Data-Driven Development

Executive summary

Data Deluge

In a sample second in July 2018, it is estimated that some 2.7 million emails were sent and received, 74,860 YouTube videos watched, and 59,879 gigabytes of internet traffic carried.1 Clearly, we generate huge and growing volumes of data.

The digital economy has become more information intensive, and even traditional industries, such as oil and gas or financial services, are becoming data driven. By 2020, forecasts Cisco, global internet traffic will reach about 200 exabytes per month, or 127 times the volume of 2005, with much of the growth coming from video and smartphones (figure ES.1). And that data may hold huge value. McKinsey Global Institute estimates that crossborder data flows in 2014 were worth about US$2.8 trillion, up 45-fold in value since 2005.

Figure ES.1 The growing internet

The vast majority of the data that exists today was created in just the past few years. The challenge is to extract value from it and to put it to work – for firms, governments, and individuals. Every citizen is producing vast amounts of personal data that, under the right protective frameworks, can be of value for the public and private sectors. Firms are willing to pay everincreasing amounts for our attention on social media sites and to mine the data we produce. But even data that is produced unintentionally – a byproduct of other processes, known as "data exhaust," such as call data records or GPS coordinates – can have value when effectively analyzed. Both types of data, their potential uses, and associated risks are all growing exponentially. Figure ES.2 shows common sources of personal data.

Figure ES.2 Types of personal data

Source: World Bank, https://openknowledge.worldbank.org/handle/10986/30437
This work is licensed under a Creative Commons Attribution 3.0 IGO License.

Who Benefits?

This new report, the fourth in the Information and Communications for Development series (figure ES.3), examines data-driven development, or how better information makes for better policies. The report aims to help firms and governments in developing countries unlock the value in the data they hold to improve service delivery and decision making and empower individuals to take more control of their personal data. The report asks just how we can use this data deluge to generate better development outcomes.

Figure ES.3 The Information and Communications for Development series

People's lives can benefit greatly when decisions are informed by relevant data that uncover hidden patterns, unexpected relationships, and market trends or reveal preferences. For example, tracking genes associated with certain types of cancer or explaining the potential links of Neanderthal DNA with resistance to the common flu virus or Type II diabetes can help improve treatments. As argued in chapter 3, development partners therefore need to establish strategies to better use data for development, while intervening appropriately in the data ecosystem and respecting data protection and privacy.

The World Bank Group, for instance, has established a Technology and Innovation Lab for improving data use in its projects, including using artificial intelligence and blockchain. This is part of a broader work program that aims to leverage data and technology in its work. The International Telecommunication Union has so far hosted two editions of the Artificial Intelligence for Good, Global Summit. And the team from UN Global Pulse, another partner in this report, is working with UN partners to responsibly harness big data and artificial intelligence for development and humanitarian action (see chapter 3).

However, firms and organizations that can make the best use of the data are not necessarily the ones that collect it. An "open data" mind-set is critical to data-driven development. Thus, an open marketplace for data is to be encouraged within limits. It is important therefore to develop appropriate guidelines for data sharing and use, and for anonymizing personal data. Governments are already beginning to release value from the huge quantities of data they hold to enhance service delivery, though they still have far to go to catch up with the commercial giants. To use data intelligently for better development outcomes, national statistical offices will continue to play a core function, including that of objectivity and impartiality, producing data "without fear or favor". But many statistics offices are struggling technically and financially. To remain relevant in an on-demand world, they need to strive for real-time data availability, striking an informed balance between accuracy and timeliness.

Data-Driven Business Models

Companies are also developing new markets and making profits by analyzing data to better understand their customers. This is transforming conventional business models, as explored in chapter 2. For years, users paying for calls funded telecommunications. Now, advertisers paying for users' data and attention are funding the internet, social media, and other platforms, such as apps, reversing the value flow. The share of the value extracted by the network providers is shrinking, threatening future investment. Good business models for investment in telecommunication networks typically have high up-front sunk costs, but very long-term returns. Twenty to thirty years ago, companies that built networks – such as NTT, China Mobile, AT&T, or Deutsche Telekom – were the champions of their respective national stock markets. Their assets, like the infrastructure that they put in place, represent the backbone services operate on. But their market values have fallen in comparison to the businesses gathering and storing data – such as Google and Alibaba Global – thanks to these existing infrastructures. Stock markets, in turn, assign huge potential to these data-rich companies, and undervalue the companies that keep the digital plumbing working.

We have seen this pattern before. In the early part of the nineteenth century, the markets of the time afforded optimistic valuations to the companies that built railroads. But as the century drew on, railroad investors went bankrupt or were nationalized because of their huge debts, even as the companies whose products they carried, such as mail-order companies, thrived in the early twentieth century.

Once again, we face an inflection point. For more than a hundred years, infrastructure companies made their money primarily from subscriptions and usage charges paid by users – who paid by the minute, by the mile, and lately by the megabyte. This is changing. The value of telecommunication networks is now not so much in data transport as in data storage. As chapter 2 shows, the companies with the highest market valuations are those that collect then monetize their customers' data through targeted advertising. Services from Facebook, Google, or Tencent are largely "free" at the point of use – yet their bandwidth requirements grow ever larger, as does their customer reach.

Beyond internet business or commercial applications, multiple opportunities also exist for harnessing the value of big data and artificial intelligence to help us achieve shared development objectives, as exemplified in chapter 3. However, global efforts to develop new frameworks for the responsible use of emerging technologies must address their implications for society and the consequences of both using data and algorithms, and of failing to use them.

Data Belongs to All of Us

People need to exert greater control over the use of their personal data. Their willingness to share data in return for benefits (real or perceived) and free services, such as virtually unrestricted use of social media platforms, varies by country and by age group (figure ES.4). Consumer research from GfK, a German research institute, shows that willingness to share is highest in China and lowest in Japan. Early internet adopters, who grew up with the internet and are now age 30–40, are the most willing to share. Many countries and regions have taken steps recently to update and reinforce rules on the use of personal data. The European Union's General Data Protection Regulation, which went into effect on May 25, 2018, imposes a long list of requirements for companies processing personal data. Violations will result in fines that could total as much as 4 percent of global annual turnover.

Figure ES.4 Are you willng to share your data?

Other countries have taken steps to restrict the flow of their citizens' data beyond their borders (data localization). In China, where data localization is strongly championed, restrictions on moving data are severe. Long-established controls over technology transfer and state surveillance of the population are predominant, and such measures form part of the country's "Made in China 2025" industrial strategy. The strategy is designed, in part, to make the country a global leader in tech-intensive sectors such as artificial intelligence and robotics. Chinese technology giants, including Baidu, Alibaba, and Tencent, are among the biggest in the world, and the country is establishing strong positions in new sectors like the Internet of Things (appliances, machines, and other items able to connect with the internet and exchange data). Throughout the world, data is regarded as a new asset class vital for industrial competitiveness.

Other emerging markets, such as India, Indonesia, the Russian Federation, and Vietnam, are also seeking data localization. The Russian Federation has blocked LinkedIn from operating in the country after the site refused to transfer data on Russian users to local servers. Divergent rules on the treatment of data impose significant costs on doing business online. Business organizations, including the International Chamber of Commerce, would like to establish rules to restrain what they call "digital protectionism". However, a serious gap exists in global governance with regard to cross-border trade in data, and a coherent approach is prevented by differing philosophies among the main trading blocs.

The ownership and control of data will continue to be a major question for society. Broadly speaking, there are three possible answers to the question "Who controls our data?": firms, governments, or users. No global consensus yet exists on the extent to which private firms that mine data about individuals should be free to use the data for profit and to improve services. Some governments argue that data from a country's citizens belongs to those citizens and should not leave the country without permission. Data dependency leads to new risks of exclusion. The data poor, who leave little or no digital trail because they have limited internet access, are most at risk of exclusion. But, equally, those who live in ways that society deems unconventional may also risk exclusion, for instance, because they lack a digital ID or are considered an insurance risk

This report espouses the view that citizens should control their own data and should be free to choose how to release it and even to commercialize it (figure ES.5), as explored in chapter 4.

Figure ES.5 Toward a new value chain for personal data

The growth of the data economy therefore requires changes in competition policy and the regulation of privacy. In a traditional, or one-sided, market, dominant firms are bad for overall market development. But when it comes to personal data, splitting the market share too many ways may inconvenience users and complicate matters for the individual if the different platforms do not connect, or if they require different passwords. As data becomes more important in shaping markets, it may reinforce tendencies toward monopoly, and thus monopoly profits, unless competition rules are modified to deal with new concepts of dominance. The emergence of multisided platforms, explored in chapter 5, poses new challenges for regulators.

