While Twitter struggles to get new users,
its revenue side is killing it. Welcome to the not-so-dismal science of Twitternomics.
In 2009 Twitter was a 50-person company punching way above its weight in cultural impact, its micro-blogging platform blasting its way into the public imagination. But its ambitions were even higher. According to leaked internal documents, the company had privately set goals over the next few years of a billion users and well over a billion dollars of annual revenue. Many people thought that the first goal was inevitable and that the second would be a slog. After all, Twitter’s user base was skyrocketing much in the way of other Internet phenoms like Facebook and Google. But there was virtually no income and no monetization plan in sight. When a Twitter co-founder appeared on the Colbert Report that spring, the host remarked, “So I assume that ‘Biz’ in ‘Biz Stone’ does not stand for ‘Business Model’?”
It wasn’t only comedians who held that view. In that summer’s Sun Valley mogulfest, media icons Barry Diller and John Malone, on a “Making Money on the Internet” panel, called out Twitter as an example of a fad product that might never generate serious cash. Cable-TV pioneer Malone said it was unlikely that the company could build a significant advertising business.
Twitter now is a public company with a headcount of 3600 and a valuation of over $23 billion. But it has not come close to a billion users (not even 300 million). The shortfall has unleashed a torrent of criticism, and even intimations that an executive putsch will be required to address endemic product woes.
On the other hand, in 2014 Twitter did generate that billion dollars in revenue. (The final number will come in when the company reports earnings later this week, but its earlier estimate was $1.375 billion.) And the stash has been rising at an annual rate of over 100 percent. Twitter’s bottom line is not in the black, but blame that on continuing expenses in growing its business and (mainly) paying out huge amounts of “stock-based compensation” to employees. That’s not uncommon in the Internet world: more notable is that Twitter has cracked the code to making money on the net.
You heard that right: while the headlines about Twitter describe a company in crisis, the business end of the company has been nailing its targets, and its revenue team is the envy of the industry. “Twitter sold the wrong thing to investors — it sold the user story,” says analyst Brian Weiser of Pivotal Research Group of the company’s emphasis during the 2013 IPO rollup. “In spite of obvious evidence that user growth was slowing, they picked that as the story, and they failed to deliver. But revenue is doing what it should do.”
So how is Twitter making its money? The obvious answer is advertising — those “Promoted Tweets” that appear unbidden in a user’s timeline. Getting there wasn’t easy. And it takes effort to understand how those ads work. But in contrast to the well-documented turmoil in the company throughout its history, the road to revenues for Twitter has been a steady, coherent, patient and innovative path. And more recently, the company has embarked on a strategy to go beyond those ads to make money from selling products directly, and even monetizing people who don’t use Twitter. (More on this in Part Two of the series.)
What’s more, Twitter has turned this trick in its own unique way, creating a system that’s true to its peculiar culture. So here, in somewhat more than 140 characters, is the tale of Twitternomics.
In mid-2009 Twitter’s CEO was co-founder Evan Williams. (Williams is still on Twitter’s board of directors — and, I should disclose, he is also the CEO of Medium, which makes him my boss.) Just after that Sun Valley panel — Williams was in attendance— I asked him about the incident in an interview for a story for Wired. “I didn’t argue my case, but all the Internet guys I talked to were laughing at the media guys,” he said. “Are you kidding? There’s obviously a huge business there.” Later in the interview, though, he blithely admitted that Twitter had made little headway in that direction, mainly because it wasn’t yet a priority. But he noted that since Twitter was widely used by corporations to connect with their customers, the transition would be natural.
Today he marvels how people missed what seemed obvious to him. “I always joked that the press reported as if we were looking for a business model that we lost somewhere — maybe it’s under the couch, where is the business model?” says Williams now. “But no one has ever built a consumer Internet thing that reached scale that they couldn’t earn money on.”
Nonetheless, back in 2009 Williams knew that Twitter had to make good on its prospects—sooner, rather than later. So several things happened that fall.
First, Williams hired Dick Costolo to be his chief operating officer. Costolo, like Williams, had built a company acquired by Google (Costolo’s was Feedburner; Williams had earlier sold Blogger) and spent a couple years at the Googleplex before leaving. Williams says a big thing in Costolo’s favor was that the Chicagoan understood advertising: the newcomer’s explicit mission was to build the business side. Twitter’s revenue at the time Costolo arrived on September 1, 2009? “Zero dollars and zero cents,” says Costolo.
