networks die. Early on, network effects run in reverse: with too few users, people don’t find who or what they came for, so they leave, and the network shrinks. Chen calls these anti-network effects. The cure is the atomic network, the smallest network that holds together and grows by itself. For Zoom that is two people; for Airbnb, hundreds of listings in one city. Next comes the Tipping Point, where each new network tips faster than the last, like a row of dominoes. Third is Escape Velocity, where the single network effect splits into three forces: one that brings users in, one that makes them stay, and one that makes the business pay. Fourth is the Ceiling, where a growing network pulls toward growth and toward breaking apart at once. Last comes the Moat, because every competitor in the category has network effects too. Chen closes with a line from his friend Naval Ravikant: “Humans are the networked species.”
The Cold Start Problem gets its fullest portrait in a failure. Tiny Speck has a star team that has already built and sold a company. It raises seventeen million dollars, hires forty-five people, and spends years on Glitch, a multiplayer game played in a web browser. Glitch is set inside the heads of eleven giants who have dreamed a world into being, and players wander around clicking things and learning skills like fox brushing and potion making. Stewart Butterfield, the chief executive, later gives the problem a number: about ninety-seven percent of the people who signed up were gone within five minutes. A multiplayer game is only fun when many people are playing, and Glitch never gets there. Nearly four years after founding, Tiny Speck gives up on it, with most of the money spent and most of the staff laid off. Yet the team leaves with something it never meant to build. Its engineers, scattered from San Francisco to Vancouver and beyond, have been talking all along over IRC, a text-only chat system from 1988 with no search and no history. So they built a layer on top that saved messages and photos and made them searchable. It never had a name. An early employee, Ali Rayl, calls it a frankentool.
Butterfield and his cofounders decide to build what is already in front of them. The tool gets a back end of its own and a run of code names: Linefeed for a few days, Honeycomb on others, Chatly.io and more. At last it becomes Slack, short for Searchable Log of All Conversation and Knowledge. There is no marketing team. Butterfield goes to friends at other startups, like Rdio, and sometimes it takes dozens of meetings to win one over. He signs up forty-five companies for the beta. Each customer team is a small network of its own, and even today Slack’s smallest team is three people. At Rdio it spreads from the front-end developers to all of engineering to the whole company of a hundred and twenty. At its debut in August 2013, eight thousand companies join the waitlist. Years later Slack has more than twenty million daily users and sells to Salesforce for twenty-six billion dollars. It took nearly four years to fail and two more to start again from zero. Chen calls it “the opposite of an overnight success.”
Slack is pointless until your colleagues are on it, just as Uber is useless without drivers, who won’t drive without riders. The startup myth skips it: a few founders in a loft launch, the curve shoots up, the press arrives. In real life there is a spike, then a trickle, then features added in a hurry, then the money runs out. So how many users are enough? Butterfield tells Chen that Slack works with two people but needs three to really work. Elsewhere he has said that a team which has exchanged two thousand messages has truly tried it. Ninety-three percent of those teams are still using it. For Zoom, Eric Yuan says, it takes two people, one who wants to call the other. Airbnb’s Nate Blecharczyk finds that a city takes off at about three hundred listings, a hundred of them reviewed. At Uber, William Barnes aims for fifteen to twenty cars online at once, and the Los Angeles launch, poured into West Hollywood, is notoriously expensive. Find the kink in your own curve, Chen says. Below it, a sparse network feels broken. Message someone on Slack who isn’t there, and as Chen puts it, “Your coworker will just wonder why you didn’t send them an email.”
The atomic network has a birthday: September 18, 1958, in Fresno, California. Bank of America wants a card that shoppers will carry and shops will take, which means signing up both at once. So it picks one town. Fresno has about a quarter of a million people, and forty-five percent of its families already bank there. On that single day, Joseph Williams’s team mails sixty thousand residents a BankAmericard. Nobody has applied. It just turns up in the post, already active, with a few hundred dollars of credit on it. Alex Rampell, Chen’s colleague, sees the brilliance in it: “On day 1, cardholders simply existed.” The bank skips giants like Sears and goes after small shops like Florsheim Shoes, and more than three hundred Fresno merchants sign up. Within three months the card reaches Modesto and Bakersfield. Thirteen months after the drop there are two million cards and twenty thousand merchants. Uber’s first atomic network isn’t San Francisco but the Caltrain station at five in the evening. Staff use a tool named after the game StarCraft to text nearby drivers: go to the train. As Chen puts it, “The next big thing will start out looking like it’s for a niche network.”