Data and the internet have predominantly been regarded by pioneers and campaigners as a decentralized, self- regulating community. Activists have tended to regard government intervention with suspicion, except for its role in protecting personal data, and many are wary of legislation to enable data flows. But that position is under pressure from the increasing centralization of the internet and a series of revenue data breaches and media exposés of questionable business practices by social media platforms. The use by political parties in Kenya, the United States, and elsewhere of data harvested from social media profiles does not appear to have broken any rules, but it has led politicians on both sides of the Atlantic to take a closer look at social media giants, such as Facebook and Twitter. The proliferation of "fake news" has also spurred calls for action.

Data collected by governments, and thus paid for by taxpayers, arguably belongs to all of us. But there are limits to the openness paradigm. Citizens may not want data about themselves to be exposed without protection. And governments often lack the resources to extract value from their data without private partners. Data-driven development needs greater dialogue between the custodians of a country's data and its users. The key to unleashing the power of datadriven development for developing countries lies in intelligent management, use, and supervision of data.

Chapter 6 reviews data-related policy issues relevant to the digital economy. It considers policies geared toward building consumer trust, policies that facilitate or can affect access to data, and the use of data as infrastructure. The chapter also covers mainstreaming policies, such as those that facilitate the use of data for innovation or those that build digital skills. At least 35 economies are currently drafting data protection laws (map ES.1). In addition, a number of economies are considering reforms to their legal frameworks. One factor driving this consideration is the European Union's adoption of the General Data Protection Regulation. While the regulation introduces, or confirms, many important principles for data protection and privacy, it also extends these principles to firms from other parts of the world that wish to do business in Europe.

Map ES.1 Data protection and privacy legislation worldwide, 2018

Ironically, although data is becoming ever more important, data about data is still hard to find. The Data Notes to this report set out some of the indicators that should exist and present data that do exist on an internationally comparable basis for indicators such as the price and affordability of data transmission and the availability of open government data.

This report aims to stimulate wider debate within the development community on the nature of data for development. It is not the first word on this topic and certainly will not be the last. But it is a topic of growing importance that cannot be ignored.

Data: The Fuel of the Future

Data, Data, Everywhere

A self-driving car, one of the most anticipated developments of the next decade, is expected to generate some 4,000 gigabytes of data for each hour of driving, according to chip maker Intel. To put it another way, just 3 million autonomous vehicles would generate, or consume, more data than the combined human population of more than 7 billion. Vehicles provide just one example of how data generation and use are growing explosively. Other machines generating an overload of data include satellites, environmental sensors, security cameras, and, of course, the ubiquitous mobile phone.

We are undoubtedly experiencing a data revolution in which our ability to generate, process, and utilize information has been magnified many times over by the machines that we increasingly rely upon. By 2016, according to IBM, some 90 percent of data that exists had been created within the previous 12 months, a rate of 2.5 quintillion bytes per day. Firms are increasingly finding hidden value in some of that data. Some 7 of the top 10 companies worldwide, by market capitalization, are data driven in that they create value primarily from the data they collect from or sell to their customers. The remaining 3 firms in the top 10 ­– in the more traditional financial services, energy, and health care sectors ­– also increasingly build data into their products and services or use it to improve them (table 1.1).

Table 1.1 Data hogs: Top 10 private companies globally, by  market capitalization, May 2017

With some justification, therefore, data has been called the new gold, the new oil, or the world's most valuable resource. Like oil, unprocessed data has relatively little value and needs to be mined, refined, stored, and sold on to create value ­– albeit in data centers rather than in oil rigs. But unlike oil, the quantity of data is ever increasing, not diminishing. Even though data is a nonrivalrous good, in the sense that my consumption of it does not affect yours, it is also excludable, which means it can be sold for profit, many times over. This makes it what economists sometimes call a club good, like privately owned safari parks or pay-per-view television. But because of the ever-increasing quantity of data, extracting value from it requires ever-greater computer power. Thus, the spoils from data-driven markets typically go to the largest players; those with the deepest pockets, the most users, the largest data centers, and most wide-ranging ability to collect and analyze data. Consequently, it is possible for market capitalization in companies like Facebook, Tencent, or Alibaba to exceed their annual revenue by 15 times or more and for the market capitalization of Apple and Amazon to touch the US$1 trillion mark in mid-2018, because investors view them as well positioned to take advantage of future data trends.

How Data Is Changing Development

This report is about how the data revolution is changing the behavior of governments, individuals, and firms. Specifically, the report examines how these changes affect the nature of development – economic, social, and cultural. How can governments extract value from data to improve service delivery in the same way that private companies have learned to do for profit? Is it feasible for individuals to take ownership of their own data and to use it to improve livelihoods and quality of life? Can developing-country firms compete with the internet majors on their own turf and even be more innovative in their use of data to serve local customers better? Several potential audiences could therefore benefit from this report:

• The primary audience is government policy makers, though not in a single line ministry, such as information and communication technology or finance, but rather across government, given that data is a multidisciplinary concern

• A secondary audience would be individuals concerned about how their personal data is used and those interested in how the data revolution might impact future job prospects.

• Beyond that, private sector firms, particularly in developing countries, looking to expand their markets and improve their competitive edge will find interesting examples of how other firms are doing that.

• Finally, development professionals should find the report relevant as they seek to use data more creatively to tackle long-standing global challenges, such as eliminating extreme poverty, promoting shared prosperity, or mitigating the effects of climate change.

A Data Typology

Data-driven development is an emerging and rapidly changing field. So it may be useful, at the outset, to define terms recurring across the chapters. These are not fixed or official definitions, but rather working usage for this report:

• Big data, a commonly used term, describes data sets so large or so complex that traditional data processing techniques are inadequate. The field of big data analytics uses advanced computational techniques to extract meaningful information (such as patterns, trends, repetition) from data. For the moment, big data is largely the domain of large private companies. But as tools to mine it become cheaper and more readily available, smaller companies and governments will also use big data. It can be useful to further distinguish between big data produced intentionally or unintentionally and that produced by humans and by machines, as in table 1.2. 

• 	Data generation	Intentional	Unintentional
  Data agent	Human	Primary content	Data exhaust
  	Machine	Secondary content	Internet of Things data

• Data exhaust, which is unintentionally created by humans. This can include metadata (data about data), such as call-data records derived from mobile phones, or the trail of data left by users engaged in other activities, such as keystrokes. Data exhaust generally has low value, but the trail left by millions of users can be mined or combined to extract value or to hack into an otherwise secure system.
• Internet of Things data, which is intentionally created, but from sensors and other internet-connected devices, rather than from humans. This mainly has value in the aggregate ­– and over time ­– but can also be used to provide alerts for impending events, such as extreme weather conditions.
• Primary content is intentionally created by humans, typically users. An example here might be a social media profile or a browser search history. When thousands or more of these are combined and anonymized they can be used, for instance, for analyzing popular or emerging trends. Humans also create primary content in the form of videos, academic papers, blogs, and the like that can be mined, for instance, for sentiment analysis.
• Secondary content is intentionally created, but through artificial intelligence rather than directly by humans. A benign example would be a chatbot that helps a user fill in a form online by giving suggestions. A malign example would be a fake social media profile that seeks to influence buying habits or political opinion.
  

• Personal data relates to an individual and is generally concerned with private information. Personal data can form large, complex data sets (such as multiple health indicators including weight, blood pressure, or heart rate measured over a lifetime) but more normally constitute small data, which can be easily monitored. Personal data may be willingly exchanged in return for convenience (such as a phone number or email address), but it can also be given away unwittingly (such as date of birth provided to enter an online competition) or unwillingly (such as data hacked from a personal email account). The consequences of loss of personal data, explored more in chapter 4, might include loss of privacy and loss of control (over the future use of personal data) and a loss of agency (such as being exposed to a more limited range of news sources or opinions as a result of previously expressed preferences). What is relatively new is how persistence, repurposing, and spillovers from big data increase the risk and uncertainty about how private data can be used in the future.

• Open data refers to data made freely available and deliberately stored in an easily read data format, particularly by other computers, and thereby repurposed. For instance, data on airline schedules could be easily read by travel companies to generate customized itineraries for travel websites. Governments may use open data to promote transparency and accountability in their operations and allow voters to measure the performance of different government functions. As a philosophy, therefore, open data is intended to encourage the juxtaposition of data from different sources to create value and new applications. It is estimated, for instance, that some 500 different applications use London transport data, and the savings to the UK economy from its open data policy amount to some £6.8 billion (about US$9.5 billion) a year. Open data tools are particularly useful in the transport sector (see box 1.1).