Within a couple of months Costolo had completed some deals previously set in motion to move the needle from zero: Google and Microsoft agreed to pay Twitter for direct access to its information (its “firehose” of data). But Costolo knew that Twitter’s commercial future lay in advertising. There were obstacles. For one thing, a substantial percentage of the Twitter service accessed it via third-party sites and apps like Tweetie and TweetDeck. Twitter didn’t even own the Android app for its product. “If we’re going to monetize Twitter, we have to make money everywhere the tweets go,” says Costolo. “The tweets go all over the place so how are we going do that?” (Eventually Twitter would buy out some of the bigger third party clients and limit the data it provided to the remaining ones, so that the vast majority of users would actually use a Twitter-owned product to access the Twittersphere.)
To build an ad system Costolo turned to a physicist named Kevin Weil.
Weil had joined Twitter in the spring of 2009. His resume was standard issue for a Silicon Valley high flier: undergrad at Harvard (math, physics), physics doctorate at Stanford, and abandonment of plans to be a researcher in order to join the tech circus bustling all around him. After working for a couple startups, he agreed to head Twitter’s very small revenue team, at first trying to develop a paid analytics system for companies that wanted to use Twitter to track events and consumer sentiments in real time. When it was time to build an ad product later in 2010, Weil’s five-person team was the obvious choice. The technical lead on the team would be eventually be Alex Roetter, a former Googler.
The answer seemed so simple that it is now impossible to imagine otherwise. The ad would be a tweet.
The person probably most responsible for this concept was Anamitra Banerji, an early Twitter employee who had brainstormed Twitter’s ad strategy and got things in motion even before Weil’s team began building the final system. In November, 2009, he’d presented a long memo to Twitter’s leaders outlining a road map for Twitter monetization, including some dummied-up sponsored tweets.
“Now it seems obvious, but back then it was not obvious,” says Weil. “People said why don’t you just throw up AdSense on the side of the page and make a bunch of money? But we needed to build something that felt like Twitter. We didn’t want to build something where we knew the ads weren’t very good, so we’ll shove them over to the side of the page and maybe you won’t hate them that much.”
Even now, Costolo marvels at the simple genius of the approach. “An ad can go everywhere tweets go because it’s just a tweet,” he says. “You can do anything to it that you can do to a regular tweet, and we could actually use the signal from how many favorites and replies and re-tweets it gets as a way of thinking about how much farther to broadcast the ad, the Promoted Tweet.”
Only later did another huge advantage of dropping the distinction between a tweet and an ad emerge: the approach allowed for a seamless transition to mobile. (About 80 percent of Twitter use is now on mobile devices.) “Promoted Tweets fit naturally within the product, wherever it goes,” says Weil. “Everybody else had a struggle when they switched to mobile. But we were already there.”
Also, because ads would literally be tweets, they would instantly take advantage of any improvements that Twitter made in its general experience — like more prominent photos, videos, Vines, and app downloads (the latter has become a huge source of income in recent months, as developers consider Promoted Tweets on a par with Facebook as the best way to get appropriate customers to consider their apps). “Promoted tweets instantly get the same superpowers, which is nice,” says Weil.
In hindsight those virtues indicate that ads-as-tweets preclude any other approach. But at the time, Twitter could not be sure. Would Twitter users even tolerate ads in their timeline? Could it produce a pricing system that piled up revenues while giving advertisers value? It was up to Weil’s team to find out.
Despite the pressure to make money — and a silent but undeniable urgency to develop a financial profile that would eventually justify an IPO — Twitter was determined to take baby steps. The first tweet-as-ad experiment for instance, did not show up in anyone’s main timeline: those Promoted Tweets appeared only in the relatively little-used search feature.
That first Promoted Tweet, by the way, came from Starbucks. It appeared on April 13, 2010. It was indicative of the fact that while Twitter was itself not a true business yet, its service hosted virtually every corporation you could name. (In contrast, Google had started its ad business with a self-serve search ad from a small business selling mail-order lobsters.)
“Instead of building self-serve right away we said, let’s work with these big global brand advertisers we can have direct relationships with, and understand how it’s working or not working,” says Costolo. The coffee giant was one of six charter Twitter ad customers, along with Red Bull, Sony Pictures, Best Buy, Virgin America, and Bravo. The initial deal they got from Twitter was irresistible. “For a month or two we didn’t charge them a dime,” says Weil. “Our focus at that point wasn’t about money— it was entirely about how users would react. We wouldn’t make our first dollar until June.”