Every network has a hard side: the few who do most of the work and are hardest to win and keep. They are the creators on a social network, the sellers in a marketplace, the developers in an app store. Wikipedia shows how few they can be. It draws about five hundred million visitors a month, yet only about four thousand people make more than a hundred edits a month. One of them is Steven Pruitt, a government records officer who edits in the evenings and on weekends. By 2019 he has made nearly three million edits, written thirty-five thousand articles, and touched a third of all the articles in English. Chen asks how much he earns for this, and answers in a word: “Nothing.” Bradley Horowitz’s rule is that one percent create, ten percent take part, and everyone benefits. What keeps creators going is feedback: likes, comments, thanks. Chen’s test is a single question. “If a piece of content was created, and no one saw it, would the creator be disappointed?”
To win the hard side, solve its hardest problem. Online dating in the early 1990s works like the newspaper classifieds: browse profiles, send messages. At a bar, a suitor who sees a line waiting to talk to someone attractive might walk away. Online there is no line to see, so the most attractive people, especially women, are buried in messages, and everyone else hears nothing back. In 2012 comes Tinder. Its cofounder Sean Rad tells Chen, “The older dating sites made it feel like you were doing work, like you were inside the office.” Tinder has no forms. It shows photos, and you swipe, in five minutes while waiting in line. Everyone signs in with Facebook, so you see mutual friends, and you are never shown to your own. Men swipe right on about forty-five percent of profiles, women on about five, so women mostly match with men they chose. Marketplaces follow the same rule: bring the supply first. Sidecar’s founders, hunting for a legal model that lets ordinary people drive, find Homobiles, a San Francisco nonprofit run since 2010 by Lynn Breedlove. A hundred volunteer drivers answer text messages and take donations to give the city’s LGBTQ community safe rides. Sidecar copies the model almost exactly, donations, front seat and all, and an industry with more than a hundred billion dollars in gross revenue follows.
A killer product for a network is simple, often embarrassingly so. Over Mediterranean food in a hotel restaurant in San Jose, Eric Yuan tells Chen that people thought Zoom was a terrible idea when he started it. The investors who actually read his pitch deck hated it, because WebEx, GoToMeeting and Skype had already won. Zoom’s edge is a link you click and video that works. Chen calls the one feature that matters the “it works” feature. To set the price, Yuan studies Dropbox, which gives away two gigabytes so people have time to fill them. He makes meetings free for forty minutes. In the beta, Zoom is just a download button on a web page, and some people at a Stanford continuing studies program want to pay for it. Yuan doesn’t know what to charge, so around Christmas they give him a check for two thousand dollars. He still keeps a copy. When the pandemic comes, Zoom’s meeting participants leap from ten million at the end of 2019 to over three hundred million within months.
How do you know a network has started? It has a Magic Moment: you open the app and it is full, alive and connected. Its opposite is a ghost town. Paul Davison and Rohan Seth reach theirs after years of audio apps: Seth’s Phone-a-friend, then together Uncalendar and Talkshow, a podcast studio that asks too much of its hosts and sounds scripted. In 2020 they strip it down to a single room where you can just talk. Chen signs up early in that pandemic year, officially user number 104. Often the room holds only Paul, and some days no one at all. But Chen remembers what kept him coming back.
…there were little bursts of magic: sometimes I’d drop in and start having one of the most wonderful conversations with friends I hadn’t seen in months—after all, it was during the COVID pandemic year. Or I might just listen to others having a fascinating chat about robotics, Bitcoin, the history of technology, or some other nerdy topic. I was hooked.
— Andrew Chen, The Cold Start Problem, Harper Business, 2021, p. 112
Chen goes on to lead Andreessen Horowitz’s investment in Clubhouse when it has two employees, at a value of nearly a hundred million dollars. Inside a year, millions are joining every month. Its first network is the founders’ friends in tech. The next fifty thousand or so bring it into the mainstream, led by Black musicians, comedians and creators hosting shows from the middle of 2020. At Uber, the opposite of a magic moment has a name. A zero is opening the app to get a ride and finding no drivers anywhere. Uber tracks them city by city, and Chen urges every team to count its own. Even then the work isn’t finished. Enough people in IT on Slack doesn’t make it work for someone in marketing, or for the next office. The Cold Start Problem, Chen writes, “needs to be dealt with over and over.”
