Viktor Mayer-Schönberger & Thomas Ramge. Foreign Affairs. Volume 97, Issue 5, September/October 2018.
Over the last two decades, a few technology giants have come to dominate digital markets. Google performs about nine out of every ten Internet searches worldwide. Facebook, the world’s leading social media platform, has well over two billion users. Together, the two companies have seized well over half of the online advertising market. Apple, originally a computer manufacturer, now runs the world’s largest mobile app store in terms of revenue, with about 80 percent of the market, and the second-largest music streaming business, approaching a third of the market. And Amazon captures close to every other dollar spent online in the United States. These companies are what the economist David Autor calls “superstar firms,” able to gain huge market shares and translate their market power into enormous profits.
Their success has brought tremendous benefits to users-and grave dangers to societies and economies. Each company hoards the information it collects and uses centralized systems to run its huge businesses. That hoarding has hampered innovation and allowed the companies to abuse user data, and their centralized systems leave online markets vulnerable to unexpected shocks, posing risks to the wider economy.
The most common answer to the problem of overly powerful firms is to break them up, as U.S. regulators once did to Standard Oil and AT&T. Yet that would destroy much of the value that these digital giants have created and probably do little to improve competition in the long run, since without structural reforms, killing today’s digital superstars would simply generate opportunities for new ones to emerge. There is a better solution: a progressive data-sharing mandate. This would leave these companies intact but require them to share anonymized slices of the data they collect with other companies. Such a mandate would decentralize digital markets and spur innovation as companies competed to extract the best insights from the same data. Much is at stake; if governments fail to act, they will leave key parts of Western economies and democracies vulnerable to sudden failures.
Markets in Everything
What sets the new digital superstars apart from other firms is not their market dominance; many traditional companies have reached similarly commanding market shares in the past. What’s new about these companies is that they themselves are markets. Amazon operates a platform on which over $200 billion worth of goods are bought and sold each year. Apple runs gigantic marketplaces for music, video, and software. As the world’s largest music-streaming service, Spotify provides the biggest marketplace for songs. The Chinese e-commerce behemoth Alibaba manages the world’s largest business-to-business marketplace. Meanwhile, Google and Facebook are not just the dominant search engine and the dominant social media platform, respectively; they also represent two of the world’s largest marketplaces for advertising space. And home-rental companies such as Airbnb and ride-sharing companies such as Uber and Didi Chuxing match workers with customers.
Markets have been around for millennia; they are not an invention of the data age. But digital superstar firms don’t operate traditional markets; theirs are rich with data, which they use to improve transactions and thereby make consumers better off. More data about the products on offer and the preferences of buyers and sellers helps people find what they are looking for and allows businesses to figure out how best to serve their customers. In traditional markets, informing everyone about which products they are likely to prefer is cumbersome and costly, so the information is usually condensed into a single figure: price. Relying on prices provides the grease that enables conventional markets to work, but it omits a lot of detail. If a potential customer, after learning the cost of an item, decides not to buy it, the seller will not know why. Was the price too high? Or was the product just not what the buyer was looking for? That loss of nuance hurts both sides of the transaction.
Prices can also be deceptive. They are easy to compare, so customers often believe that doing so is an appropriate way to choose the products that will suit them best. But behavioral economics has long shown that humans are poor judges of their future needs and that their ability to make objective comparisons can easily be skewed, resulting in many transactions that should not happen. By contrast, the data-rich markets operated by digital superstars offer lots of information about products and preferences, along with tools to search and filter the available goods. On Amazon, buyers can tick options to quickly identify the kinds of products they are looking for. Several studies have shown that Amazon rarely offers the cheapest choice, but buyers value the ability to find something easily that closely matches their needs.
Digital superstars also use data in another important way. They offer shoppers digital “decision assistants,” which sift through huge amounts of information to provide recommendations that are often superior to the choices that consumers might make on their own. Automated assistants suggest music on Spotify, movies on Netflix, and apps in Apple’s App Store. A third of Amazon’s retail sales reportedly result from customers following the advice of the company’s venerable decision assistant.
In each of these cases, the use of data improves market matches, and as customers keep shopping, the assistants learn how to make even better recommendations. Most of this learning takes place completely or largely unassisted by humans, as data is fed into machines that continuously update their algorithms. It is the combination of vast markets and decision assistants driven by the data those markets produce that makes these companies such enticing places for customers to spend their money. At least so far, consumers, and not just the digital superstars, seem to be winning.
The Return of Central Planning
Yet these gains in consumer welfare mask dangers for both individual users and the wider economy. As data-rich markets replace conventional ones, competition withers. On Amazon, for example, consumers can still choose from a variety of products, but Amazon decides what brands to carry-and it pushes its own products, from batteries to food. If Apple wants to ban a competitor from its App Store, it can. (Some argue that it has already done so by delisting apps that offer services similar to Apple’s own, although Apple disputes this.) Market dominance isn’t illegal, but if history is a guide, it commonly causes problems. As Microsoft discovered in the late 1990s, a thin line separates market dominance and abuse of power. If a monopoly’s stranglehold on the market enables it, that monopoly will almost always raise prices when it runs into trouble. Consumers may pay only a small extra cost per transaction as a result, but the profit difference for the monopoly can be substantial. That is one reason why many economists worry about the ascendancy of digital superstars. Market dominance also allows companies to squash competitors, either by undercutting them on price before they have a chance to grow sufficiently large to survive a price battle or by simply acquiring them.
