Large swaths of the Internet economy are based on an advertising business model. Many content platforms, social media sites and mobile applications provide a free service to users in exchange for personal data that, once collected, is sold to advertisers and data brokers to generate corporate revenue. Metcalfe’s Law predicts that corporate revenues should increase exponentially as a company’s number of connections goes up and that costs should increase linearly. The combination of exponentially increasing revenues and linearly increasing costs should generate large profits for ad-based Internet companies. However, the opposite tends to be the case. Many established Internet companies are deeply in the red and only an estimated 0.01 percent of mobile applications will ever earn a profit. This disjunction raises the question: why do so many ad-based Internet companies perform so badly?
The answer lies in the interaction of two factors. First, the costs faced by advertising-based Internet companies tend to increase faster than their core resource (i.e., users). Many ad-based Internet companies, therefore, do not seem to benefit from economies of scale on the cost side of the equation. The implication is that such companies need a very large number of users in order to reach profitability. Secondly, profitability’s large user base requirement turns out to be extraordinarily rare in the Internet ecosystem because the network tends to structure itself into what are known as power law distributions, where most companies get only a few users (millions maybe) while some get literally billions. This suggests that the advertising model that underwrites so many Internet companies is broken. It works for a few, but not for most.
Selling ads: The basic business model of most content platforms, social networking sites and mobile applications
An expensive proposition: The cost of getting and retaining users
80/20: Understanding the shape of networks
Conclusions: The broken premise of the advertising business model
Some content platforms, social media sites and mobile applications — what we could call ‘Internet companies’ — are swimming in wealth. In 2015, for example, Facebook connected 1.59 billion users, generating US$5.81 billion in revenue in just the last quarter of the year, up 51 percent for the same quarter in 2014 (Constine, 2016). Google, for its part, generated US$21.32 billion in fourth quarter revenue and soundly beat analyst expectations of adjusted earnings per share (D’Onfro, 2016).
With surpassed expectations and unprecedented revenues for many firms, everyone wants to be a major Internet company. As a result, there are unsurprisingly tens of thousands of new Internet-based tech start ups each year. By one estimate, there were over 50,750 new apps in the iTunes app store alone in January, 2015 (Statista, 2016a). Annualized, that pace of new start-ups comes in at well over 600,000 new applications per year, not including those unique to Android devices. What is more, the pace at which new apps are entering the market per month is also growing fast.
On the face of it, the Internet ecosystem appears to be healthy and vibrant. However, many well-established Internet companies fail to make a profit and less than 0.01 percent of mobile applications are estimated to be profitable though 2018 (Gartner, 2014). There are, then, two diverging trends. On the one hand, there are huge revenues to be had for Internet companies. Yet, on the other hand, most still fail to make enough profits to stay afloat and must eventually fail. On the face of it, there is a seeming contradiction here, which begs the question: Why, when there is so much money to be had, do so many Internet companies perform so badly?
The interaction of two factors explains poor corporate performance for a specific universe of cases involving Internet companies that adopt the advertising business model — companies such as content platforms (Google; Yahoo!), social media sites (Twitter; Facebook) and mobile applications (Snapchat; Tinder) and conventional Web sites (media sites, etc.).
First, while the advertising business model worked well for television and radio companies, the Internet’s radical reduction of barriers to entry into a given sector means that there is always a new challenger to an incumbent company. These new entrants dilute market share by diffusing users across the ecosystem. Moreover, users — the ultimate resource for advertising-based companies — can decide to vote with their feet (or mouse) by moving to more satisfying services if the current platform does not innovate well enough or with sufficient alacrity. As a result, the costs that Internet companies need to pay in order to collect and retain users are surprisingly high and aggregate surprisingly quickly, often outpacing the growth in users themselves. Since profit comes when revenues are greater than costs, the implication of this finding is that Internet companies need a very large user base if they are to be in the black.
The second factor at work is that the networks of the Internet tend to be shaped according to what are known as power law distributions. Power law distributions — unlike normal, bell-shaped or Gaussian curves — exist where select few companies get a lot of users, site links or traffic and where most get almost none, relatively speaking of course. The implication of this finding is that most ad-based Internet companies will not have a large base of users and so cannot satisfy the high user base requirement of profitability as detailed in point one.
The interaction of these factors has two main implications for ad-based Internet companies. First, most of these companies will likely never reach profitability because they won’t ever get enough users. Instead, these firms need to focus again on what makes traditional businesses successful — providing a service people want — and get over the idea that free applications, services and widgets will always sell. Secondly, the way in which these two factors combine also suggests that Metcalfe’s Law, at least as it applies to companies aiming to use advertising to generate revenue, needs a corrective. Unlike in Metcalfe’s model where revenues increase exponentially as the costs of new connections accrue linearly, the evidence presented here suggests that costs of collecting and retaining users — who have agency — is surprisingly high and often not marked by economies of scale. The problem is users, as opposed to the hardware assumed in Metcalfe’s original formulation, have feelings and desires and can leave a network if it fails to give them what they want.
By way of a roadmap of what is to come, the first section elaborates upon the Internet’s advertising business model, showing how Metcalfe’s Law suggests that firms should get an exponential growth in revenue as their user base increase. The second section discusses how the cost of getting and retaining users tends to be surprising high, with many Internet companies failing to reach profitability even with hundreds of millions of active users. The third section discusses how the Internet tends to form what are known as power law distributions, which entail that most content platforms, social media sites and applications get few users (millions, maybe) while some get a great deal (>billion). The final section concludes with some ruminations on what the implications of the faulty premise of the Internet business model are for companies and the Internet economy.
Selling ads: The basic business model of most content platforms, social networking sites and mobile applications
While all firms have different revenue generation strategies (Obar and Wildman, 2015), many Internet companies aim to make money in a similar way. Their basic business model involves providing a free service and then collecting user-generated data. The data is then aggregated, analysed and sold to data brokers and advertisers, who combine data from a variety of sources to paint sometimes frighteningly accurate portraits of what people want, what they do and who they are (Schneier, 2015; Hampson and Jardine, 2016). The advertising model is certainly not new, as it has been employed by media companies since before the commercialization of the World Wide Web, but it is amplified by current technologies that expand the potential audience from the reach of a radio wave to a global scale (Napoli, 2003).
The basic free service/data collection/advertising business model is prolific, underpinning most content platforms, social networking sites and mobile applications — both large and small — in the Internet space. As Bruce Schneier put it at the Boston SOURCE Conference, “Surveillance is the business model of the Internet. We build systems that spy on people in exchange for services. Corporations call it marketing” (cited in Rashid, 2014). Even former U.S. President Obama has noted the importance of data collection to the basic functioning of many Internet companies, highlighting in a public speech on new privacy legislation in January of 2014, that “Corporations of all shapes and sizes track what you buy, store and analyze your data, and use it for commercial purposes” . Indeed, one estimate of the prevalence of this activity in the mobile application industry forecast that upwards of 94.5 percent of all applications will be free to users (with the obvious data collection hitch) by 2017 (Gartner, 2014).