On buses and subways around the world, twentysomethings with headphones in move their thumbs the same way: swipe, swipe, swipe, then one the other way. Tinder now handles more than two billion swipes a day, and Chen calls it “what love looks like at scale.” Dating is a terrible market to tip: when it works, two happy users leave. Chen hears the backstory from Sean Rad over drinks at the Chateau Marmont. “It all started with a single party at USC,” Rad says. In 2012 the app is called Matchbox, and it has no swipe, only a green heart and a red X. Jonathan Badeen, the iOS developer, keeps a deck of cards on his desk to fiddle with between bouts of coding, and one day he adds a swipe because it feels good. Texting their own contacts brings in only about four hundred users. So the founders throw a birthday party for a popular friend of Justin Mateen’s younger brother, and bus the students to a luxurious house. There is one catch. A bouncer at the door checks that every guest has downloaded the app. Next morning they all wake up with Tinder on their phones, and a second chance at the people they never spoke to. It is, Rad says, five hundred of the right people. Then they throw another party, at another school. Downloads go from four thousand to fifteen thousand within a month, then five hundred thousand a month later. Each party is easier than the last, and that repeatability is the Tipping Point.
A curated network can copy itself, and invitations are how. When LinkedIn is founded in 2002, many doubt that people will share anything professional online. Reid Hoffman, its cofounder, tells Chen over a video call, an orca carving on the wall behind him, that a professional network is a hierarchy. Bill Gates gets more requests than he can answer. LinkedIn is built for the middle: successful people, still hustling, who actually take the meetings. For its first week the site is invitation only. The founders and investors invite their contacts, perhaps a couple of thousand people, and by the second week the core is strong enough to open up. Every invited user arrives knowing at least one person, like a guest greeted at the door of a dinner party. And the earliest users have the biggest address books. In Chen’s words, “The connected bring in the more connected.” Scarcity adds hype. Gmail opens by invitation on April Fool’s Day 2004, offering a gigabyte of storage. The reason is practical, not hype: it runs on three hundred old computers nobody else at Google wants. The hype comes anyway. Invitations sell on eBay for a hundred and fifty dollars.
Another way past the empty network is a tool that is useful on its own. In the App Store’s early years, two friends from Wisconsin make a photo app with vintage filters. It gets millions of installs, a feature in the New York Times. Its name, Chen reveals, “drumroll, please,” is Hipstamatic. But it is only a tool. You swipe through virtual lenses, wait about ten seconds between shots, and pay a dollar ninety-nine, and your photos just sit in your camera roll. That same year, on Pier 38 in San Francisco, Kevin Systrom and Mike Krieger are building Burbn, a cluttered app for check-ins and plans. They cut everything but the photo, the comment and the like, and call it Instagram, partly, Systrom says, because it sounded camera-y. It is free, with filters one tap away, and it is a network from the first day, with a feed, followers and sharing to Facebook. It launches on October 6, 2010, and passes a hundred thousand downloads in its first week. Six months in, sixty-five percent of its users follow no one; they came for the filters. In 2011 Serena Williams, Drake and Justin Bieber post their first photos, and the feeds fill up. Eighteen months after launch, Facebook buys it for a billion dollars. Years later, most photos carry no filter at all. Chris Dixon’s phrase sums it up: “Come for the tool, stay for the network.”
Sometimes the fastest way to tip a market is to pay for it, and the idea is older than software. In 1888 Coca-Cola’s founders print cards good for a free glass, and over two decades they give away eight and a half million drinks. A grocery store has its own chicken and egg: grocers won’t stock what shoppers don’t ask for. The advertising man Claude Hopkins solves it for Van Camp’s evaporated milk by paying the hard side, the grocers. Each coupon is good for a ten-cent can, and Van Camp’s pays the grocer full price. New York is dominated by a rival brand. In three weeks, mostly by letter, Hopkins gets the milk into ninety-seven percent of its stores. Then, one Sunday, a full-page ad with the coupon runs across Greater New York. One million four hundred and sixty thousand homes try the milk, Hopkins writes, “all in a single day.” The campaign pays for itself within nine months. Uber starts the same way, guaranteeing its first drivers thirty dollars an hour just to leave the app on. The common wisdom, Chen notes, is to stop waiting for the egg and simply buy the chicken. Microsoft pays in effort instead. In 1981 it strikes a deal to supply the operating system for IBM’s new PC. It keeps the right to sell that system to every other maker, and the clones that follow hand it the market. Losing money can be smart, Chen argues, as long as the subsidies stop once the market tips.