Market dominance is not the only problem raised by the ascent of digital superstars. These companies gather comprehensive personal data on their customers and store that information in databases that are enticing targets for hackers. Reports of major data breaches are mounting, and many people rightly worry about how safe their personal information is.
But digital superstars pose an even bigger, more systemic threat to the wider economy. By hoarding data and operating their own, dominant decision assistants, they undo the resilience of traditional, decentralized markets. In a typical market, if a buyer or a seller makes a mistake, it may hurt that person or that company, but it will not bring down the market as a whole. In contrast, in many of the datarich markets run by digital superstars, the company controls all the information about customer preferences and transactions, and the company uses that data to help its own decision assistant learn. Buyers still make individual decisions, but they are greatly influenced both by explicit recommendations and by the ways in which options are filtered and presented. These markets are much closer to centrally planned systems than to resilient and decentralized traditional marketplaces.
Digital superstars don’t want to operate a centrally planned economy. But given the complexity of the algorithms that recommend products, it’s always tempting for these companies to manipulate the markets they operate. Last year, for example, the European Commission fined Google 2.4 billion euros after finding that the company had distorted its search results to favor its own shopping service. And a company’s motives don’t always matter: the hoarding of data makes it far easier to systematically influence consumers’ decisions even without intending to do so.
Central control of data means that faults in a system have the potential to influence not just a single consumer but also every market participant. If there is a systemic problem in Amazon’s recommendation engine, everyone heeding its advice may make needlessly poor decisions. People may buy products that don’t really match their preferences, shun a particular brand without even being aware of doing so, or pay too much for what they purchase. They may never be shown the search results that best answer their queries and never see the news that really matters. This not only exposes millions of consumers to the risk of wasting their money and their time but also endangers the market as a whole.
The Failure of Antitrust
Existing regulatory tools fail to address the risks posed by today’s superstar firms. Antitrust laws prohibit certain kinds of anticompetitive behavior. Companies can be fined for fixing prices or bundling products together to stifle competition. But such laws do not regulate market power directly; superstar firms operating data-rich markets can exercise a dangerous degree of control without violating any antitrust laws.
The legislators who wrote the existing laws assumed that competition would make it hard to concentrate market power for long. To be sure, as companies grow, the marginal cost of production usually decreases. As long as there are sufficient buyers, it pays for a company to get bigger. And for some products and services, new users improve the value for everyone else: when more users join a social media platform, the existing users have more people to interact with. But the advantages of being a big, well-connected company have long been balanced by the ability of start-ups to take on established firms by developing superior products or better production processes. Antitrust laws have been needed to stop only the most egregious abuses; innovation has eventually taken care of ordinary market concentration.
That is changing. Innovation is shifting to data-driven machine learn- ing. Insights are no longer the product solely of human ingenuity. They are now the result of the automated analysis of huge amounts of data. More and more, the success of a firm rests on its ability to use the information it possesses. Because only the largest firms have access to enough data to compete, innovation is losing its power to make markets fairer.
To solve this problem, some experts have suggested breaking up digital superstars, so that they no longer control the marketplace, the information that flows among market participants, and the decision assistants. The model would be the robust antitrust enforcement that led to the breakup of Standard Oil, in 1911, and AT&T, in 1984. A less drastic alternative might draw inspiration from the steps taken by regulators in the 1990s to force Microsoft to stop bundling a Web browser with its operating system and, more recently, to prevent Google from favoring its own services in its search results.
But by reducing firms’ ability to use large amounts of data, such measures would reduce market efficiency and leave consumers worse off. If, for instance, Amazon were broken up into a marketplace and a separate tool to provide recommendations, the latter would no longer have access to the huge streams of data generated by the former. Nor would a breakup improve competition. Alternative recommendation engines would not see the market data either, so their suggestions would be no better. It would not really matter how regulators broke a firm up-whether they created many little Googles, for instance, or split YouTube from Google Search- because after the breakup, all the new entities would have less information to learn from, leading to inferior products and services overall.
Similarly, although restricting the ways digital superstars can collect or use data-through tougher privacy laws, for instance-might fragment markets and thus improve their resilience, the quality of recommendations would deteriorate absent sufficient data, leading to inefficient transactions and reduced consumer welfare.
Trustbusting for the Digital Age
Luckily, regulators do not have to choose between structurally vulnerable but efficient markets and resilient but inefficient ones. There’s an easier way to foster both market diversity and resilience: a progressive data-sharing mandate. Under this system, every company above a certain size, say, those with more than a ten percent share of the market, that systematically collects and analyzes data would have to let other companies in the same market access a subset of its data. The larger a firm’s market share, the more of its data others would be allowed to see. Data would be stripped of personal identifiers, augmented with metadata to make clear what sort of information the data provided and where it came from, and selected randomly to prevent companies from gaming the system (by granting access only to largely useless data, for instance). Participants would have to agree to certain restrictions, including rules against sharing data with third parties. The role of regulators would be limited to assessing market share, an area in which they have already accumulated expertise. If necessary, regulators would also enforce access to data, but they would not actively organize or operate the sharing system.