One example of this business model in action is the Brightest Flashlight app. The app provides a bright flashlight by turning a user’s smartphone into a powerful source of illumination. In 2013, over 50 million people had downloaded and used the application. In exchange for free software that turned their smartphones into a flashlight, the users of the app turned over real-time location data on their movement to the company, who then sold it to generate revenue. “It was, in other words, a stalking device disguised as a flashlight” (Roberts, 2014).
The Brightest Flashlight app has some not so good company in this regard. One look into the proverbial ‘genome’ of the app industry found that, just as with the Brightest Flashlight app, 28 percent of apps in the Android store and 34 percent in the Apple App Store collect user-generated location data (Lookout, n.d.). Appthority’s 2014 App reputation report paints an even starker picture. According to their analysis, upwards of 95 percent of the top 200 freely available applications in the iOS and Android stores collect sensitive user-generated data. Some 70 percent of the top 200 free apps collect location data. Another 69 percent of applications access people’s social networks. Fifty-six percent collect uniquely identifying information on users, such as names and addresses. Thirty-one percent go through people’s address books and contact lists, while four percent collect details from people’s calendars (PR Newswire, 2014). The corporate philosophy seems to be you will never know what bits of information might be relevant when trying to sell a product.
Many apps also integrate together to play off of each other’s strengths and weaknesses in order to access more data so as to better target advertisements. This interconnection can effectively overcome efforts to keep users anonymous. One study, for example, found that with knowledge of only four mobile applications on an Android device, advertisers are able to reconstruct the real identity of a user 95 percent of the time. If an advertiser gets access to all the applications on a person’s device, they can predict the user’s identity with 99 percent accuracy. Even with information on just two applications, advertisers can uniquely identify users a startlingly high 75 percent of the time (Achara, et al., 2015). Since many applications are interconnected (Foursquare and Tinder both link to Facebook, for example), data sharing between apps is pretty common.
Like with mobile applications, the major social networking sites such as Twitter and Facebook earn the vast majority of their revenue from selling ads, as do content platforms such as Google, Yahoo! and others. As shown in Figure 1, Twitter routinely derives over 80 percent of its revenue from selling advertising. On average, from the start of 2012 to the end of 2016, 88 percent of Twitter’s revenue has come from selling advertisements. Facebook is much the same, if even more dependent upon connecting users with ad buyers. Over the quarters for which Facebook data is available, the social networking site has gleaned an averaged upwards of 94 percent of its income from advertising.
These corporate trends are repeated across the Internet ecosystem, with digital advertising revenues increasing significantly each year. As noted in the 2016 Internet advertising report, annual Internet advertising revenue in the United States alone has grown from US$16.9 billion in 2005 to US$72.5 billion in 2016. Advertising revenue from mobile applications, for its part, has experienced the even faster year-over-year growth, rising from just US$1.6 billion in 2011 to US$36.6 billion in 2016.
The precise form that Internet-based advertising takes varies considerably across platforms. Google has promoted results sold via its AdWords program based upon a complex “generalized second price” algorithm (Edelman, et al., 2005). Twitter uses ‘promoted’ tweets. Facebook has promoted stories and uses lead generation techniques to direct traffic to services. Some sites use banner advertisements; others use rich media that interacts with users. Some have pop-up videos; others use ad-based landing pages.
Not all advertising methods are equally popular, however. For instance, Figure 2 shows how the available US$59.6 billion in U.S.-based Internet advertising revenue in 2015 was divided up between the various forms of advertising (Interactive Advertising Bureau (IAB), 2016). As a proportion of allocated revenue, search ads are among the top of the buckets. Mobile ads make up the second largest bucket, but are actually an agglomeration of search (43 percent), display (53 percent) and other (four percent) advertising types (Interactive Advertising Bureau, 2016). Banner ads, digital videos, lead generation and classified placements trail behind.
The basic revenue proposition for ad-based Internet companies is that they can sell the attention of their users to advertisers. Decades ago, Robert Metcalfe helped pioneered the classic worth generation proposition of computer networks. The basic formula is simple: the systemic value of a network is equal to the number of connections squared — or, in more formal terms, systemic value = n2 (Metcalfe, 1995).
The intuition behind Metcalfe’s Law is a simple one. The cost of adding and retaining new connections accrue according to a linear function, while benefits (revenue) tend to aggregate exponentially. Figure 3 above depicts the relationship using simulated data. With few connections, the costs of the system outweigh the benefits. At a certain level (15 hypothetical devices), a critical mass point is reached and the company’s benefits start to outpace its costs from there on forward, leading to handsome profits.
At one large level of abstraction, network effects help to explain why Internet access rates have expanded with such rapidity. Since each new entrant increases the benefits of the network exponentially, the decision to join the system by later adopters becomes easier with each new user. Across 14 early adopting countries, for example, the network expanded in the early 1990s at a fairly slow rate. At around five percent Internet penetration, the pace with which new user joined the system rapidly increased as new user flooded online. The same sort of S-shaped pattern is apparent today in lower middle income countries (Hampson and Jardine, 2016).
Scaled down from the networks of the Internet as a whole, Internet firms tend to benefit from network effects, with corporate revenues fitting the n2 estimation of systemic value for networked systems remarkably well. Revisiting his law some years later, Robert Metcalfe (2013) took data on average monthly Facebook users and corporate revenues and plotted them in relation to the predicted n2 values. In line with the predicted values of his theory, he found that Facebook’s actual revenue grew in almost perfect lockstep with the predictions of his law . Other researchers replicated this same test using data from both Facebook and the Chinese social media giant Tencent, again finding a good fit between theory and evidence (Zhang, et al., 2015).
Certainly, these results do not imply that all users are created equally, as Briscoe, et al. (2006) imply with their n log(n) revision to Metcalfe’s classical formulation. The addition of some new users adds far less than the predicted exponential return in revenue. The inclusion of others adds even more than the exponential amount. For example, during 2015, Facebook’s average advertising revenue for the world was US$2.85 per user. Facebook’s ad revenue from the Asia-Pacific during this period was only US$1.36 per user, or roughly 48 percent less than the average. In contrast, Facebook earned an average of US$10.41 per user in the U.S. and Canada, or basically 3.6 times the worldwide average (Facebook, 2015). Clearly, some users contribute more to the revenues of a company than others. At the same time, the average contribution of a number of new users is really what matters when assessing a company’s revenue and there we see evidence that Metcalfe was largely right — at least on the revenue side of the equation.