When the hard side isn’t there yet, you can play it yourself. Chen calls this Flintstoning, after the cartoon car that runs on Fred’s feet. Reddit launches in 2005 as a single page of links that users submit, and the only users are its founders, Steve Huffman and Alexis Ohanian. “No one wants to live in a ghost town,” Huffman tells Chen, on a sofa in his San Francisco office. So the founders post every day from dozens of fake accounts, and Huffman writes code that pulls stories from news sites and posts them under made-up names. It still needs him. About a month after launch, in July, he goes camping with his family, and when he checks, the front page is blank. Others play the hard side too. DoorDash lists restaurants that haven’t signed up, and sends a courier to buy the food like any customer. Nintendo builds the Switch’s first great games in its own studios. But Flintstoning needs an exit, because fakes can crowd out the real creators. Huffman remembers the day he knew he could stop.
…one day I was hanging out in Boston and didn’t submit anything into Reddit. I was worried I’d left an empty homepage, but when I pulled it up, it was full of links! I clicked on the usernames of the people who’d posted that day and saw that yes, they were real people.
— Andrew Chen, The Cold Start Problem, Harper Business, 2021, p. 170
By then, Chen writes, Reddit has a few thousand users and runs on its own, and each community it later splits into needs at least a thousand subscribers to sustain itself. Uber’s networks are never finished, because every city is its own cold start. In 2015 Uber quietly flies more than four thousand employees to Las Vegas to celebrate ten billion dollars in gross revenue. The logo on the T-shirts and water bottles is just two white Xs, so tourists won’t notice. David Guetta and Kygo play one night, and Beyoncé gives a private concert the next. Chen still keeps the water bottles. The credit, he says, belongs to Operations, the thousands of people who launch cities. “Product can solve problems, but it’s slow,” Kalanick tells the product teams. “Ops can do it fast.” Many cities launch with a local celebrity as Rider Zero. Teams send trucks of puppies to offices for an hour, or soft-serve ice cream on demand. The hustle has a gray side. Ordinary people driving for pay is not legal at first in many places, and police raid local Uber offices. But the system works: young general managers who feel like the chief executives of their cities, racing their sister cities. One night in Vegas, in a stadium, Kalanick unveils fourteen company values. One of them is Always Be Hustlin’.
Reaching Escape Velocity doesn’t mean the hard part is over. Chen often meets Drew Houston on unhurried Sundays at Yank Sing, a dim sum restaurant near the Ferry Building, and they talk about Dropbox’s teenage years. Tired of losing USB drives, Houston builds a magic folder that syncs across computers, and in April 2007 he narrates a four-minute demo video. The beta waiting list, he says, went from five thousand to seventy-five thousand “literally overnight.” By 2012 Dropbox has a hundred million registered users, a four-billion-dollar valuation and nearly two hundred employees, mostly engineers. It is an awkward teenager. When its small sales team is swamped, it simply takes the team’s email address off the website, and in the engineering culture, Chen writes, revenue isn’t considered “Dropboxy.” Then the hosting bills from Amazon’s cloud climb like a second hockey stick. A new growth team splits users into high-value and low-value actives. A count of file types says photos, and Dropbox builds a photo app, Carousel, which disappoints. A better question, which files people return to, edit and share, gives a different answer: documents, spreadsheets, presentations. By its public offering in 2018, Dropbox calls itself a global collaboration platform, and it has reached a billion dollars in annual recurring revenue faster than any subscription-software company before it.
At this stage, thousands of people work furiously inside a company. Instagram had thirteen employees when Facebook bought it; Dropbox, before its offering, has more than two thousand. Yet ask a team to improve its network effects, and it will just stare back. So Chen breaks the idea apart. The network effect, he argues, “is not one effect.” It is three. The Engagement Effect makes a product stickier as the network fills in, as Twitter grows from a nerdy friend or two to news outlets, celebrities and politicians. The Acquisition Effect lets the network bring in new users by itself. The Economic Effect improves the business, as when Slack charges to search every coworker’s messages. Each force moves a number teams already track: new users, plus returning users, minus the users who leave.