Box 1.1 Open data tools for improving transport through big data

Rise in digital data. Digital data has proliferated with the rapid increase in smartphone ownership in advanced and emerging market economies, alongside advances in global positioning systems and digital sensors. This data has the potential to transform transport systems worldwide. The location-tracking data provided by smartphones, for instance, can reveal how and why people travel, information critical for optimizing transport networks. Accordingly, opensource tools and cloud-based platforms have been developed to help collect, manage, and analyze the ever-increasing volume of digital data. These easily accessible tools provide individuals, governments, and private entities with sophisticated analysis capabilities, empowering them to improve all aspects of transport. Such tools will be particularly beneficial in developing countries that have limited resources.

Open-source tools. The World Bank has developed a variety of free-of-charge tools that capitalize on big data to facilitate transport-related development projects across the globe (see Figure B1.1.1). These tools provide numerous capabilities, including transit system analysis, route planning, and road condition and incident reporting. Open Transit Indicators allows public transit administrators to assess existing services and identify improvements through the collection and analysis of standardized transit data. This approach has been used to address transit problems in China, Kenya, Mexico, and Vietnam, among others.b The Rural Accessibility Platform uses freely available OpenStreetMap data to evaluate the accessibility of rural population centers to points of interest.c Indices of rural accessibility have been used to identify needed transportation improvements in countries including Bangladesh, the Lao People's Democratic Republic, and Zambia.d These open-source data and tools make transport analysis accessible for a broad range of users.

Citizen engagement. The increasing ubiquity of smartphones and internet connectivity is allowing individuals to provide valuable data and contribute to development efforts. Citizen engagement is prioritized in many of the World Bank's transport-focused open-source tools. For example, the smartphone application RoadLab uses a crowdsourcing approach to obtain route information and roadway infrastructure conditions from users.

Figure B1.1.1 How open data tools can assist transport



The related tool RoadLab Pro was used to assess the conditions of unclassified road networks in Mozambique, demonstrating the potential of citizen provided smartphone data in transport planning. These tools provide an easy-to-implement way for traffic engineers to obtain roadway information, particularly when professional pavement testing equipment and base geographic information system maps are not available. Similarly, DRIVER capitalizes on crowdsourced data to collect road incident information, which can then be visualized and analyzed to improve enforcement and resource allocation. In the Philippines, DRIVER has been applied to identify and prioritize problematic road areas for interventions.

• Metadata, or "data about data," is used to classify, categorize, and retrieve data files. For instance, metadata might include the date on which it was created, the number of pages or data size, and keywords that can be used to search. Data attributes may be added to data according to the way it is typically used, for instance, how popular it is as a function of how frequently it is downloaded. Metadata helps with data analysis and can be applied to data users, such as by giving them attributes, sometimes based on inferred data, that equate to a "reputation".

• Data platforms offer a convenient and cost-effective way to link customers and suppliers. Some platforms may connect only peers (such as a dating website) and others might be internal to an organization (such as an intranet). But most of the popular platforms using the internet are multisided platforms. Uber, for instance, connects drivers with riders; AirBnB connects property owners with guests; and Jumia connects sellers with potential buyers. But the biggest platforms are those that connect advertisers with consumers, usually in return for some kind of free service, such as social media or web browsing. As explored in chapter 5, multisided platforms, driven by advertising, are now among the most powerful firms in the world.
  

How Governments Use Data

From e-government to digital government

How governments use data runs throughout this report, although because it is the subject of a separate World Bank Group report it gets no separate chapter here. Chapter 3 nonetheless focuses on big data for social good, by international, nongovernmental, and humanitarian organizations, as well as by governments.

In the first generations of "e-government," much of the emphasis was on channels ­– using web browsers, and more recently, mobile devices, to access government information and services and to perform transactions. In this period, data was often seen as just the payload of the transaction ­– information supplied by the citizen or the business to support the request for service and the information supplied by the government in return.

However as "e-government" has evolved into "digital government," data is seen increasingly as a strategic asset with value lasting beyond a particular transaction and able to strategically transform the efficiency and effectiveness of government through:

• Making "e-government" transactions more attractive and useful to citizens and businesses by eliminating the need to supply the same information again and again, making transactions more suitable for the mobile channel, and allowing continuity of transactions over time, through different channels, and among different government institutions

• Allowing governments to become more "data driven" at all levels, from policy making through operational management and risk management to individual decision making

• Underpinning the creation of "smart cities" (and "smart nations") whose systems and infrastructure adapt automatically to the needs and behaviors of their inhabitants

• Providing, through "open data" and other programs, authoritative reference, geospatial, and other data to the national economy and society as a whole to improve transparency, to enable economic growth and business innovation, and to increase the engagement of citizens in the co-creation of public services. An example of this would be a National Spatial Data Infrastructure, or Digital Maps.

Viewing government data as a strategic asset leads to requirements for effective and strategic data governance and data management across the entire life cycle of data, including how data is collected, described, and catalogued, as well as secured and controlled (not just to protect confidentiality, but also to ensure availability and integrity). Preparing government data for wider use will require elimination of unnecessary duplication or avoidance of re-collection of data. It will also require a strategic view of how data is shared across government, used within government and other public services, and made available to other economic and societal actors to generate additional economic and social value.

The changing role of national statistical offices

These requirements are in turn leading to demands for new skills and roles, including "data scientists" and "chief data officers," and new functional capabilities such as "data analysis," "big data," and "visualization". Historically, national statistical offices were, appropriately, the central repository of data, along with national archives, as they have the skills and resources to catalogue and manage data. The skills of information scientists and librarians in these offices may not be so readily available to more casual users of data in line ministries. But national statistical offices have had to reinvent themselves in the internet age, in which a simple web search can come up with many more possible sources of information than even the most dedicated librarian can track.

International authorities also need to collaborate on standards for information management. The General Statistical Business Process Model, for example, is a framework for organizing business processes in statistical organizations adopted in more than 60 countries.1 But as national statistical offices modernize and partner with nongovernmental entities that provide data for official statistics, there is recognition that such frameworks and business models may be too rigid. For instance, Statistics New Zealand's 2020 strategy affirms the organization's role as a producer of official statistics but moves beyond this to acknowledge its place in a broader ecosystem and its renewed purpose of "adding value to New Zealand's most important data" through increased data cooperation, integration, and analysis.

The challenge for national statistical offices, therefore, is to ensure the information they hold is properly catalogued (metadata) and easily searchable, and to offer this expertise throughout government. This requires a changing business model for the offices. They can no longer expect to cover costs primarily through sale of publications, though this may still provide an additional source of income. Instead, they must rely on the central treasury for most of their income, and on payment for services provided to other parts of government. Where the central treasury is already overstretched, as in many developing countries, national statistical offices frequently struggle. Thus, just as the value of data is becoming all the more evident in the private sector, it is too often neglected in government, especially at the regional or local level.

Sharing data across government

Data collected and held by one government agency may be valuable to another agency in its operations. For instance, it may relieve the second agency of responsibility for collecting the data itself. And in countries such as Belgium, Estonia, or the Russian Federation the government is not allowed to ask citizens again for data that it has already collected from them.

Of course, if personal or classified data is shared between ministries, it is important that it is shared securely. In the United Kingdom, two compact discs containing personal details of some 25 million children were lost in transit between two government agencies. This led to the mandatory use of encryption when moving confidential data between government agencies.

A number of countries have taken the concept of data sharing further by explicitly recognizing the importance of unified databases accessible to and used across the public sector, rather than each agency keeping its own records. In 2012, Denmark published a strategy for "Good Basic Data for Everyone ­– A Driver for Growth and Efficiency".2 Public authorities in Denmark register various core information about individuals, businesses, real properties, buildings, addresses, and more. This information, called basic data, is seen as important for reuse throughout the public sector because it is an important basis for public authorities to perform their tasks properly and efficiently, "not least because an ever-greater number of tasks have to be performed digitally and across units, administrations and sectors". Some of the registers do not contain personal information and are released as open data (for instance, addresses). In the Netherlands. there is a similar initiative for the sharing of 17 "base registers". The United Kingdom, despite past political controversy, is collaboratively developing a data-sharing policy that will allow the use of key databases across the public sector and, in some circumstances, beyond.

In federated countries those data sets need to be available not just between national agencies, but also regional and municipal agencies. Since changes to master data may first be notified to other agencies, robust processes are essential for the maintenance of the master data using notifications of change as early as possible; this is even more important in federated systems, where important changes, such as of address, may well be notified locally first.