However, when it comes to costs, the verdict is far more complex. Right now, the basic assumption built into Metcalfe’s Law — and at play in the business decisions of ad-based Internet companies — is that collecting and, often more importantly, keeping connections (users) is cheap. The problem is this assumption is wrong. Physical connections are easy to get and cheap to retain in Metcalfe’s original formulation because he is talking about hardware inputs (Ethernet cards). The trouble is that today it is human users that matter (at least until the Internet of Things really takes over), but users have preferences and can choose to leave a network if they are unhappy. This simple point affects the cost proposition behind Metcalfe’s Law.
An expensive proposition: The cost of getting and retaining users
The exponential aggregation of revenues as users increase implies that most Internet companies should accrue handsome profits. However, while Metcalfe’s Law seems to work well on the revenue side of the equation, the fact that users have opinions and preferences and can leave a network if they are dissatisfied entails that the cost structure at play for contemporary firms is far removed from Metcalfe’s initial vision. In many cases, costs tend to increase faster than users, which means that Internet companies do not benefit from economies of scale. This happens for the simple reason that companies need to satisfy more opinions to retain more users. As a result, costs are actually higher than anticipated and companies often need hundreds of millions of users, at bare a minimum, to just break even. Turning a profit usually requires more users still.
Take YouTube — the immensely popular video sharing site — as an example. YouTube has well over a billion users who view hundreds of millions of hours of online content. The number of people watching YouTube each day, for example, was up 40 percent year-over-year since March 2014, showing an increasingly active number of users (YouTube, n.d.). As a result of this trend, YouTube earned Google (now Alphabet), its parent company, some US$4 billion in revenue in 2014, up from US$3 billion in 2013. Taken on revenue terms alone, YouTube appears to be a success story.
These rosy numbers, however, mask an underlying economic reality that shows just how costly it can be to get and retain users. Despite pulling in US$4 billion in revenue in 2014, YouTube did not contribute to Google’s bottom line. As Rolfe Winkler (2015) put it, “while YouTube accounted for about six percent of Google’s overall sales last year, it didn’t contribute to earnings. After paying for content, and the equipment to deliver speedy videos, YouTube’s bottom line is ‘roughly break-even’.”
YouTube is not alone in this regard. Russia’s social media site, VK, is another example. VK’s user base is increasing, growing from 48.8 million users in December 2012 to 55.7 million users in December of 2014 (Statista, 2016b). As Metcalfe would expect, part and parcel of its expanding user base has been expanding revenue. In 2012, VK recorded US$172 million in total revenue, up 44.1 percent from the year before. Again, things seem to be going well on the revenue side of the profitability equation.
But growth in both users and revenue masks larger profitability problems due to runaway costs. From 2011 to 2012, for example, VK’s profits actually fell almost 95 percent to just US$1,000,000 (Hopkins, 2013). The company attributed the fall to a large one-time capital expenditure on new data centres. A move like this could actually reduce costs over the long term. However, the trend in declining profits at VK continued into 2013, even as its user base and revenue continued to grow. Marking another colossal drop in profits, VK earned a paltry US$225,000 over the course of the 2013, attributing the decline in profits this time around to a large increase in administrative costs (Trevisan, 2014).
Table 1: Users, value and profitability. User base in 2013 Corporate valuation in 2013 Profitable in 2013? Spotify 6 million US$3 billion No 128 million US$3.8 billion No 150 million US$1 billion No Fab 14 million US$1 billion No Snapchat 60 million US$800 million No Tumblr 300 million US$1.1 billion No Yelp 100 million US$4.4 billion No
The more you pick at the veneer covering many Internet companies, the longer becomes the list of platforms with a large base of users, impressive looking valuations and little-to-no profits. Table 1 above showcases the scope of the problem for a number of well-valued companies in 2013 (Luckerson, 2013). Many of the companies on the list, of course, have since been sold to larger companies that have incorporated them into a broader business plan, but the point stands. The cost of gaining and retaining users is often greater than the revenues generated by even a very large — sometimes as high as 300 million — base of users.
The quarterly earning reports of publicly traded social media sites, such as Twitter, provide additional insight into the precise cost of getting and retaining users. Twitter, for example, seems a success in terms of market capitalization, revenue and, to a certain extent, user accumulation. In June of 2017, for example, Twitter had over a US$13 billion market capitalization. Between the first quarter of 2012 and the latest quarter of 2016, Twitter’s user base increased from 138 to roughly 328 million users. Over this same period, Twitter’s revenues have also gone up considerably, rising from US$54 million to US$717 million (Twitter, 2016). Still, despite these trappings of success, the company’s GAAP revenue has been consistently less than its costs, meaning it has yet to turn a profit (Twitter, 2016) .
For companies like Twitter that rely upon the advertising business model, costs get inflated due to the combination of the low barriers to entry in the Internet space and the fickle nature of Internet users. The basic product that these Internet companies sell is not the content that they produce, but the attention of the users that frequent their services (Napoli, 2003). Yet, users are hard to keep when preferences evolve and there is always a new, more tantalizing rival just on the horizon.
The first problem is that Internet users can behave fickly as their preferences evolve over time. The fickle nature of Internet user behavior often produces Internet fads. Such fads, as in the physical world, occur when an Internet meme, platform or company gains a short 15 minutes of fame, but then crashes and burns (Bauckhage, et al., 2013). A clear cut case of faddish users can be found in the launch, rapid spread and quick decline of Pokémon Go. According to analysis done by comScore (Frommer, 2017), Pokémon Go launched on 6 July 2016. With remarkable alacrity, use of the augmented reality game exploded. By 13 July, some seven days later, there were 28.5 million unique daily Pokémon Go users. Yet the game’s success story was very short lived. Within the year, daily usage rates had plummeted by some 80 percent, leaving the average unique visitor rate at a paltry five million users per day (Arif, 2017). Despite its massive initial appeal, Pokémon Go grew rapidly stale and fickle users voted with their proverbial feet moving on to other things.
Faddish behaviour of this sort actually marks a surprisingly large portion of user interaction with all mobile applications. A lot of people download an app, but then quickly lose interest. For free applications (those most likely to be based upon an advertising business model), for example, the vast majority of downloaded apps are quickly left fallow. One study by Pinch Media found that only about 20 percent of users return to applications that they have downloaded after the first day. The longer out you go from the point of download, the worse the return rate becomes. By 30 days, people only return to downloaded apps a little under five percent of the time. The numbers keep getting worse from there for app developers, as the percentage of returning users asymptotically approaches zero after 90 days (Schonfeld, 2009). The point is that even if an app has been downloaded, active use of the service is likely a mere shadow of these numbers because users are fickle and prone to faddish behavior.