Engagement starts, oddly, with scurvy. In 1753 a Scottish naval surgeon, James Lind, reports on twelve sailors with the disease aboard HMS Salisbury. He splits them into six pairs and doses them daily with remedies such as cider, vinegar and sea water. One pair, luckier than the rest, also gets two oranges and a lemon, and by the time the fruit runs out, they have nearly recovered. Lind’s method, comparing groups over time, becomes the cohort retention curve. For most apps the curves are grim. Seventy percent of the people who install an app are gone the next day, and ninety-six percent within three months. Chen’s benchmark at Andreessen Horowitz is sixty percent of users still active after a day, thirty after a week, fifteen after a month. Networked products can beat it, and the rarest draw a smile, with retention rising over time. The levers are new uses as the network grows, like Uber’s company-wide Slack channel called Pool Party; a tighter loop between users; and bringing lapsed users back. Uber finds several million people a week failing to recover their passwords. And a note that your boss has shared a folder with you, Chen notes, pulls you back far better than any promotional email.
The Acquisition Effect is viral growth, and Chen learns its science from the PayPal Mafia. PayPal begins as FieldLink, for beaming money between Palm Pilots, and moves to the web. In November 1999 growth is still tepid; the team, David Sacks recalls, doesn’t know whom to target. The turn comes by email, from an eBay PowerSeller who has designed a “We accept PayPal” button and asks to put it on auction listings. The team searches eBay and finds hundreds of listings already mentioning PayPal. It says yes, then goes further: sellers can enter their eBay login, and PayPal inserts the button into every auction. “In other words,” Sacks says, “we productized the idea.” Buyers see the badge, sign up, and put it on their own listings. Ten dollars for every user who invites a friend, and ten for the friend, adds fuel. Fewer than ten thousand users become a hundred thousand within a few months, then a million, then five million within a year. Because growth like this is built into the product, it can be measured and tuned, one step of the loop at a time. Acquisition alone isn’t enough, though. In the 1930s a Denver chain letter called the Prosperity Club asks for a dime and promises more than fifteen hundred dollars back. Its last line asks, “Is this worth a dime to you?” It reaches hundreds of thousands, and swamps local post offices until the Post Office outlaws letters like it. A network that only acquires, and never keeps anyone, runs out.
The Economic Effect has the oldest roots. In 1776, five hundred and fifty London merchants form a Society of Guardians to protect trade against swindlers and sharpers, and each member must report anyone unfit to trust. More merchants mean more data, and more data means better judgments of risk; that loop grows into credit bureaus like Experian and Equifax. At Uber the same arithmetic decides price wars. In early 2017, after burning nearly fifty million dollars a week in China, Kalanick announces a new theme: Efficiency over Subsidy. Say Uber guarantees a driver twenty-five dollars an hour. In a small network the driver gets one ten-dollar trip an hour, so Uber pays fifteen dollars per trip to make up the difference. Chen’s verdict is one word: “Ouch!” In a big network the driver gets two trips, and the cost falls to two dollars fifty a trip. So the bigger network can pay more for the hard side, cut prices every January, and still come out ahead. Features can be cloned, Chen writes, but it is “nearly impossible to copy a network.” Even so, every network eventually grows more slowly, sometimes to a standstill.
At scale, an exponential curve turns into a squiggle, and Twitch is born from a company that stalled. Justin.tv begins with Justin Kan wearing a camera on his baseball cap and streaming his whole life, the laptop that sends it riding in his backpack. Its first network is Justin and the tech insiders who watch him, Chen among them. By the end of 2010 it streams everything from singing to bootleg football games. It is profitable, and it has stopped growing. On the internet, Kan says, not growing means you are on the brink of decline. Gaming is perhaps two or three percent of the traffic. Emmett Shear and Kevin Lin interview dozens of its most devoted users and plan a new site, at first called Xarth. The board hates the plan, which would turn a profitable company into one losing millions. The team goes ahead anyway, and builds for the streamers rather than the viewers. It gives them tools, high-definition video, categories by game with the most-watched streamers on top, and tipping, because an extra fifty dollars a month is a big deal. Lin explains the atomic network as one streamer and one viewer: playing for even one person is more fun than playing alone. Within a month of launching at the E3 game show in 2011, Twitch has eight million unique viewers. It sells to Amazon for nine hundred and seventy million dollars, and its main boardroom is still called Xarth.