This also exemplifies the increasing extent to which leading governments see databases, not functions, as the key asset of government administration, along with developing strategic plans to introduce interoperability standards and middleware that allow seamless integration of these databases through open application programming interfaces (widely known as APIs).

Structure of the Report

Following this overview chapter, with its focus on government use of data and presentation of definitions, Part I of the report look s at the "supply side" of the data sector.

• Chapter 2 looks at data connectivity and capacity, considering where data comes from, how it is stored, and where it goes. Specifically, the chapter looks at the technological drivers that make data ever cheaper to collect, store, and transmit, and the relationship between data and economic growth.

• Chapter 3 examines data technology, specifically big data analytics and artificial intelligence, and how this is contributing to development, especially in humanitarian interventions. The enthusiasm for the uses of these new tools is tempered by awareness of the ethical issues.

Part II looks at the "demand side" of the data sector:

• Chapter 4 looks at people's use of data and asks whether scope exists for a new model for a data market in which individuals may be able to trade access to their personal data. The underlying principle is that the data itself has no value, but the use of it has. The chapter goes on to examine the potential costs of a data market in possible losses of privacy, control, and agency

• Chapter 5 examines how firms use digital platforms in the data economy, and how that contributes to competitiveness, particularly for small and medium enterprises (SMEs). The chapter details several developing-country platforms and emerging business models, and concludes by considering how SMEs in developing countries can make better use of data to improve competitiveness and thereby compete against the dominant international social media companies.

Part III of the report brings together the policy implications for developing country stakeholders:

• Chapter 6 discusses the policy issues surrounding the use of data, notably over privacy, data localization, and security issues. The chapter also considers the value of digital ID systems, which many countries have adopted in recent years, though some have specifically rejected them. Finally, the chapter returns to the themes of open data and big data and offers recommendations.

The Data Notes appendix to the report looks at statistical indicators associated with the use of data. It also presents the 2018 update of the Digital Adoption Index (DAI), a composite indicator first introduced in the 2016 World Development Report: Digital Dividends. The DAI is an analytical tool that compares the relative adoption of digital technologies by governments, people, and firms within a country.

Supply: Data Connectivity and Capacity

The Ever-Expanding Data Universe

The rapid growth of internet users and faster network speeds is driving an avalanche of electronic data. About 3.5 billion people globally were using the internet in 2017, up 73 percent, or 1.5 billion, since 2010 (figure 2.1, panel a), and penetration has risen to almost half the world (48 percent in 2017). The rapid increase in users is driving demand for more internet content.

End-user internet speeds are also increasing rapidly, in turn driving use of broadband content and applications (figure 2.1, panel b). Global average wired broadband speeds are projected to nearly double from 25 megabits per second (Mbps) in 2015 to 48 Mbps by 2020 as more users move to fiber and higher-speed coaxial cable. As speeds rise, so does demand for video content.1 Mobile broadband speeds, which are much lower than fixed speeds, averaged just 2 Mbps in 2015. This will more than triple to 6.5 Mbps by 2020 as more users switch to fourthgeneration technologies. Mobile speeds vary greatly by device; smartphones are nearly three times faster than the global average, which results in more time spent online. In the United Kingdom, time spent on the internet more than doubled between 2005 and 2015 from 10 hours a week in 2005 to 23 hours in 2015.

Figure 2.1 Internet users and broadband speeds

Data can either be measured as stock (the amount of data stored in a location) or flow (amount of data transmitted from one location to another). One stock indicator is the number of websites providing partial information about content growth on the internet. According to Netcraft, a leading research firm covering the internet, 170 million websites were active in June 2016, up from just 8 million in June 2000. Worldwidewebsize.com puts the number of indexed web pages at 4.5 billion.2 Although useful, these numbers still lack the ability to portray the full scale of data accessible over telecommunication networks. They do not include the so-called dark web, ranging from innocuous private sites collecting sensor data to nefarious sites carrying out illegal or semi-legal activities. Furthermore, not all data going over the internet is from websites; it can also arise out of voice over internet protocol, video conferencing, gaming, and machine-to-machine communication.

More is known about data flows over the internet. In the past, separate networks existed for specific content and functions: for instance, telecommunications for voice and, later, text messages; broadcasting for television and radio; and private networks for businesses. The development of the internet and internet protocol (IP) communications has changed all that. Communications networks have generally shifted from circuit-switched to packet-switched IP networks, enabling virtually any type of content, from voice to text to multimedia, to be encoded and distributed digitally. According to information technology (IT) company Cisco, traffic over the internet will grow by more than 20 percent a year between 2015 and 2020. This data deluge has popularized a new vocabulary of petabyte and exabyte that spell checkers have not yet caught up with. Cisco proclaimed that the world entered the zettabyte era (an amount equivalent to 250 billion DVDs) in 2016 when annual global internet traffic surpassed 1 zettabyte.

It is useful to understand how internet traffic is classified to understand how devices, users, applications, and services are driving this growth. Internet traffic consists of IP and managed IP traffic. The former is exchanged between internet service providers (ISPs), whereas the latter is end to end within the same ISP's network. IP traffic can be further disaggregated by whether it emanates from fixed or mobile networks. The two accounted for three-quarters of internet traffic in 2016, with fixed making up more than 90 percent (figure 2.2, panel a). Managed IP traffic is forecast to decline by 10 percentage points between 2015 and 2020, and the share of mobile data in total traffic is projected to rise from 5 percent in 2015 to 16 percent by 2020.It is useful to understand how internet traffic is classified to understand how devices, users, applications, and services are driving this growth. Internet traffic consists of IP and managed IP traffic. The former is exchanged between internet service providers (ISPs), whereas the latter is end to end within the same ISP's network. IP traffic can be further disaggregated by whether it emanates from fixed or mobile networks. The two accounted for three-quarters of internet traffic in 2016, with fixed making up more than 90 percent (figure 2.2, panel a). Managed IP traffic is forecast to decline by 10 percentage points between 2015 and 2020, and the share of mobile data in total traffic is projected to rise from 5 percent in 2015 to 16 percent by 2020.

Figure 2.2 Global IP traffic and global consumer IP traffic

Businesses and consumers generate traffic, with the latter accounting for more than 80 percent in 2015, a share not projected to change much through 2020. Video dominates consumer IP traffic. It accounted for 38 exabytes a month of traffic in 2016, 71 percent of consumer IP traffic, and 43 percent of total IP traffic. It is forecast to grow by more than 30 percent a year so that, by the year 2020, it will account for 82 percent of consumer traffic and 57 percent of total IP traffic. Online gaming is projected to be the fastest-growing traffic stream between 2015 and 2020, at 47 percent per year. However, it accounts for a tiny share of total consumer traffic and its contribution will only rise from 0.2 percent to 0.4 percent.

Goodbye Data Carriers, Hello Data Creators

The rise of the internet has altered communications network value chains. In the past, little value was perceived in the content of traffic carried over communications networks. In the telecommunication world, this traffic was mainly telephone calls. While some of the calls may have triggered wealth, the direct income accrued to telecommunication carriers that transmitted the calls and billed them. In the case of broadcast and private networks, content was more financially significant but intra-industry (such as broadcast transmissions to satellite and cable television companies, banking transactions).

The development of the internet in the 1960s, the World Wide Web in the 1990s, and its iteration in Web 2.0 more recently has modified the way content is obtained and created.

Traditional content providers such as the media and audiovisual companies have moved online either with their own websites or by licensing content to streaming platforms. Take the BBC, which has 98 million global internet users viewing 1.5 billion pages a month. Or O Globo, one of Brazil's largest newspapers, whose online readers (23 million) outnumber print readers (300,000) by 75 times. A big difference is that not only can anyone access content anywhere on the public internet, they can also create it. Users become creators by sharing their own content with others through blogs, videos, social networking posts, and product and service reviews. Attention has shifted from the carrier of the data to the creator; from "the medium is the message" to the messages delivered over the medium. A company's telephone number is arguably no longer as important as its website, and individuals increasingly exchange their email or social networking links. Similarly, in video entertainment, power is shifting from the company broadcasting the content to the creator. This is reflected in the rise of companies offering internet-delivered video such as Amazon and Netflix and television content creators now embracing the internet (HBO NOW streaming service).