A few months later, Twitter introduced a more traditional product placement, Promoted Trends. It came from Twitter’s sale of a spot on its “trends” list signaling hot topics in the Twittersphere. An advertiser could spend a sum — in the US it’s typically around $200,000—for an otherwise unearned 24-hour placement on that list.
When Porsche rolled out its new 911, the automaker doubled down, buying a Promoted Trend that linked a Promoted Tweet in the search timeline. The engagement rate for that campaign, says Weil, was 87 percent. (Yes, that sounds crazy-high—but Twitter says it was a rare case where the ad content—a super-slick video—and a fascinated audience just worked perfectly.) “Those kinds of things gave us the confidence that promoted tweets in search could be a good user experience,” he says.
Later that year, the revenue effort at Twitter got a new leader: a NewsCorp executive named Adam Bain. Bain had helped build out the company’s online ad division, but had become convinced that his bosses didn’t understand its value. He and some others at NewsCorp were trying to buy the division to spin it out as a separate venture. In January 2010, Bain had deflected Costolo’s offer, because he felt that neither the employees nor the customers of Twitter were ready to see the company monetize the product.
“I found a profound determination to show the world that this was going be a big and meaningful company and a big and meaningful business,” he says. Also, Bain was a big fan of Twitter itself, in contrast to his disinterest in the actual product he was hawking at NewsCorp. Finally, Bain had become disgusted with the traditional Internet display ad model, and welcomed a chance to try something less odious.
He began working at Twitter in September. His first day was a Monday, and he learned that he was expected to deliver his monetization plan to the board in three days. Bain presented the board two options. The first was to pursue a traditional form of display ads, which would give an instant revenue spike. This was sort of a sandbag, representing the kind of digital marketing that Bain had come to loathe. The second alternative was more exciting: Twitter would invent a new product, with a completely different way of targeting users, and a completely different way of measuring success. The two lodestars he pointed to were Google’s search ads — already a model for Weil and Roetter’s team—and television.
Google’s search system, called AdWords, was an obvious model: it was the Internet’s most successful product. Using television as a model did require some explanation: “In a television, you have a thirty-minute story that’s told to you, has a story arc, a beginning, middle and end, and it plays out inside those thirty minutes,” says Bain. “Interspersed between those stories are other stories, stories by marketers, nano stories. They are thirty-seconds instead of thirty minutes, but they also have a story arc, a beginning, middle and end. So the only difference on TV is who is telling the story at that time.” In other words, a Twitter timeline would have two narratives: organic and paid. But, he cautioned, both must be compelling.
The system that Weil was building was totally in synch with that sentiment. To put Promoted Tweets in actual user timelines, the key would be making those ads welcome to users and low-risk to advertisers. (Bain admits to worries of “folks with torches and pick axes at the door.”)
The eventual system did in fact adopt some key concepts of the Internet gold standard for an ad model: Google AdWords. Fortuitously, Weil’s engineering lead Alex Roetter had helped create AdWords’ sister product AdSense during his six and a half years at Google. As with AdWords, Twitter advertisers buy via auction. (That way, they knew pricing was fair.) And Twitter does not charge advertisers by how many people see the promoted tweets: the payments would only come when users actually engaged with those ads, via retweets, replies, favorites, clicks to follow links. (This guarantees advertisers not only that their money won’t be wasted, but also that their results will be endlessly measurable.)
That’s the trickiest part of the system. In order to conduct a fair auction — with advertisers bidding against each other for a certain amount of engagements, rather than simply a number of tweets inserted in timelines — Twitter has to be able to accurately predict which people are most likely to engage with a given ad. That sort of prediction was the secret sauce of Google’s AdWords: if Google believed that an ad would be relevant to the users who saw it, the price would be lower.
Google could do this with confidence because its users explicitly express what they’re interested in at that very moment — by typing keywords into the search engine. But Twitter didn’t have any signals as solid as that one. And it also didn’t have the deep knowledge of one’s connections that Facebook calls “the social graph.” But Twitter does have a lot of information that can help it predict which users would welcome a given Promoted Tweet. It calls this “the interest graph.”