How fast is fast enough? Chen does the venture arithmetic. Of the startups that venture firms back, more than half fail, and only one in twenty returns ten times its money. A company worth a billion dollars needs about a hundred million in recurring revenue within seven to ten years. For software, the investor Neeraj Agarwal’s path is to reach two million dollars a year, then triple, triple, double, double, double. A marketplace that earns its first million in year four must grow about two and a half times a year to reach two hundred million by year ten. Growth rates tend to fall, even as spending rises, while success inflates belief. A question once met with a laugh, whether this is the next Facebook, gets a serious yes. In pitch decks Chen sees growth of five times, then four, then three, more often than the reverse. When the easy ideas are taken, a product has hit the ceiling.
In 2000, in Jeff Jordan’s first month running eBay’s American business, it fails to grow month over month for the first time. Auctions intimidate many newcomers, and Jordan’s team proposes something radical: a fixed price, called Buy It Now. It is controversial inside eBay and among its users, and Jordan recalls swallowing hard before launching it. Today it is sixty-two percent of eBay’s sales. Stores and PayPal at checkout follow, each another layer in what Jordan calls a layer cake. What looks like an exponential curve, Chen writes, is often “a series of lines layered quickly on top of each other.” Networks saturate too. A search for a vintage Rolex Daytona gets far better with the first few listings, maybe a few dozen, but not with the five thousandth. At Instagram, Bangaly Kaba finds growth slowed to a straight line at about four hundred million users, and goes looking for adjacent users: people who tried the product and couldn’t get into it. At first they are American women aged thirty-five to forty-five with Facebook accounts. Three years and a billion users later, by Kaba’s account, the adjacent user is a woman in Jakarta with an old Android phone on a prepaid plan. In Bangkok, Uber becomes almost unrecognizable: sign up with just a phone number, pay cash, and hold on tight on the back of an Uber Moto.
Every marketing channel wears out. Chen gives it a blunt name, the Law of Shitty Clickthroughs, and traces it to 1994. Until then the web has no ads at all. That year Hotwired, a cousin of Wired magazine, runs the first banners, for advertisers including AT&T, Volvo and Zima. The internet is dial-up, the ads load slowly, and perhaps two million Americans are online. One banner asks, “Have you ever clicked your mouse right HERE? You will.” Seventy-eight percent of the people who see those first ads click on them. Today it is around one percent or less. The decay compounds through a network. Suppose a hundred users invite seventy-five, who invite fifty-six, and so on. Halve the share of invitations that convert, and the total of new users falls by eighty percent. The answer is to keep adding channels, to jump on new platforms early, and above all to strengthen the network rather than outspend the decline.
The hard side can turn against you. In 2016 Chen lives in a loft in Hayes Valley, a short walk from Uber’s headquarters on Market Street. Each morning starts with an americano from a little alley coffee shop, then a stroll past bars and boutiques to the office. A few times a year, the walk ends differently.
…dozens of upset and highly vocal Uber drivers, holding signs outside the lobby. They shouted, banged drums, and were irate, and they often circled for hours. … I once had an irate driver find out I worked at Uber via my social media profiles, and they waited in the lobby for hours to accost me when I arrived. I walked extra quickly to the elevators when he shouted my name.
— Andrew Chen, The Cold Start Problem, Harper Business, 2021, pp. 273–274
Upstairs, Chen’s meetings keep returning to the same paradox. Drivers are about five percent of Uber’s users, yet a new driver costs ten times as much as a new rider, and in San Francisco as much as two thousand dollars. Power drivers, who work forty hours a week or more, are the ones outside with handwritten signs, asking for better pay and treatment. They are fifteen percent of drivers and give over forty percent of the rides. Chen sympathizes, and every network has the same tension. Vine’s eighteen biggest stars ask for one point two million dollars each, in return for a dozen posts a month. Vine says no, and a few years later it shuts down. Professionalizing the hard side brings quality and conflict together, and Chen sees no choice but to embrace it. Sometimes it brings chaos too. Uber’s car-leasing program for would-be drivers loses five hundred and twenty-five million dollars, and some of its cars turn up in Iraq and Afghanistan with their GPS still running.