Many of the world's top websites (ranked by a combination of users and page views) are platforms for usergenerated content such as social network posts, video sharing, blogs, and collaboration (for example, Wikipedia) (table 2.1). All of the top sites are headquartered in either the United States or China, the two countries with the most internet users (more than 900 million combined, or just over a quarter of all internet users). While the US sites are mostly global, Chinese ones are mainly local. The concentration of so much data on so few sites is concerning, particularly as the giant internet companies behind most of them branch into other domains. Many aspire to be the single window to communications, news, and shopping.

Table 2.1 top 10 global websites

Data centers: Greener and further away, or closer to home?

The growth of internet content is driving the need for places to store it. A data center is a location with networked computers providing remote storage, processing, and distribution of data. They are mainly operated by global IT companies, governments, and enterprises that host other companies' data (that is, colocation). Data centers vary in size, capability, security, and redundancy. A so-called tier 1 data center provides basic nonredundant connections between computer equipment and may be prone to electrical outages, and a tier 4 center has redundant components, multiple connections between computers, continues to operate during maintenance, and is protected against most physical events.

Statistics vary widely about the number of data centers in the world. One challenge is that most centers are "small racks in computer rooms in smaller companies". Although the number of data centers has grown rapidly, growth is forecast to slow because of the trend toward larger spaces. More information is available about giant data centers, referred to as "hyperscale" because of their size and ability to add servers and storage as needed. They are operated by about two dozen global IT companies, including heavyweights such as Amazon, Microsoft, and IBM, as well as enterprises providing cloud-computing services. The 259 hyperscale data centers in 2015 are projected to grow to 485 by 2020 (figure 2.3).

Figure 2.3 Hyperscale data centers

Although the majority of hyperscale data centers are in developed nations, data center growth in emerging markets has ticked up. As more users are connected to the internet in lower-income nations, demand for data is rising. IP traffic is forecast to grow fastest in developing regions during 2015–20. Some countries are concerned about data sovereignty, insisting that government data be stored in the country, driving demand for national data centers. Software parks have also become popular in developing nations as a way to grow their digital economies, and data centers are essential for these facilities. Although the operation of a data center does not create many jobs, they are an essential platform for companies using them to generate revenue and employment.

Telecommunications carriers are particularly keen on data centers as a way to offset declining revenue from traditional voice services. Japan's biggest carrier, NTT, is one of the largest data center operators in the world, with more than 140 across the globe. Leading carriers have formed a working group of the Open Compute Project for the adoption of common standards for data centers. Operators in developing nations from Paraguay to the Philippines are busy constructing state-of-the-art data centers. Mobile group Millicom has been launching data centers in African countries, and its Paraguayan data center won an award in 2016 for its modular design. The Philippines Long Distance Telephone Company has constructed eight data centers across the country to be close to IT parks and support its cloud-based service offerings.

But data centers require significant electricity to power and keep equipment cool. According to a study, data centers in the United States accounted for 2 percent of that country's electricity consumption in 2014.

The data center industry is therefore constantly looking for ways to reduce reliance on fossil fuels, particularly given the possibility of data rationing due to shortages of electricity:

This has made geographies with cool climates and abundant hydro or geothermal energy attractive locations for data centers. Google's Finnish data center is built in a restored machine hall designed by renowned architect Alvar Aalto and draws on water from the Bay of Finland for cooling. Its data center in West Dublin does not need air conditioning units because of Ireland's cool climate. Some developing nations have similar environments, making them ideal for data centers. The Data Center Services data center in the Thimpu TechPark draws on mountainous Bhutan's abundant hydropower and year-round cool climate and, in 2017, the government launched its first data center.

Many developing countries face a challenge competing with hyperscale data centers abroad given that electricity costs tend to be relatively high. In Rwanda, the government is considering subsidizing data center electricity costs to attract more digital companies to the country and for local firms to transition their websites to local hosting enterprises. Small island developing states generally have high electricity costs due to the absence of local energy sources: 8 of the 10 most expensive countries for electricity are such states. However, they are surrounded by a useful resource: cool seawater. As noted, Google uses seawater to cool its Finnish data center and Microsoft is testing underwater data centers. Mauritius has also experimented with ocean water to cool data centers. Some Pacific small island developing states and other coastal economies have taken advantage of new submarine cable connections to bundle data centers into the landing station, lowering construction costs. Samoa, which has among the highest electricity costs in the world (figure 2.4), installed an energy efficient prefabricated data center in its new cable landing station. Many developing nations also have abundant sunshine with great potential for solar energy to lower costs. This is the thinking of mobile group MTN, which deployed Africa's first solar data center at its head office in Johannesburg.

Figure 2.4 Price of electricity (US cents per kilowatt-hour)

Reliability is critical for data centers. Developing countries, particularly in Africa, will need to improve the quality of the electricity supply to create the proper environment to attract investment in data centers (box 2.1). This will require electricity sector reform and prioritizing reliability for firms.

IXPs and caches: Closer to the edge

Although trends suggest a move toward larger data centers, the tendency is toward pushing data closer to the user or the "edge" to reduce latency and lower costs. Having data close to end users is critical, particularly in the financial sector, in which a few milliseconds advantage has a huge potential impact. This is raising traffic on internet exchange points (IXPs), places where telecom carriers and content providers come together to exchange their traffic (peering). This is cheaper, particularly for developing countries, since internetwork traffic does not need to be sent over costly international links only to return. In addition, ISPs do not need to make peering agreements with each potential partner. IXPs also improve quality since they are situated closer to the user and hence have less latency. "Soft" benefits are also associated with IXPs, such as developing technical skills and fostering a culture of cooperation, helping to sustain the internet. As the volume of data transmitted over the internet accelerates, IXPs have become even more relevant for ensuring that it is quickly exchanged among different parties.

The largest IXPs (measured by traffic or members) are mainly in Europe, with its long tradition of multistakeholder internet cooperation. The biggest is the German Internet Exchange (DE-CIX) founded in 1995, with locations in Dusseldorf, Frankfurt, Hamburg, and Munich. DE-CIX Frankfurt is the world's leading internet exchange, with peak traffic of 5.6 terabits per second in March 2017. More than 700 networks are connected, and access is available from 20 data centers across the city. The networks connected to DE-CIX are a smorgasbord of giant telecommunication carriers (such as Deutsche Telekom, China Telecom, Verizon, and NTT) and emerging country operators (such as Sri Lanka Telecom, Telkom South Africa, and Telkom Indonesia), big IT firms (such as Apple, Google, and Microsoft), and content and service providers (such as eBay and Facebook). DE-CIX began expanding abroad in 2012 and now operates IXPs in Dallas, Dubai, Istanbul, Madrid, Marseille, New York, and Palermo.

Although IXPs are burgeoning in most developed markets, growth has been uneven in developing nations. According to one source, 78 economies are still without an IXP (map 2.1).

Map 2.1 Internet exchange points around the world, 2018

The establishment of an IXP is often hampered by small markets, vested interests, and limited or unbalanced competition. Powerful incumbents with a high level of control over international gateways prefer that ISPs use their overseas links for IP transit. Nevertheless, developing regions, such as Latin America and Africa, have been adding ISPs at relatively high levels (table 2.2). Where no powerful incumbent exists, IXPs can thrive. This is the case of the Rwanda IXP, where the historical operator no longer exists. The IXP has 13 members, including all of the country's infrastructure-based ISPs. Peak traffic load was over 1 gigabit per second in March 2017, up more than 50 percent from the previous year. An IXP is particularly relevant in landlocked countries like Rwanda, which is far from undersea fiber-optic cables.

Table 2.2 Internet exchange points by region

Cloud Computing: Back to the Future

The ability to store and process data remotely dates to the early days of computer networks. Back then, end-user devices were "dumb" terminals hooked up to large mainframe computers that did all the work. The invention of the personal computer was revolutionary in providing users with their own device that could run applications and store data. The process is again reverting to centralized control, where data is increasingly stored and processed over the "cloud" on anonymous data servers. Three main reasons explain this:

• Faster networks. Rising internet speeds are making the transfer of data between device and cloud increasingly transparent. According to Cisco, average global fixed broadband speeds were 25 Mbps in 2015, up from 7 Mbps in 2010. Average mobile speeds were considerably slower at 2.0 Mbps in 2015, but with large device differences; smartphones averaged 7.5 Mbps around the world in 2015 and are forecast to rise to 12.5 Mbps by 2020.

• Greater storage. Storage available over the cloud is vastly superior to what can be saved on a desktop, laptop, or tablet computer or smartphone.

• Proliferation of smart devices. As the number of devices a person owns increases, the cloud provides a useful way of keeping them all synced. There were 2.2 devices per person worldwide in 2015, projected to rise to 3.4 by 2020.