By analyzing what you tweet, it can infer your intent, much the way Google knows intent by what you’re searching for. It also can figure out who you are, to a certain extent. “Twitter never asks you whether you’re male or female,” says Weil. “But we’ve found that we can infer that data with about ninety percent accuracy just based on having a little bit of ground truth data and then understanding look-alikes and other things.”
Using information it has about users and their behavior, as well as analysis of the Promoted Tweet itself, Twitter comes up with a predicted rate of engagement. Making those calculations is not one of those things you can try at home, unless you live at the Stanford computer science department. “Basically, it’s a machine learning problem where you’re trying to predict certain behaviors from users,” says Roetter, who goes on to describe how all sorts of user information — who they follow, how they interact with content relating to the advertiser, what their response is to Promoted Tweets — is combined with information about the advertiser, like how certain kinds of users have responded to its ads and similar ones in the past. Then, applying a sophisticated set of machine-learning algorithms, Twitter gets its precious expected rate of engagement.
But that isn’t enough. Before completing the auction, Twitter applies a final algorithmic quality hurdle to make sure that totally irrelevant and annoying ads won’t appear in people’s timelines. “If we just don’t think that the ad is going to be engaging enough or a good enough experience—even if it was the best ad that we could have shown out of all of those bidding— if we think the ad is not going be the best ad for the user, we just won’t show it,” says Weil.
While Weil and Roetter were honing the system, Bain was going on a months-long tour of potential customers, visiting around 140 chief marketing officers of major companies. “It was tough for Adam and his team,” says Costolo. “There was a lot of education work to do. Customers wondered why they couldn’t move over their search ads. We would say, just take one of your tweets that you know is already good because people are engaging with it, and you can promote it!”
One advantage that Twitter had was that businesses were already taking advantage of Twitter — they just weren’t paying for it. “Most of the big brands were already here,” says Bain. “Our opportunity was to take this activity, maybe it was mostly in customer service, and figure out how to redirect it into a marketing opportunity.”
Bain recalls visiting the headquarters of a major airline where the CEO spoke of how much they’d invested in using Twitter as an alert for customer complaints. When Bain asked for more details, the CEO took him to another building. “It was like a scene from James Bond,” says Bain. “You open the door and instead of people doing karate moves or whatever, you had a massive sea of customer service people on the phone. He said there a ton more of these overseas. Look at this sea of investment.” So Bain’s job wasn’t selling Twitter, but convincing advertisers that promoting ads could do more for them.
The first Promoted Tweets in people’s timelines, which appeared in October 2011, came only from brands that users explicitly followed. These were tweets that people would have seen for free — the difference is that by paying to promote those posts, advertisers could guarantee that followers would be exposed to them. For example, if a follower of Red Bull checked his or her timeline five hours after the brand made a tweet, in normal circumstances that tweet wouldn’t be seen. By promoting the tweet, though, Twitter would fiddle with the chronology and include the ad among the first few tweets in the current timeline.
“We weren’t going to start by inserting content that you wouldn’t have ever seen,” says Weil. “These were brands you already followed. We started with that and we gauged user reaction.”
To everyone’s relief, there were no pickaxes and torches. “One of the first things we did was run a long-term retention study to understand whether people use Twitter less if they had a version with ads,” says Roetter. “And it turns out, they don’t. The behavior is the same.”
Still, Twitter kept rolling like a Florida driver in a school zone. Costolo remembers a Reuters reporter asking him early in the process how many advertisers Twitter had. Costolo told him it numbered in the dozens.
Oh, the reporter said, will you scale up to tens of thousands by the end of the year?
“I said, ‘No, no, no,’” recalls Costolo. “We’ll have hundreds by the end of the year. We are not in any rush. We have to make sure the model works, that we have advertisers who like it, and who are willing to spend more the next time.” Costolo even had to preach patience to some Twitter board members, who would remind him how quickly Google scaled its AdWords product. “I would say, ‘Well, we’re not going to do it that way — we’re figuring this out for the long term.’”