Implicit in the idea of fickle user behavior is the notion of a user stampede. Stampedes occur when users of a service cascade with tremendous rapidity either toward or away from an online service. Pokémon Go’s rapid expansion is an example of a positive stampede, with daily usage rates jumping by 28.5 million people in the span of seven days. The classic example of a damaging stampede is the rapid decline of Myspace. The former social media giant reached its peak user base of 75.9 million monthly average users in December 2008. Only a couple of years later in May 2011, the number of monthly active users declined to 34.8 million people (a negative change of 48.8 percent). Based as it was upon the data-driven advertising business model, Myspace’s revenues took a corresponding hit. In 2009, Myspace brought in US$470 million in revenue. By 2011, it was projected that its revenue would decline to US$184 million (Gillette, 2011).
The Myspace example obviously hails from a period before smartphones and the age of near-constant interconnection. The implication of these differences for stampede behaviour is not trivial. The effect of smartphones and constant connectivity probably make extreme outcomes more likely. On the positive side, smartphones and constant connectivity can keep users connected to their platform of choice, making it harder to topple an Internet company. At the same time, constant, real-time connectivity can also increase the pace with which stampedes can occur, leading to faster, more pronounced changes. It is like going viral in reverse.
Fickle users and evolving preferences would always suggest that Internet companies need to invest in product innovation in order to keep their users happy, but the problem is compounded by the Internet’s low barriers of entry. Because starting up a new Internet company is fairly cheap, established Internet companies are constantly challenged by new and often highly innovative rivals. Across various industries, the common disruptive effect of the Internet has been to reduce the costs of entry, creating rivals where previously there were none. The effects are visible in industries as variable as Internet search engines (Gandal, 2001), the music industry (Lewis, et al., 2005), healthcare provision (Estrin and Sim, 2010), and the video streaming and e-mail industries (Gallaugher, 2016). Of course, not all forms of cost have declined significantly due to the Internet, but overall it is now a lot cheaper and easier to launch a new company that could displace a gigantic rival than it previously. This trend leads to a situation, as Leslie Daigle  puts it, were there are potentially “no permanent favourites” in the Internet ecosystem.
Myspace is again a perfect case exemplifying this process. Like the fall of most companies, Myspace’s demise was the result of a number of factors (Lee, 2011). But one of the primary culprits was the company’s lack of constant innovation. Connie Chan, an analyst with Chess Media Group, for example, argued in 2011 that “Myspace was created by people in the entertainment industry, not by technology gurus, therefore they could not innovate at the pace that they needed to compete” (cited in Lee, 2011). Likewise, Facebook’s former president Sean Parker argued that Myspace thought that they were invulnerable and did not properly recognize that low barriers in the online ecosystem rapidly produce strategic rivals. In such a context, only constant innovation provides any sort of short run inoculation against user stampedes and corporate collapse. As he put it in an interview with the Huffington Post in 2011, Myspace’s staff “weren’t successful in iterating and evolving the product enough, it was basically this junk heap of bad design that persisted for many, many years. There was a period of time where, if they had just copied Facebook rapidly, I think they would have been Facebook. The network effects, the scale effects were enormous. There was so much power there” (cited in Lee, 2011).
When a world of “no permanent favourites” is combined with the fickle nature of Internet users, the end product is a realm where even established companies need to invest heavily in corporate and technological innovation just to stay current. These investments involve the near constant reinvention of platforms, the optimization of algorithms and the expansion of provided services. These efforts are bound to be costly, making it hard for ad-based companies to turn a profit.
Twitter and Facebook’s investment patterns support the idea that Internet companies need to engage in a nearly constant updating, adaptation and innovation of their services if they are to remain competitive. Twitter’s financial records provide some useful insights. The company spends a lot on infrastructure to provide users with the best possible experience. Indeed, historically, most of Twitters cost of revenue (which averages about 27 percent of Twitter’s overall costs) “consists primarily of data center costs related to our co-located facilities, which include lease and hosting costs, related support and maintenance costs and energy and bandwidth costs, as well as depreciation of our servers and networking equipment, and personnel-related costs, including salaries, benefits and stock-based compensation, for our operations teams” .
Additionally, Twitter, like all companies, also needs to get the word out in order to convince late joiners and digital holdouts that their platform is worth using. These essential sales and marketing efforts are quite costly and consume around 27 percent of the company’s overall costs from the first quarter of 2012 to the last quarter of 2016. As Anthony Noto, Twitter’s CFO, put it in an interview in 2015:
We’ve invested a lot of time in understanding users’ attitudes and behaviors and preferences and the choices that they have on a global basis. ... In addition to that market research, we spent a fair amount of time around our product roadmap and how we can marry making the product easier to use with marketing, with content choices and media choices and communication choices. And our goal is to bring the combination of those functional areas together to answer the question both from an awareness standpoint of why they should use Twitter (cited in Peterson, 2015).
Twitter, again like other ad-based Internet companies, also needs to continually invest in research and development in order to reinvent themselves at the margins. In its 2014 Annual report, Twitter noted that “We intend to continue to invest in research and development to improve our products and services for users and advertisers and to grow our active user base in order to address the competitive challenges in our industry” . These activities, which can range from tweaking an algorithm used to display content to adding ‘moments’, are essential if the platform is not to become a stale taste in the mouths of users. However, staying current is expensive, costing around 34 percent of Twitter’s total costs from the beginning of 2012 to the end to 2016.
In absolute terms, as shown in Figure 4, Twitter’s costs have increased along with its user base from the beginning of 2012 to the end of 2016. The close, positive relationship between Twitter’s user base and its absolute costs is intuitive. Overall, managing the online experience of more users should cost more than managing less users, everything else being equal.
But an interesting trouble for social media companies that aim to monetize user’s data (and quite possibly for other Internet companies as well) is that Twitter’s costs as a proportion of its user base is actually increasing as well. This trend essentially means that costs are aggregating faster than the rate of new user accumulation. As shown in Figure 5, it cost Twitter US$54,952 per 100,000 users to maintain and enlarge the company’s user base in the beginning of 2012. By the end of 2016, these same efforts cost Twitter US$262,449 per 100,000 users. Twitter, in other words, is not only seeing increasing revenues and rising absolute costs, but also growing per user costs.
Twitter is not alone in this regard. Facebook — which is a profitable company — suffers from the same problem of expanding cost per user. From the beginning of 2013 to the end of 2016, Facebook’s user base expanded from 1.1 to roughly 1.8 billion active monthly users. According to Facebook’s financial statements for this same period, Facebook has been paying more and more to retain, as a proportion of its user base, this active following. At the beginning of 2013, it cost Facebook US$97,748 per 100,000 users. By the end of 2016, Facebook’s cost per 100,000 users had increased to US$232,065. This growing per user cost amounts to a 137 percent increase in just four years. As Figure 6 below highlights, growing costs per 100,000 users is a very real trend for Facebook, just as for Twitter.