Several acronyms are used to identify different cloud services. Infrastructure as a service (IaaS) offers computing power and storage. Platform as a service (PaaS) offers computer programs and other tools for users to develop their own applications. Software as a service (SaaS) offers complete applications and supporting upgrades and maintenance.

There are a number of benefits for cloud users, including reduced need for IT expertise, flexibility for scaling, and consistent application rollout and maintenance for large organizations. Free cloud services also exist that provide office-like application tools useful for small and medium enterprises (SMEs), as well as social network pages and blogs. This is particularly relevant for developing countries where the cost of licensed software can be an obstacle to creating applications and services.

Though cloud computing offers a number of benefits, it comes with costs and risks. Users will utilize more of their data allowance accessing cloud services, and businesses face migration costs either converting to the cloud or changing cloud providers. Risks include security and privacy breaches as well as potential loss of service due to communications or electrical failures. These risks have been well publicized through headline stories detailing cyberattacks, such as against IT giant Yahoo, affecting some 1.5 billion accounts.

Internet of Things: Data Is All Around

According to Swedish IT equipment manufacturer Ericsson, "things" connected to the internet will overtake devices used by humans in 2018. Cisco reckons some 12 billion devices will be connected to the internet by 2020 that talk to other devices or computers, up 20 percent a year from 2015 (figure 2.5, panel a). These so-called machine-to-machine connections form the heart of what is referred to as the Internet of Things (IoT), an interconnected ecosystem of sensors, meters, radio frequency identification chips, and other gadgets. Traffic from these things will grow at twice the rate they are being connected, or 40 percent a year from 2015 to 2020, from 1 exabyte per month to 6.3 (figure 2.5, panel b).

Machines have talked to each other for years over communication networks using electronic data interchange and other formats, largely to exchange financial information such as transactions from bank automated teller machines or companies ordering products or services from each other. The IoT expands scope, as the things doing the communicating are generally small devices and sensors tracking everything from utility use to automobile movements.

A report from the International Telecommunication Union (ITU) and Cisco argues that the IoT could be beneficial for developing countries, since it lowers the cost of service monitoring and delivery, allowing countries to gain in areas such as health and energy over a shorter timeframe than ever before. The ITU has formed a study group to enhance global standardization and collaboration on the IoT. One example is Ghana, where sensors are helping improve the vaccine supply by indicating whether refrigeration was affected during transport

The gap in IoT adoption is wide, according to statistics. For example, adoption of machine-to-machine communications, a subset of the IoT, varies tremendously in Organisation for Economic Co-operation and Development (OECD) countries, with Sweden and New Zealand ahead by some margin in respect to machine-to-machine connections per 100 people (figure 2.6). Sweden's telecommunication companies are striving to be leaders in Internet-of-Things services, and the country also has a sizable Internet-of-Things startup ecosystem. And a major reason for New Zealand's high penetration is that the main gas and electricity company has installed more than a million smart meters in homes and businesses across the country.

Figure 2.6 Machine-to-machine connections per 100 people, OECD member countries, june 2016

Many analysts see fifth-generation (5G) wireless technology as critical for the IoT because of its expanded data handling. According to one report, 5G networks can process about a thousand times more data than today's systems. Of particular relevance for the IoT is 5G's ability to connect many more devices (such as sensors and smart devices) than previous generations of wireless. A major milestone was reached in December 2017 when the first 5G specification was approved by the 3rd Generation Partnership Project and endorsed by many of the world's leading telecommunication equipment manufacturers and operators. The ITU's World Radiocommunication Conference 2019 aims to establish standards for spectrum management and harmonization for 5G.15 Some countries cannot wait, such as the United Kingdom, which earmarked some 3.4 GHz for 5G and auctioned it in March 2018. Operators in several countries have announced commercial deployments of 5G before the end of 2018.

Data-Driven Business Models

This section looks at how access to the internet and its content is priced and which parties earn the revenue. It examines the different ways for earning revenues from network access and content, including subscriptions (postpaid and prepaid), advertising, and transaction fees. It looks at the consequences of data-driven business models for the traditional pricing approach used by telecommunication operators, with potential impacts on network investment, market concentration, and net neutrality.

In reviewing alternative pricing structures, the difference between access to the internet and to its content varies in their financial significance. Users typically pay for access to the internet based on the volume of data they use, whereas content is not generally priced by volume. While a user may view free video streaming services that generate large amounts of traffic, an online shopping transaction generates little data traffic but, in aggregate, creates significant value for the seller. This contradiction poses a significant challenge for measuring the economics of data:

Consumer pricing at telecom network operators has evolved with technological change, from price per minute, to price per speed, to price per data consumed. Before the internet, voice was king and operators generally charged for it on a per-minute, metered basis. Before broadband, consumers mainly accessed the internet using dial-up connections with pricing as if it was a voice call (that is, minutes of use). The emergence of fixed broadband led to pricing by speed. With mobile broadband, it is harder to guarantee speeds because of different coverage ranges and a variety of handsets. Three main models exist for pricing data for mobile networks. One is flat rate, in which data usage is not metered but there is a fine-print data cap that, if surpassed, results in the speed being lowered. The second is postpaid, in which a certain amount of data is included with a monthly subscription.

The third, and most predominant, particularly in developing countries, is prepaid, in which the price is tied to a fixed amount of data. Prepaid must generally be used within a specific time, but does allow for flexibility in that a small amount of data can be purchased for a day or a week.

To justify investment in higher-bandwidth infrastructure, a number of telecom operators around the world have diversified into providing video services, enabling them to provide so-called triple-play offers (such as voice, data, and video). This has led to bundled offers in which all three services are offered together at a price higher than that for buying any single service alone, but at a discount compared with purchasing each service separately. Ironically, although traditional broadcast operators moved to protect their markets through legal challenges, neither they nor the telecom operators foresaw the greater threat looming from streaming video services.

The challenge today is that data traffic is exploding while the traditional revenue earner for telecom operators, voice and text, is declining as users shift to over-the-top (OTT) services delivered over the internet. Data traffic from smartphones surpassed voice traffic in 2011 and since then has grown at a tremendous rate, accounting for 96 percent of mobile traffic (figure 2.7, panel a). However, operator revenues have not kept pace. In 2015, voice still accounted for almost half (46 percent) of telecom operator revenue (figure 2.7, panel b), down sharply from two-thirds of operator revenue in 2010. Total telecom revenue has only been growing at 2 percent a year.

Figure 2.7Global network traffic and retail telecom revenue, selected countries

How to pay: Advertising-funded versus subscriber-funded networks

The internet has largely been characterized by "free" content. Although some sites, mainly news and streaming audio and video, charge subscriptions, the most popular websites are free. Most of these sites earn revenues from advertising through a two-sided market strategy of providing users with incentives to join their platform. As the internet audience has grown, firms are increasingly flocking to place ads on the web and smartphone apps.

The growth of digital advertising is astonishing. Global internet advertising accounted for more than a third of advertising expenditure in 2016, slightly behind television (figure 2.8, panel a).

Digital advertising spending is highly concentrated, with the vast majority going to two companies: Google and Facebook, who between them received US$106 billion in internet ad revenue in 2016, 64 percent of the global total (figure 2.8, panel b). Their share is growing, up 20 percentage points since 2010. They both have huge user bases, with each reporting more than a billion users of their services, which is attractive to advertisers. But reasons for placing ads on them differ, partly explaining why advertisers often put ads on both. According to one digital ad analyst, "Facebook believes the most important thing is identity in ensuring ad effectiveness . . . they know who you are and so much about you" whereas "Google believes identity is secondary to intent. What's important is what you want right now because advertising products and services fulfils a want or need".

Figure 2.8 Global advertising revenue

Digital advertising is causing a huge transfer of wealth from traditional advertising outlets (such as newspapers and radio) to internet companies. Telecom network operators have also not largely benefited from the value advertisers place on internet content. However, in a highly contested decision, the U.S. government decided in 2017 that ISPs can sell browsing histories to advertisers without the user's consent.

The concentration of advertising spending reinforces the large properties at the expense of the millions of smaller ones. It also threatens infrastructure investment, since the large websites sit on top of telecommunication networks, yet the networks do not necessarily receive advertising revenues. Although there are payments from content businesses to carriers for bandwidth, it is not clear that they make up for the huge amount of traffic generated. At the same time, if not transparent, the payments between content providers and telecommunication carriers pose net neutrality concerns, since a payment may imply an enhanced traffic service, to the detriment of other content providers.