Usenet, created in 1980, is the granddaddy of social networks, a web of discussion groups where the World Wide Web, Linux and the modern graphical browser are all first announced. Its first atomic network is Duke University. Every September a new class of students arrives, learns the netiquette over months, and fits in or is flamed out. Then, in September 1993, AOL starts mailing millions of discs to American homes, and the newcomers never stop coming. The Usenet pioneer Dave Fischer calls it “the September that never ended.” Spam, piracy and pornography follow, the good groups grow hard to find, and in 2010 Duke switches off its servers. Adam D’Angelo of Quora describes the deeper danger, which Michael Wesch named context collapse. You join with close friends, then strangers arrive, then your parents and your boss, and the photos from a party might get you in trouble. Scams are older still. In 1828 the French detective Eugène François Vidocq describes the Jerusalem Letter, from a marquis’s valet offering a share of lost gold to whoever helps recover it. Reddit’s answer is self-government in software. The downvote, Steve Huffman writes in testimony to the US House of Representatives, “is where community culture is made.” Usenet could have been saved, Chen thinks. But it was never a company, and no one had the job of changing it.
Overcrowding is the last ceiling, and YouTube meets it fast. It starts in 2005 as a dating site for video profiles, and within weeks opens up to anything. Its first video is nineteen seconds of Jawed Karim in front of the elephants, admiring their long trunks, and signing off: “That’s cool, and that’s pretty much all there is to say.” The first thousand videos, Steve Chen tells Andrew Chen, were the hardest stretch of all. Views climb from a thousand a day to a million in under a year, and finding anything gets hard. Google buys YouTube in 2006 for one point six five billion dollars and brings search and related videos. Sorting by popularity has a trap: the rich get richer, and Eugene Wei warns of old money that newcomers struggle to catch. YouTube now takes in nearly six hundred hours of video every minute. Keeping a network healthy is an everyday battle.
Network effects don’t protect you, because your competitor has them too. In 2011 Airbnb meets Wimdu, a clone out of Berlin, built by the Samwer brothers’ Rocket Internet down to the colors and the text. It launches with ninety million dollars, the largest investment ever in a European startup, and hires more than four hundred people in under a hundred days. Airbnb is two and a half years old, has forty employees, and takes payment only in dollars. Brian Chesky has heard that the brothers kill whatever they clone, and to him it is as if a “giant dragon” has appeared. Wimdu’s bots scrape Airbnb’s listings, and its staff book stays with Airbnb hosts to sign them up in person. The Samwers offer to merge. Chesky refuses. His revenge, he says, is to make them run the company: “So you had the baby, now you gotta raise the child.” Airbnb translates its site and adds thirty-two currencies to the dollar. In 2012 the managers of its new country offices gather in a rented house in Spain to plan their march across Europe: seven offices in four months. Airbnb’s organic listings are far better. Wimdu’s best ten percent, an early Airbnb employee says, matched Airbnb’s worst ten. Within two years Wimdu is laying people off, and by 2018 everyone is gone.
Warren Buffett looks for moats, and for a networked product the moat is the network, because features can be cloned and networks can’t. Once Airbnb has three hundred listings in a city, a newcomer must solve the Cold Start Problem all over again while Airbnb grows to four hundred and five hundred. Uber’s moats run city by city, so it fights trench warfare; Airbnb’s are global, because hosts in Miami serve travelers from everywhere. Still, every company in a networked category has network effects. Believing they will hold off competitors, Chen says, is “a lie that entrepreneurs tell to themselves.” If a network’s value grows as users join, it can fall apart just as fast when they leave, all the way back through the Cold Start Problem. Lose Berlin, and Wimdu loses Germany, then Europe. The fight is lopsided in both directions. Goliath has scale but moves slowly; David has speed and no sacred cows, and needs only one atomic network.
Airbnb’s first rival wasn’t a clone at all, but Craigslist. Craigslist still looks like the 1990s, all blue links and gray boxes. It serves five hundred and seventy cities, earns an estimated billion dollars a year, and runs on a staff of a few dozen. In 2010 the investor Andrew Parker notices it being unbundled: Indeed takes the jobs, StubHub the tickets, Etsy the crafts. Craigslist is a network of networks, and as Chen puts it, an upstart needs only one entry point while the incumbent has to protect them all. Its room rentals are poor, with no reviews and no way to check dates, so Airbnb builds photos, reviews, payments and host profiles. Early on, a host who finishes an Airbnb listing can post it to Craigslist with a link back, through a bot built without Craigslist’s help. By the time Craigslist shuts it down, Airbnb has its atomic network. The danger is leaning on a platform too long. Apps built on Twitter and Facebook became worthless overnight when their access was cut.
