International Think-Tank on Innovation and Competition
Multi-sided Platforms in the Third Industrial Revolution
February 15th, 2007
In a recent important book, Evans, Hagiu and Schmalensee (2006, Invisible Engines, MIT Press) have emphasized the crucial role that software platforms are playing in shaping our economies, the functioning and the development of many sectors, and ultimately our way of living. These "invisible engines", as they call them, power not only the PC industry but also other industries as those associated with mobile phones and other handheld devices, video games, digital music, and (with strong externalities for the rest of the economy) on-line auctions, online searches and web-based advertising. Their convincing claim is that, as the steam engine was at the basis of the first industrial revolution (1760-1830) and electric power at the basis of the second industrial revolution (1850-1930), microprocessors and software platforms are at the basis of the third industrial revolution (since 1980), which started with the introduction of commercial PCs and had a second phase starting in 1995 with the Internet.
A software platform is a software program that makes services available to other software programs through Application Programming Interfaces (APIs). Examples are the operating systems running on PCs as Windows, Mac OS or Linux, those employed by videogame consoles as the Sony one for PlayStation or Windows 2000 for the Xbox, Palm OS for personal digital assistants (PDAs), RIM for the BlackBerry, iPod OS for the Apple iPod, the Symbian operating system for cellular phones - Symbian is a joint venture founded by Nokia, Ericsson and Motorola, which left it in 2003; it is currently owned by Ericsson (15.6%), Nokia (47.9%), Panasonic (10.5%), Samsung (4.5%), Siemens AG (8.4%), and Sony Ericsson (13.1%).
To understand the peculiarities of software platforms in general it is convenient to focus briefly on the main functions of PC operating systems. The main one is to serve as a platform on which applications (such as spreadsheets or word processors) can be created by software developers. Operating systems supply different types of functionality, referred to as system services, that software developers can call upon in creating their applications. These system services are made available through APIs. When an application calls a particular API, the operating system supplies the system service associated with that API by causing the microprocessor to execute a specified set of instructions. Software developers need well-defined platforms that remain stable over time. They need to know whether the system services on which their applications rely will be present on any given PC. If they did not, then software developers would have to write the software code to provide equivalent functionality in their own applications, generating redundancy, inefficiency and a lack of interoperability. Moreover, modern OSs provide a user interface, the means by which a user interacts with his computer. User interfaces for computers have evolved dramatically over the last decades, from punch card readers, to teletype terminals, to character-based user interfaces, to graphical user interfaces, first introduced by Apple with Macintosh. Finally, operating systems enable users to find and use information contained in various storage devices: local ones, such as a floppy diskette, a CD-ROM drive or the hard drive built into a PC, or remote ones, such as local area networks that connect computers in a particular office, wide area networks that connect computers in geographically separated offices, and the Internet.
Over time, the OS of Microsoft became the most popular because Microsoft continually added new functionality to the OS and licensed it to a wide range of computer manufacturers with extremely aggressive pricing strategies. Microsoft recognised early on that an OS that served as a common platform for developing applications and could run on a wide range of PCs would provide substantial benefits to consumers. Among other advantages, development costs would fall and a broader array of products would become available because products could be developed for the common platform rather than for a large number of different platforms. By providing a single OS that ran on multiple brands of PCs, Microsoft enabled software developers to create applications, confident that users could run those applications on PCs from many different computer manufacturers. In addition, applications developed for a single platform are more easily interoperable because they rely on the same functionality supplied by the underlying OS. The winning strategy of Microsoft was the creation of these network effects between hardware producers, software developers and consumers: computer manufacturers benefit because their PCs can run the many applications written for Windows and because users are familiar with the Windows user interface; software developers benefit because their applications can rely on system services exposed by Windows via published APIs and because they can write applications with assurance that they will run on a broad range of PCs; consumers benefit because they can choose from among thousands of PC models and applications that will all work well with one another and because such broad compatibility fosters intense competition among computer manufacturers and software developers to deliver improved products at attractive prices.
Software platforms deal with multiple sides. Microsoft deals with at least three: consumers, software developers and PC manufacturers. Apple produces hardware internally, hence it deals with the remaining two sides: consumers and software developers. Sometimes relationships are even more complex, as in the platform ecosystem for smart mobile phones where, beyond OSs, software developers and handset makers, there are network operators (Vodafone, NTT DoCoMo, T-Mobile, Tim,..) to play a coordinating role and even competition between layers is strong.
In the presence of multiple sides with network effects between them, the choice of which ones should be charged more to use the platform is not simple. Rochet and Tirole (2003) and others have been the first to notice that software platforms, as other similar multi-sided platforms, give rise to market structures that are quite different from traditional ones. For simplicity, here we will refer to two-sided platforms, which connect two sides in such a way that for each side the valuation of the interactions with the other side depends on the number of agents on the others side. These network externalities, and in particular the non neutral impact of the pricing structure on both sides (and hence on these externalities) distinguishes a two-sided market from a traditional one-sided market with different consumers (and possibly price-discrimination between them).