It wasn’t until August 2012 that Twitter fully unleashed the system that Weil and his team had begun to roll out more than two years earlier, allowing advertisers to post Promoted Tweets on users who didn’t follow their brands. Even without the depth of data of Facebook or Google, Twitter could now offer micro-targeting, to what Twitter called “tailored audiences.” If an advertiser wanted to reach Beyoncé fans, Twitter’s range of targets isn’t limited to Beyoncé followers, but can also include people who have tweeted about Beyoncé or even similar artists. Twitter could also locate doppelgangers for die-hard Beyoncé fans — people whose data trails are very similar who may not have explicitly expressed enthusiasm for Mrs. Carter. Chances are those folks might like a promoted tweet just as much as the singer’s followers would.
In another example provided by Weil, McDonalds might want to target young males in an urban setting near one of its restaurants — who are tweeting about being hungry. That’s one of the lures of the Twitter system. The real-time aspect also is crucial during big events—cost-per-engagement soars during Oscars and Super Bowl time, when sponsors try to make a dent in the teeming Twittersphere.
Weil takes pains to emphasize that in no cases does Twitter turn over personal information to advertisers. It does, however, sometimes augment its information with data from outside sources. For example, Toyota might have a set of email addresses of people who recently test-drove cars. Or Samsung might know all the users who have a twenty-four month old contract up for renewal. It’s possible to match those addresses with Twitter users in a way where neither side learns any new information — but advertisers can send Promoted Tweets to just those targets.
From the start, they understood how Twitter’s auction system filtered ads for quality, and that superior ads would get better auction prices. But advertisers discovered that Twitter offered additional incentives for really great ads. Twitter charges sponsors only when users follow links or re-tweet their original ads; there is no additional charge when people engage with those second-order tweets. “When you run a great advertising campaign, people are going to re-tweet it all over the place,” says Weil. “If you get a whole bunch of views and everything else off of that, that’s free.”
That means that if advertisers create campaigns that go viral, much of the activity will be free. The classic example is the Oreo tweet during the 2013 Super Bowl, which suffered a 34-minute power outage. Even before the lights went back on in the Superdome, Oreo posted a Promoted Tweet saying, “Power Out? No problem” with an image of a cookie and the caption, “You can still dunk in the dark.” Many of the 15,000 retweets were people passing on their friends’s retweets—meaning that Oreo got free exposure. In 2014, a Budweiser Promoted Video (a tweet with video content) campaign, hit the jackpot: for every paid engagement, the brewery got six free ones.
When asked to provide different examples of a favorite Promoted Tweet, Costolo cites times when advertisers pick up on something potentially explosive, and piggyback their brands on the big interest. One example is the 2012 Daytona 500 when a NASCAR driver tweeted a picture of a fiery crash from the windshield in his car during the race: just when that was exploding on Twitter, Proctor and Gamble tweeted a image of a race technician going on the track to clean up with a huge box of Tide detergent.
The ultimate Promoted Tweets, in fact, are so Twitter-y that they make users forget the little icon and disclaimer that this post (almost always festooned with an image these days) is “promoted by” some entity paying to get in your face. Costolo cites a case where the wireless carrier O2 got into a faux Twitter fight using a cheeky Ali G-ish patois. Hey, that huge corporation soaking me for data charges is pretty cool!
Such cherry-picked victories make great fodder for Twitter sales pitches. But a more significant sign of Twitter’s success in advertising is the company’s general claim that Promoted Tweets have roughly the same three-to-five percent engagement rate as the universe of organic tweets — those unpaid comments, hot links, snark attacks, and celebrity photos that one normally finds in a timeline. In other words, Twitter has managed to slip ads into its users’ timelines without disturbing the universe.
All of this has led to a revenue story for Twitter that has kept at least some analysts optimistic, despite the company’s apparent inability to grow the user base to even half of its ten-figure dream. CEO Costolo finds all this ironic. “The court of public opinion has flipped to, Sure, you’re going to make money, but how are you going to get a global audience of multiple billions of people?” he says. “Which is just curious, because three years ago they were, Sure, you have this big audience, but how are you guys ever going to make money from this thing? That’s completely gone away now over the course of our last few public quarters, when we have announced tremendous year-over-year growth. “
On October 30 last year, Costolo extended the duties of revenue-side product manager Kevin Weil, appointing him in charge of all of Twitter’s products. Last May, Weil’s revenue counterpart Alex Roetter had received a similar promotion, moving from head of advertising engineering to VP of engineering for the whole company.
Thus the architects of the system that achieved one of Twitter’s goals — making money— now have the problem of dealing with the other one.
One suspects a billion users will be a lot tougher than a billion dollars.
Photography by Stephen Lam