The incumbent’s instinct is a big launch, like Steve Jobs in his black turtleneck announcing the iPhone at the Moscone Center in 2007. For a network it is a trap. In June 2011 Google launches Google+ to take on Facebook, linking it from the Google homepage and wiring it into YouTube. Within months it has more than ninety million sign-ups. But a Wall Street Journal report finds a virtual ghost town: about three minutes a month per user on computers, against six or seven hours on Facebook. Google+ shuts down in 2019. A wide launch, Chen argues, scatters users across many weak networks, none able to hold up alone. Apple can launch big because its products stand alone, but, he adds, “you aren’t Apple.” The giants all start in places that look too small: collectibles, college students, limos for the rich, air beds. Fred Wilson’s firm passes on Airbnb because it can’t see a blow-up bed on someone’s floor competing with hotels. For Tinder, a few hundred users at USC feel enormous, because anything is better than zero.
Back in Uber’s War Room, the North American Championship Series often meets at odd hours, at ten at night or on a weekend, and runs as long as it takes. Its premise, Chen writes, is “that it wasn’t enough for Uber to win; others had to lose.” The fight is over drivers. To flip Lyft drivers in San Francisco, Uber’s operations team sends billboard trucks to circle the block around Lyft’s headquarters. The billboards tell drivers to shave the stache, Lyft’s pink mustache, and drive for Uber. Lyft answers with trucks of its own: Be more than a number. Uber hunts for drivers who run both apps, and offers them bonuses big enough that they must choose one app when the week begins. At the peak, a single region sees more than fifty million dollars a week in driver incentives. Against Sidecar it works. Its cofounder Jahan Khanna tells Chen that once they stopped matching bonuses, all their markets went to zero within six weeks. Against Lyft and DoorDash, and in China, it doesn’t. Even with the bigger network overall, Uber is only even with Lyft in San Francisco and Los Angeles. Winner take all is a myth.
Can a giant simply bundle its way into a market? Microsoft is the famous case: Internet Explorer, shipped with Windows, beats Netscape. But Brad Silverberg led Windows 95 and the first versions of Internet Explorer, and from his retirement in Jackson Hole he tells Chen that bundling “is not the silver bullet everyone thinks.” The first Internet Explorer is so poor that people use it once, the joke goes, to download Netscape. Steven Sinofsky says Microsoft’s early Word and Excel trailed their rivals. In Chen’s telling, the move to graphical screens let them leapfrog. Microsoft wins with a good product plus its whole network, above all its developers. Visual Basic turns small businesses into programmers, and for every copy sold, ten copies of Windows go with it. And when Netscape arrives, Microsoft sets aside its rivalry with AOL, so that the CDs AOL mails by the million to American homes carry Internet Explorer inside. A decade later it has nearly ninety percent of the market. Bundling fails, though, when the product is weak. Google+ and Uber Eats are bundled everywhere, and both still fall behind.
By late 2018 Uber has a new chief executive and new values, and the War Room has been renamed the Peace Room. Chen has been gone about a year. In October an email reaches thousands of alumni about reunions in cities from Sydney to Mexico City. The San Francisco one is at Monroe, an Art Deco lounge in North Beach, and old friends trickle in through the night to trade war stories. About an hour in, the music stops for an announcement.
The room hushed as Travis grabbed the microphone:
The thing that this crew does—it makes me proud to come here and just spend time with you guys and see you. … Follow your dreams. Dream big. Do big stuff.
And with that, he had addressed the troops one last time. Travis smiled, looked around the room, and handed over the mic.
— Andrew Chen, The Cold Start Problem, Harper Business, 2021, p. 372
Chen goes on to count where the troops have gone. There are several dozen new startups, in scooters, virtual kitchens, payments and furniture, and some alumni join venture firms, as he did, to back whoever might be the next Uber, Dropbox or Slack. It is Silicon Valley’s circle of life, the way PayPal’s alumni once went on to build YouTube and LinkedIn. Networks will reshape more industries yet, Chen believes, with crypto among the most important. And the people who learned these lessons at Slack, Dropbox, Twitch, Microsoft, Zoom, Airbnb and PayPal are already carrying them into whatever networks come next.