Another concern is that as a few global sites thrive they are branching into areas they have no expertise in and for which their automated content controls pose challenges. One area is news, with traditional news outlets hit hard by a rapid decline in advertising. The rise of news on IT content sites is of particular concern, with objectivity and facts increasingly called into doubt given the explosion of sources. At the same time, legitimate news and information is sometimes blocked, illustrating the weaknesses of robotic software agents trying to determine what is appropriate. For example, Facebook blocked a 1972 Pulitzer Prize winning photo of a Vietnamese girl over concerns about nudity. The company was accused of abusing its power and the photo was later reinstated. Google has been under fire for placing ads next to extremist content on its YouTube video-sharing site. These examples have led to a growing argument that IT firms posting news stories should be subject to regulations similar to those that media firms face.

Advertising does support free services for users, which is particularly attractive for developing countries. These services include email, office applications, storage, and maps delivered over the cloud. The availability of free and legal cloud-based applications and services is putting a dent in software piracy, which has been dropping, and the focus has shifted from loss of revenues to the cybersecurity dangers of unlicensed software.

Besides advertising, some internet companies charge subscription fees, particularly those involved in content. One example is Netflix, which offers streaming television and film delivered over the internet. E-commerce sites such as Amazon, eBay, Rakuten, and Alibaba earn revenue in two ways. One is as a normal retailer, charging a markup over price. The second is when the platform is provided to third-party sellers, in which case the platform owner earns transaction fees.

Table 2.3 contrasts average revenue per user per month for the various internet payment models and for large internet-based companies. Prices have also been converted to purchasing power parity to adjust for differences in the cost of living. Subscription-based video viewing has the highest average revenue per user, and advertising on social media the lowest (but more subscribers). The telecom carrier in the list has the second highest average revenue per user (purchasing power parity).

Table 2.3 Average revenue per user from internet data, 2016

Some content companies are becoming involved in developing networks, feeling that infrastructure development is not keeping up with the vast growth in data traffic. A desire to capture more users from developing countries by making it easier for them to get online is also driving this effort.

Facebook and Google are investing in a variety of communication ventures.

At Google, this includes providing fiber internet access in several U.S. cities, offering fiber backbones and Wi-Fi in Ghana and Uganda, using hot air balloons to extend internet connectivity, and supporting the use of white spaces for making more spectrum available. It also controls Android, the leading mobile phone operating system.

Facebook has been looking at satellites, drones, and solar-powered airplanes for extending internet access, and is getting involved in networking gear. The growing involvement in data transmission by large content firms raises questions about the separation of carriage and content, with the possibility of a few companies dominating both internet content and access.

The rise of "over-the-top" service providers

OTT refers to data services provided over telecommunication networks. The impact of OTT on telecommunication operators comes from either competing with traditional revenue sources, such as voice calls and messaging, or depositing a lot of traffic on their networks. A lack of clear metrics makes understanding the OTT market a challenge. Some of the most popular OTT services have been purchased by larger companies that do not provide disaggregated financial or operating indicators. Nevertheless, considerable circumstantial evidence suggests the impact is significant.

Notable OTT providers include the following:

• WhatsApp, purchased by Facebook in 2014, offers messaging and calling services and claims to have about a billion users in more than 180 countries.

• Viber offers calling, video, and messaging services to more than 800 million people and is owned by Japanese e-commerce company Rakuten.

• Skype, owned by Microsoft, offers calling, video, and messaging services.

• Netflix provides paid video to 94 million subscribers around the world. It competes with telecommunication service providers that also offer video (such as IPTV). However, the bigger impact on the telecoms is the volume of traffic Netflix generates.

The impact of OTT has been particularly strong in what has long been a traditional profit center for telecommunication carriers: international voice telephone calls. By 2012, Skype was already handling a third of international telephone traffic. By 2016, OTT traffic using voice over internet protocol exceeded international traffic provided by telecommunication operators (figure 2.9).

Figure 2.9 International carrier and over-the-top traffic (billions of minutes)

Similarly, telecom carriers have seen a sharp fall in conventional messaging services (such as SMS) used on their networks. In 2015, an analysis of 17 countries found that the number of messages declined in 14 of them (figure 2.10).

Figure 2.10 Mobile messages, year-over-year change (percent), 2014-15

The danger is that if carriers cannot offset the loss of revenues, less money may be available for future infrastructure investments to handle the rapid increases in data traffic. The ITU has created a study group to examine this issue. The West African telecommunication operator Sonatel paints a dire picture, estimating that between 2016 and 2020, its losses from OTT in the international segment will be CFA francs 256 billion (US$432 million) in Senegal, CFAF 164 billion (US$277 million) in Mali, CFAF 79 billion (US$132 million) in Guinea, and CFAF 12 billion (US$20 million) in Guinea Bissau. It also estimates that taxes paid to the government and dividends for its shareholders will fall CFAF 243 billion (US$410 million).

Telecom operators have developed several responses to OTT. They have argued for regulating OTT providers that provide voice and text services in the same way that telecom operators are regulated. Some operators are developing their own OTT products. Others are including large bundles of their own offerings, such as free calls or text in packages. Many are diversifying into opportunities in areas such as cloud computing, IoT, and mobile money. Some operators are trying to do all of the above. Also relevant is taxation for OTT firms without a physical presence in the country in which they are providing service. Though they may offer the service for "free," digital advertising revenues often subsidize this. The lack of OTT taxation in most developing countries gives them a cost-structure advantage over domestic telecommunication operators.

Some OTT services compete with traditional telecommunication services such as voice and text. However, others provide a different challenge, particularly OTT video providers. These include those that provide television and films through subscription services such as Netflix as well as free services such as YouTube and Facebook, in which video posts are increasing rapidly. They not only compete with telecom operators that provide video services, but are also responsible for a substantial portion of traffic going over the networks. Netflix, YouTube, and Facebook are among the top traffic applications in most regions, comprising 42 percent of fixed-access and 34 percent of mobile-access, peak-period traffic (table 2.4). When all of the properties of Facebook (that is, Instagram, WhatsApp, and so on) and of Google (YouTube, Google Cloud, Google Market, and so on) are considered, they account for an even larger share. The two account for more than 60 percent of total traffic on Latin American mobile networks, for instance.

Table 2.4 Percentage of aggregate peak period traffic by region, 2015-16

The promise and perils of zero-rated services

Many internet applications are based on the so-called freemium model, in which consumers get basic features at no cost and can access premium functionality for a subscription fee. However, users must still pay for the data consumed using these applications. Zero-rated services provide access to certain content without it applying to a user's data cap. Some firms with desirable content, such as Facebook, have worked with operators, mainly in developing countries, to provide access to their services without its affecting a user's data allowance. At the same time, some operators are striking agreements to provide access to some services free to their customers. For example, T-Mobile in the United States does not charge data usage for customers on certain subscription bundles when they stream music. Other operators provide discounted access to a bundle of social networking services, but this is technically not zero rated, since users still pay for data access but at a discounted rate.

It is argued that zero-rated services provide a taste of a slimmed-down version of a service and users will eventually pay for access to the full internet. Text-only versions may also be relevant for users who do not have access to high-speed mobile internet and rely on slow 2G connections. One example is Facebook Zero, launched in 2010, providing access to a text-only version of the service. It has now been renamed the Free Basics service, available in more than 60 economies, with about 40 million users. In addition to Facebook, access is provided to other websites, such as Wikipedia, Accuweather, and Bing, as well as to local social impact sites for health and employment.

A variation on zero-rated services is sponsored data, in which companies pay for data usage if a user agrees to receive ads. Sponsored data is also used for companies to pay for their employees' mobile data usage for work. AT&T, a U.S. mobile operator, offers sponsored data in which users are not charged for usage if they access a sponsor's site. Syntonic, one of the AT&T sponsors, notes: "Millions of prepaid consumers ration their data, impeding discovery and exploration of mobile apps and content" and is expanding its product reach outside the United States to southeast Asia, India, and Mexico.

While providing free content seems commendable, only making certain content available runs contrary to the net neutrality principle of the internet. In February 2016, the Telecom Regulatory Authority of India issued a regulation prohibiting the use of what it called discriminatory tariffs for data services. The authority formed its decision based on "the principles of Net Neutrality seeking to ensure that consumers get unhindered and nondiscriminatory access to the internet. These Regulations intend to make data tariffs for access to the internet to be content agnostic". Several other countries have also banned zero-rated services. On the other hand, the United States Federal Communications Commission struck down net neutrality provisions in December 2017. The 3–2 vote by commissioners was along political lines in an allegedly "contentious and messy" public comment process. ISPs in that country are now allowed to block, throttle, and prioritize content. About 20 states have filed lawsuits against the ruling, and the United States Congress is considering overturning the ruling if it can muster the votes.