The trend of increasing costs to scale is not as prevalent among all online social media companies. Weibo, China’s version of Twitter, differs from both Twitter and Facebook in that it does have marginally declining costs per 100,000 users, as shown in Figure 7. Weibo’s base of users has expanded from 107 million users in the first quarter of 2013 to roughly 313 million users in the end of 2016. Its average cost per 100,000 users has exhibited a marginal decline over this same window of time.
One of these two trends is the rule and one is the exception: either Facebook and Twitter are the norm or Weibo is. When compared to Facebook and Twitter, Weibo has some singular advantages that could help to reduce its cost of getting and retaining users. As a result, Weibo might be the outlier and Facebook and Twitter may be closer to the norm for online social media companies.
First, Weibo’s user base is — relative to the two other globe-striding giants — very homogenous. A homogeneous user base allows Weibo to focus its advertising and marketing efforts in ways that are simply not possible most ad-based Internet companies, which more explicitly aim to have users from around the globe. There are cost saving advantages here. Developing a single country (or single language) marketing plan is far less expensive than having to market a service in 190 different locations.
China’s business environment also probably drives Weibo’s declining cost per user. The list of banned Web sites in China, particularly potential competitors to Weibo such as Facebook and Twitter, continues to expand. Although there are many social media platforms within China that compete for users, the normal worry that a different platform will spring up from the grass and steal users is less acute. In short, Weibo’s cost of collecting and retaining users is plausibly lower than for typical social media firms that aim to do business globally.
Lastly, since China has a domestic Internet using population of around 650–700 million people, Weibo has room to expand its user base without needing to invest in costly platform or infrastructure expansions that span borders and run afoul of multiple regulatory jurisdictions. Facebook and Twitter, in contrast, need to build (or rent) infrastructure around the world and comply with the domestic regulatory environment of each unique jurisdiction, although this should not minimize the regulatory and compliance barriers within China itself.
One point of contrast illustrates the disjunction well. Unlike Weibo’s relatively cost-friendly environment, Twitter’s 2014 Annual report highlights the challenges of its business environment. While Weibo targets a single population, in a low competition environment and has room to expand domestically, Twitter faces “challenges in increasing our advertising revenue internationally, including local competition, differences in advertiser demand, differences in the digital advertising market and conventions, and differences in the manner in which Twitter is accessed and used internationally. We face competition from well-established competitors in certain international markets” .
Despite its myriad business environment advantages, Weibo’s cost per 100,000 users is relatively flat (a couple of bad quarters would turn the trend positive). Indeed, while the overall trend is slightly negative, Weibo’s costs have certainly increased. In the first quarter of 2013, it cost Weibo US$39,416 per 100,000 users. By the last quarter of 2016, the cost for the same number of users had increased to some US$47,632, which amounts to an increase of roughly 21 percent (Weibo, 2016). Plausibly, then, a less favorable business environment (or expansion of Weibo outside of China into international markets marked by more competition) could quite probably lead to a similar pattern of increasing cost to scale as the one exhibited by Facebook and Twitter.
Based upon the evidence detailed in this section, it is fair to say that, for many Internet companies, profitability based upon advertising revenues is not as common as people might otherwise suspect when looking at just a company’s number of active users, its revenues, or its corporate valuation. The problem is that attracting and retaining users is fairly costly because users are fickle and often have readily available alternatives due to low barriers to entry. These realities give rise to a situation where, at least in the case of social media services detailed here, the activities of Internet companies might not be marked by economies of scale. Companies can still earn a profit, but to do so requires a far larger base of users than is commonly thought.
This large user base requirement of profitability has additional implications. Getting a large user base would not be a trouble if high density clusters were a common occurrence on the networks of the Internet. Unfortunately, the final challenge for ad-based Internet companies is that the networks of the Internet do not form large user clusters very often. It is not in their nature.
80/20: Understanding the shape of networks
The basic advertising business model of most Internet companies poses some serious concerns for the privacy of users (Pasquale, 2015; Taylor, 2016). The access that companies have to user-generated data is prolific and often largely unregulated. The privacy implications of corporate data practices have already generated some all-too-real economic ramifications by eroding people’s trust and changing how they behave online (Hampson and Jardine, 2016). For this reason alone, companies should consider revising their advertising business model in order to better protect people’s rights. But there is also another, far more hard-nosed reason why the basic advertising business model of so many content platforms, social media sites and mobile applications often does not work: it does not jive with the shape of networks, which tend to form what are known as power law distributions. Companies with exponential revenue and growing cost per user could still be profitable if they were able to collect a very large user base, but networks do not tend to work that way.
A lot of the world is organized into normal Gaussian distributions that form a familiar bell-shaped curve. There is a lot of regularity and predictability in bell-shaped distributions. The mean of normal distributions tends to be stable and the variance of the sample tends to be limited. Human height is a classic example. If the average height of the male population in the United States is around, say, 5’10, then we know with a high level of mathematical certainty that roughly 68 percent of the population falls within one standard deviation from this mean. We also know that 95 percent of the population falls within two standard deviations and that over 99 percent fall within three standard deviations.
Power law distributions are different. They are characterised by a long tail and huge outliers, which make the concepts of mean and variance less useful. You could calculate both statistics, of course, but they are not likely to tell you very much as the inclusion of new observations can potentially cause both to change significantly (Boisot and McKelvey, 2007). Power law distributions exist all over both the social and the natural worlds. The size of cities, firms and earthquakes all form power law distributions, as do academic citations and the frequency of words used in English, among numerous others (McKelvey and Andriani, 2005; Newman, 2005).
Comparing a normal and a power law distribution side-by-side, as done in Figure 8, provides some insight into how observations are distributed in the two situations. In normal distributions, most observations cluster in the middle of the range. The distribution then tapers off at a relatively equal rate both above and below the mean. In the case of power law distributions, in contrast, the bulk of the observations are close to zero (origin) and the frequency of observations diminishes very quickly as the size of the event increases. The implication is that most earthquakes are very small but a few are hugely devastating. The implication is that most academic papers are never cited but some receive thousands upon thousands of citations. The implication is that a select few words in the English language are used all the time and many are almost never used. There are, in other words, extreme outliers.
Outliers drive so much of what happens in power law distributions that the whole category of phenomenon governed by power laws are characterised by what is known as the Pareto Principle — or roughly the idea that 20 percent of the observations produce 80 percent of the outcome (Newman, 2005; Barabási, 2016). The clustering of income is a typical example. In many societies, a small fraction of the population earns most of the available income, potentially along the lines of 20 percent of the people earning 80 percent or more of the wealth. Depending upon the value of the exponent for the actual data in question, the severity of the 80/20 rule will be more or less pronounced. Regardless, all power law distributions have the same basic structure where a large fraction of the outcome (wealth; earthquake size; academic citations) are explained by a very small number of observations.