Eventually, data sharing should be mandated across the board. But countries should start with online markets, as these are particularly vulnerable to the dangers of concentration. In the United States, Congress would have to amend the country’s existing antitrust regime to develop a comprehensive data-sharing regulation, and in Europe, the EU would have to act as a whole, but a transatlantic consensus would not be necessary. Both the United States and the EU have enough regulatory power and important enough markets to make a mandate enacted in either jurisdiction effective.
A progressive data-sharing mandate would offer several advantages. Unlike a tax, it would not impose any direct cost on firms; companies would remain free to use the data they collect, just as they do now. It would allow many firms and people to use the same data, which would spur innovation; today, although huge quantities of data are collected, it remains underused. If a wide variety of firms had access to market data, a firm’s competitive advantage would rest on its ability to extract insights, encouraging companies to develop smarter algorithms and analytics.
The policy would not differentiate between different players that crossed the necessary threshold; even Amazon would have access to data from smaller competitors as long as their market shares were greater than ten percent. But since smaller firms would have less data to share and machine-learning algorithms produce diminishing returns for each new data point, a company like Amazon would gain far less than its smaller competitors. A data-sharing mandate would lift all boats, but to different degrees. That would support diversity, innovation, and competition.
Once companies had access to the necessary raw material, they would launch alternative decision assistants. People might still shop on Amazon or listen to music on Spotify, but they might use a third-party recommendation tool to choose products and songs. Today’s decision assistants mostly serve the digital superstars. Tomorrow’s more independent decision assistants could far more convincingly represent the interests of consumers. Price-comparison sites already let people find the seller offering the lowest price for a wide range of products. Independent decision assistants would help them identify the best product match, as well.
Creating competition among assistants and markets would eliminate the need to break up digital superstars, because they would no longer enjoy an insurmountable competitive advantage. And because the shared data would be chosen randomly, each competitor would train its systems on slightly different data sets, reducing the risk of systemic failures.
Some may worry that mandated data sharing would only boost the power that firms have over consumers. But so far, regulators have mostly failed to counter power imbalances between users and companies by strengthening individual privacy rights. Even in Europe, which has enacted the strongest privacy protections, most people routinely click “OK” and accept companies’ privacy terms rather than exercise their right to choose what information they share. The problem is that current privacy regulations focus too narrowly on the relationship between each consumer and each firm, rather than considering the market as a whole. Mandating progressive data sharing would not solve the privacy challenge. But preventing a small number of digital superstars from monopolizing data would better distribute the power that flows from exclusive access to information. Requiring firms to strip obvious personal identifiers from their data before sharing it would also greatly limit the potential exposure of their users, spreading power more equally among companies without compromising privacy.
Making It Work
The idea of forcing companies to share their data may seem novel, but it has been employed successfully around the world. In the United States, the federal government sometimes requires firms to share data as a condition of allowing an acquisition. For example, in 2011, when Google bought the airline reservation company ITA Software, the Justice Department forced Google to continue to offer access to ITA Software’s travel data to third parties, including Google’s competitor Microsoft. The situation is similar across the Atlantic. In Germany, for instance, large insurers must let smaller insurers access some of their actuarial data, so that the smaller firms have enough information to calculate risks.
Recent EU legislation has boosted the idea of data sharing. Thanks to new rules, a consumer can now force companies to provide competitors with access to his or her banking data (as part of so-called open banking), as well as any other personal data (as part of the new General Data Protection Regulation). This approach, which the EU has dubbed “data portability,” will not have the same effect as a datasharing mandate, since it requires each consumer to act independently. But the legislative aim is similar: to break the stranglehold incumbents maintain on data and create a competitive market full of innovative data-driven services. Both the European Commission and members of the European Parliament are studying how to legislate progressive data-sharing mandates, and Andrea Nahles, a leader of Germany’s ruling coalition, has thrown her support behind the idea.
Requiring companies to share their data will come with challenges. Governments will have to overcome several organizational and technical hurdles, and making such a policy effective will require some fine-tuning. But the successful cases of regulated data sharing in the United States and Europe, and the rising interest in such policies, suggest that regulators will be able to find pragmatic solutions.
If, on the other hand, governments leave accelerating market concentration untouched, they will threaten more than competition. They will allow the most efficient digital markets to become also the most vulnerable, both to deliberate attacks and to accidental failures that cause systemic breakdowns. Allowing superstar companies that run centrally planned systems to maintain their dominance will erode the most important quality of capitalist markets, the decentralized way they make decisions. Worse still, this centralization has begun to spill over into politics. Just as healthy markets rely on people making independent decisions, democratic systems are founded on the individual choices made by voters. Yet today, much of public discourse is shaped by a few companies with exclusive access to vast quantities of voter data. Ignoring the problem is no longer an option.