At the same time, it is argued that zero-rated services give an advantage to large companies to the detriment of new startups. As one report notes: "Ironically, if zero-rated services were available when large internet companies were startups, it is unlikely they would have scaled to the size they are now".

Importantly, cost is not the only or even, in many countries, the main barrier to internet access. Users often cite reasons such as no need or lack of skills as the reason they do not use the internet. In Thailand, 97 percent of those who do not use the internet said the main reason was because they did not know how to use it or it was unnecessary or a waste of time. In Brazil, 70 percent of those not using the internet cited a lack of interest as a reason. So it is not clear to what extent zero-rated services get new users online. As one report puts it: "Even with a zero-rated service, the user must still have a device and an active account with the operator that offers the zero-rated service. This raises the question of whether zero-rated services can bring people online who had not previously used the internet". The report looked at the impact of mobile data apps across eight developing countries, finding that 88 percent of users had already accessed the internet before using a zero-rated plan. This suggests that digital literacy challenges are arguably more important than affordability to get more people online.

Data Holes: Filling the Gaps

This section explains why data location, language, and limits are becoming more important than plain access. It also explores links between data and economic development.

From access to usage

Significant attention has been devoted to the uneven distribution of access to information and communication technology (ICT). As mobile phone penetration rises and access to the internet increases, the access gap is shrinking. Nine in ten people around the world were covered by a 2G cellphone signal in 2016 and 65 percent by 3G; by 2020 these figures are forecast to rise to 95 percent and more than 90 percent, respectively. The unconnected are increasingly those who are not interested in using or do not know how to use the internet rather than those who have no access or cannot afford to pay. Investment in digital literacy training is becoming as important as infrastructure.

The geography of data creation, distribution, and use is lopsided, resulting in a new global data divide. One manifestation of this gap is content concentration. More than half of the world's websites are in English (figure 2.11, panel b), yet only 984 million people speak English as a first or second language, 13 percent of the Earth's population. Another manifestation of the data divide is from where it flows. Just over one-third of IP traffic is generated by North America, with only 5 percent of the world's population (figure 2.11, panel a). On the other hand, the Middle East and Africa, home to 19 percent of the world's population, only generate 3 percent of global IP traffic.

Figure 2.11 Global internet protocol traffic and websites by language

New metrics of the data age

Data holes are reflected by uneven data consumption across communities, regions, and nations. The amount of data used per smartphone – measured as gigabytes of data per month or GB/user – varies tremendously. Smartphone users in North America consumed almost four times more data than those in the Middle East and Africa (figure 2.12, panel a). Average global use is forecast to grow more than fivefold between 2016 and 2022, from 1.9 GB per month to 11. Within North America, U.S. mobile broadband subscribers use more than 1.8 times as much data as their neighbor to the north, Canada, and 3.6 times more than their neighbor to the south, Mexico (figure 2.12, panel b).

Figure 2.12 Mobile data usage

Data usage is driven by factors such as coverage and device, with pricing a major influence. Data pricing varies significantly throughout the world, measured by the metric of price per GB per month or for comparability, US$/GB (figure 2.13, panel a). In absolute terms, average price ranges from US$5 per GB per month in South Asia to US$28 in high-income OECD nations. However, in relative terms, high-income OECD nations have the cheapest prices (0.9 percent of GDP per capita) compared with 12 percent in Sub-Saharan Africa. Prices vary significantly in Sub-Saharan Africa, with relative data prices ranging from a little over 1 percent of gross national income per capita in Mauritius to 45 percent in Zimbabwe (figure 2.13, panel b).

Figure 2.13 Mobile data pricing

Being data starved is a constraint when it comes to rich multimedia educational, health, and livelihood content. However, many useful activities require just narrowband: a quick e-commerce transaction, a text message to check produce prices, or a phone call in an emergency. Hourly, daily, and weekly prepaid options can enhance affordability in these circumstances.

Data and economic development

Could the data divide be affecting economic growth in developing nations? Various studies have looked at the impact of ICTs on economic growth. As businesses and consumers obtain more high-speed connectivity, they have realized important benefits in terms of efficiency, new businesses models, market information, and so on. Some research has focused on the impact of data on the economy. Four studies looking at public sector open data found impacts ranging from 0.4 percent to 4.1 percent of GDP. A European Parliament report states that big data and the data-driven economy will bring 1.9 percent in additional GDP growth by 2020.

A Deloitte study suggests that data usage affects economic growth. Based on mobile data usage for 14 countries between 2005 and 2010, the study found that a doubling of mobile data consumption added 0.5 percentage point to GDP growth a year.

While the study suggests an econometric link between data consumption and economic growth, the exact reasons seem fuzzy. It is puzzling, given that most internet traffic is video entertainment, which is not likely to have a tremendous economic effect. Other studies suggest that it may not be the quantity of data that is important, but rather the value of the data. In many developing countries, economic impacts have been noted from basic cell phone voice calls and narrowband 2G applications such as text messaging or mobile money, which do not use much data. For example, a study of grain markets in Niger found that prices dropped 3 percent after the introduction of mobile phones because of better access to market information. A study analyzing the economic impact of mobile money in Kenya found its use decreases prices of competing money transfer services and increases levels of financial inclusion. An econometric analysis on the impact of telecommunications in Senegal found no statistically significant effect from broadband; on the other hand, plain mobile communications had a significant contribution, with each percentage point increase in mobile penetration contributing 0.05 percent to GDP. These findings suggest that the data nuggets are small, often lost in the sea of video and social media traffic, and sometimes not even transmitted over the internet.

Conclusions: Toward Sustainable National Data Ecosystems

As the universe heads inexorably into the data era, there are winners and losers. Consumers have access to "free" content and services in exchange for their personal information and time spent posting information. It has also been great for content platforms that get free user-generated content and even more valuable, their personal information that is sold to digital advertisers and data analytics firms. On the other hand, the emerging data economy is requiring adjustments for telecommunication carriers who are finding it difficult to fund investment needed for rising data use. The money they are making from data access has not offset falling revenue from traditional sources due to OTT. Governments are finding it increasingly challenging to deal with concerns such as net neutrality, privacy, computer crime, and false or incendiary information on the internet, as well as automated platform censorship.

The globalized nature of the internet is its beauty, but also its peril. Users can access content from Argentina to Zambia using free platforms to store data and run applications. SMEs in developing countries have benefited from free online tools and global platforms to increase their visibility. This "free lunch" has resulted in just two U.S. companies – Google and Facebook – dominating the platforms by which many of the world's internet users interact, earning the majority of online ad revenue, controlling vast amounts of personal data, and generating much of the traffic. They are also extending their operations horizontally and vertically, from online shopping to the provision of telecommunication services. This has made some governments anxious about the power a few U.S. companies wield over the internet:

The rise of dominant internet platforms affects the development of national data ecosystems. Developing country telecom operators not only pay for a physical link abroad, they also need to pay for data traffic to be exchanged for transmission to an overseas hyperscale data center. Local digital businesses struggle to develop new applications and services already dominated by a few free platforms that have achieved giant scale because of network effects. The development of local internet infrastructure facilities such as IXPs and data centers suffers, since so much content resides abroad.

The challenge for many lower-income countries is how to develop a relevant and sustainable data ecosystem in the current environment. Much of the data consumed around the world is entertainment oriented. Yet governments need development-oriented data to enhance social and economic growth.

A starting point is boosting linguistically and contextually relevant local content. This needs to be accompanied by investment in infrastructure such as fiber-optic backbones and data centers to bring data closer to users. Applications and services that can enhance health and education, such as telemedicine and online learning, need to be implemented rather than talked about. National infrastructure deployment and the take-up of local content and services can be encouraged by taking a page from mobile tariff structures. Access to locally hosted sites and services such as e-government can be stimulated with a low "on-net" internet access price, particularly since access to locally stored content is cheaper than content stored overseas. Furthermore, digital literacy has to be boosted so that taxi drivers can use GPS and not just play smartphone games, and SMEs need to move from streaming music in their shops to using e-commerce. Digital scientist skills are needed, so instead of being overwhelmed by data, developing countries can analyze it and put it to good use. And rather than digital advertisers using personal data to sell something, developing country digital scientists can use data to pinpoint the locations of those living in poverty to better target assistance.

In short, developing nations need to leverage data to drive development through locally relevant content and a thriving employment-generating digital ecosystem. This will require better understanding of data's potential, investment in core infrastructure such as data centers and IXPs, and development of data-driven development applications and services.