An analogous situation to software platforms emerges in many completely different contexts. A classic example is given by newspapers. They are sold to readers, but they also sell advertising space to advertisers: the reader is not only a "customer" of the newspaper, the reader is also a supplier of "eyeballs" that the newspaper sells to advertisers. Here, network effects emerge because advertisers (the sellers for the platform) value their advertising more in a newspaper when the number of its readers (the buyers of the platform) is higher (the effect in the other direction may exist but is typically less important). This has crucial consequences on the pricing structure since a low price for the readers increases the number of sold copies and hence the value of advertising. Such a phenomenon is even stronger when the newspaper is competing with other newspapers, and a low price reduces the readers of competing newspapers and the value of advertising on these competing newspapers.
Other two-sided platforms include other media networks as television channels, real estate agencies, traditional auction houses, shopping malls, night clubs and dating clubs, payment card systems, telephone networks and many industries of the New Economy as those related with video game consoles, smart phones, digital music, PDAs, i-Mode (this was created by the main Japanese mobile network operator, DoCoMo, to connect mobile phones with Internet content providers and application developers. Both the US and the EU lag behind in the development of a similar platform), search engine-based portals (like Google), on line messaging (like Yahoo!), on line chatting (like Skype), on line social networks (MySpace or asmallworld), on line academic articles (JSTORE or SSRN), on line shopping (Amazon) and on line auctions (eBay). In many of these markets, multi-homing on at least one of the two sides is common: people often buy more than one journal or watch more TV channels (as companies advertise on multiple medias), hold multiple credit cards (as merchants accept multiple cards) and software developers prepare applications for multiple OSs (while individuals typically use only one).
In each one of these examples, network externalities are crucial to the success of a software platform, and the pricing structure toward buyers and sellers is crucial to the creation of these network effects. In particular, a platform typically ends up charging one of the two sides less than the other, taking into account demand elasticities and which side values the other side more: the aim is to get on board as many agents as possible from one side, so as to increase the value of the platform for the other side and earn more revenue from it. For instance, when the price is the strategic variable, it is optimal to charge the side whose demand is more elastic relatively more because this allows one to maximize the total volume of interactions – but this is the exact opposite of what happens normally, when it is optimal to charge less consumers with a less elastic demand! Prices will be constrained downward when there are competing platforms (especially in case of multi-homing), but the general principles on a balanced price structure between the two sides remain unchanged. In extreme cases, one side may even receive its goods or its services for free or even be subsidized so as to maximize earnings from the other side.
The above theoretical implications are surprisingly confirmed by what happens in the above mentioned two-sided markets, whose companies typically settle on pricing structures that are heavily skewed toward one side of the market, or in other words adopt what is sometimes called a "divide and conquer" strategy. Newspapers, television networks and even websites typically earn more from advertisers than from consumers, real estate agencies earn more from sellers (or from landlords) than from buyers (or renters), auction houses from sellers rather than from the buyers, shopping malls from stores rather than from the shoppers, night clubs from men rather than from women, payment card companies from merchants rather than from cardholders, phone operators (often) from call makers rather than from receivers, video game platforms from royalties on game developers rather than from buyers of consoles (that are often sold below cost), while most of the other software platforms, including PC operating systems, earn more from end users rather than from software developers. This happens in different ways however: Microsoft licensees Windows, Palm and Symbian license their OSs to manufacturers of PCs, PDAs and cellular phones, while RealNetworks licenses access to digital content and Apple sells PCs and iPods, but none of these companies charges content owners (Apple and RealNetworks actually pay them) or software developers (which are typically subsidized).
Notice that, in spite of the network effects, most of these two-sided markets are also characterized by a certain degree of fragmentation between platform providers (real estate agencies, dating clubs, traditional auction houses), often associated with a certain degree of differentiation (newspapers, TV channels and other medias, shopping malls). Only when technological innovation is particularly important and fixed costs of investment in R&D are high (while marginal costs of production are particularly low), the number of competing platforms is endogenously reduced, as in the above mentioned markets of the New Economy (but tipping on a single leader rarely happens, especially when product differentiation and multi-homing have a role, as for video games). Nevertheless, even in these cases, competition for the market can be quite effective and induce periods of persistent leadership with occasional replacement of the leader: pathbreaking innovations (or "killer applications") is what competitive firms really look for.
For instance, in the console video game industry, sequential innovations brought to leadership a number of companies as Atari (that reached 80% share of the market in 1980), Nintendo (90% of the market in 1987), Sega (leader in the early 90s), Nintendo again (in the mid 90s) and Sony with the PlayStation in different improved versions (during the last decade): recently Microsoft Xbox started gaining market shares, and Nintendo is still active, but the leadership of Sony (58% market share in 2004) does not appear under threat yet, especially after the recent successful launch of PlayStation 3. Similarly, after a number of unsuccessful attempts by many companies, Palm's PDA gained success and leadership in the market for OSs for organizers thanks to a simple handwriting recognition system (65% market share in 2000) until Microsoft competing platform and other handheld devices, including Blackberry and (in perspective) Apple's iPhone, gained success.