The networks of the Internet form power law distributions. A pioneering study from 1998, for example, looked at multiple instances of the Internet over a six-month period in 1997–98 and found that the network’s interconnections tended to form power law structures (Faloutsos, et al., 1999). Another study looking at the cross linkages between 260,000 websites found that most sites had very few links from other sites, while a select few had a great many, once again forming a power law distribution (Adamic and Huberman, 2000). As Steven Johnson and his colleagues put it, “Instead of following a normal distribution, these networks often follow a scale-free distribution of links [power law distribution] where a few participants have an extremely high number of relationships to other participants and most participants have very few” . In recent years, power law distributions have also been found in social media sites such as Orkut, YouTube and Flickr (Kumar, et al., 2006; Mislove, et al., 2007).
Although networks can form power law distributions due to a combination of mechanisms (Johnson, et al., 2014), one of the most common processes involves what is known as preferential attachment (Barabási and Albert, 1999). The easiest way to think about it is to ponder why people join — or stay on — Facebook. At one level, people use these sites because they satisfied person’s desire to make new friends and connect with old acquaintances (Raacke and Bonds-Raacke, 2008). Layered onto that utilitarian motive is a more psychological combination of ‘the need to belong’ and ‘the need for self-presentation’ (Nadkarn and Hofmann, 2012).
Underlying all of these motivations, however, is the idea that people use Facebook because everyone else does so too. Locating old friends and making new ones in digital forums can only happen if people actually use the same platform. In a similar vein, both a sense of belonging and the reach of a person’s self-presentation are mediated by the number of users on the site. Belonging and self-presentation are more pronounced when many other people use the site and are more muted when the site is infrequently used. Other parts of the Internet form power laws for the same reason. The majority of Web sites are designed to be viewed by the largest number of people possible, so each tries to interconnect new sites with the most popular nodes in the network, reinforcing early success and creating extreme outliers.
This sort of extreme clustering is exactly what happens within the inner working of the Internet ecosystem. Alexa (2017) data highlights the trend well. Taking the 40 most popular sites and the 40 least popular sites among the top 500 and plotting a histogram based upon the number of Web site interlinks results is a stark power law curve, as shown in Figure 9. A total of 91 percent of the Web sites fall within the first bin, while the remaining Web sites are scattered across the remaining groups. True to the form of pronounced power law distributions, the variance in the data is massive. The Web site among the top 500 with the fewest interlinks has only 32 incoming links. The Web site with the most has 6,534,681. Moreover, in typical power law form, the mean number of interlinks in the sample is 312,224. The median, the number splitting the observations in the sample into two, is only 22,238. In a normally distributed world, the mean and the median should be pretty well equal. Here, they are clearly far, far different. If the analysis was to expand to all Web sites in place of just the most popular 500 as ranked by Alexa, the power law distribution would likely become even more pronounced. Indeed, it is not outside of the realm of possibility that the modal number of interlinks is actually zero. While interconnections are not necessarily users, the point that networks produce extreme clusters is well supported by this simple evidence.
When users — or interlinks — cluster in this sort of way, the expectation would be that Internet advertising revenue should also be highly skewed towards a few firms. The Interactive Advertising Bureau’s Internet advertising report  records this exact trend in online ad revenue within the United States. As shown in Figure 10, during the last quarter of 2016, 10 firms controlled 73 percent of online advertising revenue. The next 15 firms (11–25 of the largest Internet advertisers) controlled an additional 10 percent. The millions of other Web sites and firms in the U.S.’s Internet ecosystem controlled only 17 percent of fourth quarter advertising revenue.
In sum, most platforms probably won’t get many users. The way in which the networks of the Internet structure themselves effectively limits the potential number of platforms that will obtain the large clustering of users that are essential for the effective operation of the ad-based business model used by so many companies. Most firms that rely upon an ad-based approach, therefore, will likely fail over the long term.
Conclusions: The broken premise of the advertising business model
The basic advertising business model of many Internet companies presents an alluring story. Supposedly, a simple line runs from start-up to profits: provide a free service, collect user data, and then sell that data to advertisers and data brokers. With revenue coming in as an exponential function of a company’s user base, as Metcalfe proposed, profit should easily come next. However, despite this rosy narrative, many well-established Internet companies perform poorly and only one-in-one-ten-thousand mobile applications is likely to actually be profitable (Gartner, 2014).
As Shakespeare noted in Hamlet (Act 1, Scene 4), ‘something is rotten in the state of Denmark.’ If the basic advertising business model that is central to so many content platforms, social media sites and mobile applications is so effective, why do so many Internet companies perform so badly? The basic reason is that while revenue growth does tend to increase exponentially as new users join a platform, the cost of collecting and retaining users tends to increase faster than the rate of user growth, which inhibits the creation of economies of scale. Costs increase so rapidly because users are fickle, as the Pokémon Go and Myspace cases reveal. As a result, companies need to invest heavily in innovation or be taken out by rivals who crop-up with routine frequency in a world with extraordinarily low barriers to entry.
In the end, revenue growth can outpace the expansion of cost, but the critical mass point of profitability requires a very large base of users, often reaching well into the hundreds of millions if not billions. However, since the Internet tends to form power law distributions, only a few outliers can ever attract and retain a sufficiently large numbers of users to generate a profit. By implication, most companies, be they Web sites, social media applications, content intermediaries or mobile applications, should perform badly, and, over the long run, many — indeed most — should fail.
About the author
Eric Jardine is Assistant Professor in the Department of Political Science at Virginia Tech.
E-mail: ejardine [at] vt [dot] edu
My thanks go out to David Kempthorne, who read and commented on a number of earlier drafts, as well as Jonathan Obar, who did the same. Mistakes that remain are my own.
1. Cited in Segal, 2016, p. 168.
2. Metcalfe, 2013, p. 30.
3. I use the Generally Accepted Accounting Principles (GAAP) measure of revenue as it is a common standard that applies to all companies. As a result, it is useful for comparative purposes whereas non-GAAP income tends to be altered for company-specific reasons that would hinder effective comparisons.
4. Daigle, 2015, pp. 8–9.
5. Twitter, 2014, p. 52.
6. Twitter, 2014, p. 48.
8. Johnson, et al., 2014, p. 796.
9. Interactive Advertising Bureau, 2017, internet advertising revenue report, p. 9.
Jagdish Prasad Achara, Gergely Acs and Claude Castelluccia, 2015. “On the unicity of smartphone applications,” arXiv (29 October), at https://arxiv.org/pdf/1507.07851v2.pdf, accessed 15 June 2017.
Lada A. Adamic and Bernardo A. Huberman, 2000. “Power-law distribution of the World Wide Web,” Science, volume 287, number 5461 (24 March), p. 2115a, and at http://science.sciencemag.org/content/sci/287/5461/2115.full.pdf, accessed 15 June 2017.
Alexa, 2017, “The top 500 sites on the Web,” at http://www.alexa.com/topsites, accessed 15 June 2017.
Shabana Arif, 2017, “The number of Pokémon Go players still logging in every day has plummeted by over 80%,” VG24/7 (3 April), at https://www.vg247.com/2017/04/03/the-number-of-pokemon-go-players-still-logging-in-every-day-has-plummeted-by-over-80/, accessed 15 June 2017.
Albert-László Barabási, 2016. Network science. New York: Cambridge University Press.
Albert-László Barabási and Réka Albert, 1999. “Emergence of scaling in random networks,” Science, volume 286, number 5439 (15 October), pp. 509–512.
doi: http://dx.doi.org/10.1126/science.286.5439.509, accessed 15 June 2017.
Christian Bauckhage, Kristian Kersting and Fabian Hadiji, 2013. “Mathematical models of fads explain the temporal dynamics of Internet memes,” at http://sfb876.tu-dortmund.de/PublicPublicationFiles/bauckhage_etal_2013a.pdf, accessed 15 June 2017.
Max Boisot and Bill McKelvey, 2007. “Extreme events, power laws, and adaptation: Towards an econophysics of organization,” Academy of Management Proceedings, at http://proceedings.aom.org/content/2007/1/1.65.short, accessed 15 June 2017.
Bob Briscoe, Andrew Odlyzko and Benjamin Tilly, 2006. “Metcalfe’s Law is wrong,” IEEE Spectrum, volume 47, number 3, pp. 34–39, and at http://spectrum.ieee.org/computing/networks/metcalfes-law-is-wrong, accessed 15 June 2017.
doi: http://dx.doi.org/10.1109/MSPEC.2006.1653003, accessed 15 June 2017.
Josh Constine, 2016. “Facebook climbs to 1.59 billion users and crushes Q4 estimates with $5.8B revenue,” TechCrunch (27 January), at http://techcrunch.com/2016/01/27/facebook-earnings-q4-2015/, accessed 15 June 2017.
Leslie Daigle, 2015. “On the nature of the Internet,” Global Commission on Internet Governance, Paper Series, number 7 (16 March), at https://www.ourinternet.org/research/nature-internet, accessed 15 June 2017.
Jillian D’Onfro, 2016. “Google beats, stock soars,” Business Insider (1 February), at http://www.businessinsider.com/alphabet-google-q4-2015-earnings-2016-2, accessed 15 June 2017.
Benjamin Edelman, Michael Ostrovsky and Michael Schwarz, 2005. “Internet advertising and the generalized second price auction: Selling billions of dollars worth of keywords,” NBER Working Paper Series, number 11765, at http://www.nber.org/papers/w11765, accessed 15 June 2017.
Deborah Estrin and Ida Sim, 2010. “Open mHealth architecture: An engine for health care innovation,” Science, volume 330, number 6005 (5 November), pp. 759–760.
doi: http://dx.doi.org/10.1126/science.1196187, accessed 15 June 2017.
Facebook, 2015. “Investor relations: Quarterly results,” at https://investor.fb.com/financials/?section=secfilings, accessed 29 June 2017.
Michalis Faloutsos, Petros Faloutsos and Christos Faloutsos, 1999. “On power-law relationships of the Internet topology,” SIGCOMM ’99: Proceedings of the Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, pp. 251–262.
doi: http://dx.doi.org/10.1145/316194.316229, accessed 15 June 2017.
Dan Frommer, 2017, “Millions of Americans still play Pokémon Go every day. Popularity peaked last July, but its long tail is still going” (23 March), at https://www.recode.net/2017/3/23/15039626/pokemon-go-daily-users-chart, accessed 15 June 2017.
John Gallaugher, 2016. Information systems: A manager’s guide to harnessing technology. Nyack, N.Y.: Flat World Knowledge, and at http://catalog.flatworldknowledge.com/, accessed 15 June 2017.
Niel Gandal, 2001. “The dynamics of competition in the Internet search engine market,” International Journal of Industrial Organization, volume 19, number 7, pp. 1,103–1,117.
doi: https://doi.org/10.1016/S0167-7187(01)00065-0, accessed 15 June 2017.
Gartner, 2014. “Gartner says less than 0.01 percent of consumer mobile apps will be considered a financial success by their developers through 2018” (13 January), at http://www.gartner.com/newsroom/id/2648515, accessed 15 June 2017.
Fen Osler Hampson and Eric Jardine, 2016. Look who’s watching: Surveillance, treachery and trust online. Waterloo: Centre for International Governance Innovation, and at https://www.cigionline.org/publications/look-whos-watching-surveillance-treachery-and-trust-online, accessed 15 June 2017.
Ben Hopkins, 2013. “Vkontakte’s annual revenue rises, but profits fall,” Rusbase.com (29 April), at http://archive.rusbase.com/, accessed 15 June 2017.
Interactive Advertising Bureau, 2017. “IAB Internet advertising revenue report: 2016 full year results” (April), at https://www.iab.com/wp-content/uploads/2016/04/IAB_Internet_Advertising_Revenue_Report_FY_2016.pdf, accessed 15 June 2017.
Interactive Advertising Bureau, 2016. “IAB Internet advertising revenue report: 2015 full year results” (April), at http://www.iab.com/wp-content/uploads/2016/04/IAB-Internet-Advertising-Revenue-Report-FY-2015.pdf, accessed 15 June 2017.
Steven L. Johnson, Samer Faraj and Srinivas Kudaravalli, 2014. “Emergence of power kaws in online communities: The role of social mechanisms and preferential attachment,” MIS Quarterly, volume 38, number 3, pp. 795–808.
Amy Lee, 2011, “Myspace collapse: How the social network fell apart,” Huffington Post: Tech (30 June), at http://www.huffingtonpost.com/2011/06/30/how-myspace-fell-apart_n_887853.html, accessed 15 June 2017.
Gerard J. Lewis, Gary Graham and Glenn Hardaker, 2005. “Evaluating The impact of the Internet on barriers to entry in the music industry,” Supply Chain Management, volume 10, number 5, pp. 349–356.
doi: https://doi.org/10.1108/13598540510624179, accessed 15 June 2017.
Lookout, n.d. “App genome project,” at https://www.lookout.com/insights, accessed 15 June 2017.
Victor Luckerson, 2013. “These 8 Internet companies are worth over $1 billion — But they haven’t made a dime,” Time (29 October), http://business.time.com/2013/10/29/these-8-internet-companies-are-worth-over-1-billion-but-they-havent-made-a-dime/, accessed 15 June 2017.
Ravi Kumar, Jasmine Novak and Andrew Tomkins, 2006. “Structure and evolution of online social networks,” KDD ’06: Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 611–617.
doi: https://doi.org/10.1145/1150402.1150476, accessed 15 June 2017.
Bill McKelvey and Pierpaolo Andriani, 2005. “Why Gaussian statistics are mostly wrong for strategic organization,” Strategic Organization, volume 3, number 2, pp. 219–228.
doi: https://doi.org/10.1177/1476127005052700, accessed 15 June 2017.
Robert Metcalfe, 2013. “Metcalfe’s Law after 40 years of Ethernet,” Computer, volume 46, number 12, pp. 26–31.
doi: https://doi.org/10.1109/MC.2013.374, accessed 15 June 2017.
Robert Metcalfe, 1995. “Metcalfe’s Law: A network becomes more valuable as it reaches more users,” Infoworld, number 17 (2 October), p. 53.
Alan Mislove, Massimiliano Marcon, Krishna P. Gummadi, Peter Druschel and Bobby Bhattacharjee, 2007. “Measurement and analysis of online social networks,” IMC ’07: Proceedings of the Seventh ACM SIGCOMM Conference on Internet Measurement, pp. 29–42, and at http://conferences.sigcomm.org/imc/2007/papers/imc170.pdf, accessed 15 June 2017.
Ashwini Nadkarn and Stefan G. Hofmann, 2012. “Why do people use Facebook?” Personality and Individual Differences, volume 52, number 3, 243–249.
doi: https://doi.org/10.1016/j.paid.2011.11.007, accessed 15 June 2017.
Philip M. Napoli, 2003. Audience economics: Media institutions and the audience marketplace. New York: Columbia University Press.
M.E.J. Newman, 2005. “Power laws, Pareto distributions and Zipf’s Law,” Contemporary Physics, volume 46, number 5, pp. 323–351.
doi: http://dx.doi.org/10.1080/00107510500052444, accessed 15 June 2017.
Jonathan A. Obar and Steven S. Wildman, 2015. “Social media definition and the governance challenge: An introduction to the special issue,” Telecommunications Policy, volume 39, number 9, pp. 745–750.
doi: https://doi.org/10.1016/j.telpol.2015.07.014, accessed 15 June 2017.
Frank Pasquale, 2015. The black box society: The secret algorithms that control money and information. Cambridge, Mass.: Harvard University Press.
Tim Peterson, 2015. “Q&A: Twitter CFO Anthony Noto, what are you doing as the company’s top marketer?” Advertising Age (28 August), at http://adage.com/article/digital/twitter-s-cfo-anthony-noto-explains/300131/, accessed 15 June 2017.
PR Newswire, 2014. ”Appthority exposes security and privacy risks behind top 400 mobile apps” (18 February), at http://www.prnewswire.com/news-releases/appthority-exposes-security-and-privacy-risks-behind-top-400-mobile-apps-245968521.html, accessed 15 June 2017.
John Raacke and Jennifer Bonds-Raacke, 2008. “MySpace and Facebook: Applying the uses and gratifications theory to exploring friend-networking sites,” CyberPsychology & Behavior, volume 11, number 2, pp. 169–174.
doi: https://doi.org/10.1089/cpb.2007.0056, accessed 15 June 2017.
Fahmida Y. Rashid, 2014. “Surveillance is the business model of the Internet: Bruce Schneier,” Security Week (9 April), at http://www.securityweek.com/surveillance-business-model-internet-bruce-schneier, accessed 15 June 2017.
Jeff Roberts, 2014. “‘Brightest Flashlight’ Android app exposes location of 50 million people, but FTC imposes no fine,” Gigaom (9 April), at https://gigaom.com/2014/04/09/brightest-flashlight-android-app-disclosed-location-of-50-million-people-but-ftc-imposes-no-fine/, accessed 15 June 2017.
Erick Schonfeld, 2009. “Pinch Media data shows the average shelf life of an iPhone app is less than 30 days,” TechCrunch (19 February), at http://techcrunch.com/2009/02/19/pinch-media-data-shows-the-average-shelf-life-of-an-iphone-app-is-less-than-30-days/, accessed 15 June 2017.
Adam Segal, 2016. The hacked world order: How nations fight, trade, maneuver, and manipulate in the digital age. New York: Council on Foreign Relations Books.
Statista, 2016a. “Number of newly developed applications/games submitted for release to the iTunes App Store from 2012 to 2016,” at http://www.statista.com/statistics/258160/number-of-new-apps-submitted-to-the-itunes-store-per-month/, accessed 15 June 2017.
Statista, 2016b. “Number of monthly active VKontakte users via desktop connections in Russia from December 2012 to December 2014 (in millions),” at http://www.statista.com/statistics/425423/number-of-monthly-active-vkontakte-users/, accessed 15 June 2017.
Emily Taylor, 2016. “The privatization of human rights: Illusions of consent, automation and neutrality,” Global Commission On Internet Governance, Paper Series, number 24 (26 January), at https://www.ourinternet.org/research/privatization-human-rights-illusions-consent-automation-and-neutrality, accessed 15 June 2017.
Zoran Trevisan, 2014. “Vkontakte suffers massive drop in profit,” Rusbase.com, at http://archive.rusbase.com/, accessed 15 June 2017.
Twitter, 2016. “Investor relations: Quarterly results,” at https://investor.twitterinc.com/results.cfm, accessed 15 June 2017.
Twitter, 2014. “Annual report,” at https://investor.twitterinc.com/annuals-proxies.cfm, accessed 15 June 2017.
Weibo, 2016. “Investor relations: Quarterly results,” at http://ir.weibo.com/phoenix.zhtml?c=253076&p=irol-reportsother, accessed 29 June 2017.
Rolf Winkler, 2015. “YouTube: 1 billion viewers, no profit,” Wall Street Journal (25 February), at http://www.wsj.com/articles/viewers-dont-add-up-to-profit-for-youtube-1424897967, accessed 15 June 2017.
YouTube, n.d. “Statistics,” at https://www.youtube.com/yt/press/statistics.html, accessed 15 June 2017.
Xing-Zhou Zhang, Jing-Jie Liu and Zhi-Wei Xu, 2015. “Tencent and Facebook data validate Metcalfe’s Law,” Journal of Computer Science and Technology, volume 30, number 2, pp. 246–251.
doi: https://doi.org/10.1007/s11390-015-1518-1, accessed 15 June 2017.
Received 1 November 2016; revised 2 June 2017; accepted 12 June 2017.
“‘Something is rotten in the state of Denmark.’ Why the Internet’s advertising business model is broken“ by Eric Jardine is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
“Something is rotten in the state of Denmark.” Why the Internet’s advertising business model is broken
by Eric Jardine.
First Monday, Volume 22, Number 7 - 3 July 2017