Profiting from Innovation in the Digital Economy PDF

Research Policy 47 (2018) 1367–1387 Contents lists available at ScienceDirect...

Research Policy 47 (2018) 1367–1387 Contents lists available at ScienceDirect Research Policy journal homepage: www.elsevier.com/locate/respol Profiting from innovation in the digital economy: Enabling technologies, T standards, and licensing models in the wireless world David J. Teece Institute for Business Innovation, Haas School of Business, U.C., Berkeley, United States A R T I C LE I N FO A B S T R A C T Keywords: The value-capture problem for innovators in the digital economy involves some different challenges from those Appropriability in the industrial economy. It inevitably requires understanding the dynamics of platforms and ecosystems. These Complementarity challenges are amplified for enabling technologies, which are the central focus of this article. The innovator of an General-purpose technology enabling technology has a special business model challenge because the applicability to many downstream Licensing verticals forecloses, as a practical matter, ownership of all the relevant complements. Complementary assets Platform Standards (vertical and lateral) in the digital context are no longer just potential value-capture mechanisms (through asset Technology policy price appreciation or through preventing exposure to monopolistic bottleneck pricing by others); they may well be needed simply for the technology to function. Technological and innovational complementors present both coordination and market design challenges to the innovator that generally lead to market failure in the form of an excess of social over private returns. The low private return leads to socially sub-optimal underinvestment in future R&D that can be addressed to some extent by better strategic decision-making by the innovator and/or by far-sighted policies from government and the judiciary. The default value-capture mechanism for many enabling technologies is the licensing of trade secrets and/or patents. Licensing is shown to be a difficult business model to implement from a value-capture perspective. When injunctions for intellectual property infringement are hard to win, or even to be considered, the incentives for free riding by potential licensees are considerable. Licensing is further complicated if it involves standard essential patents, as both courts and policy makers may fail to understand that development of a standard involves components of both interoperability and technology development. If a technology standard is not treated as the embodiment of significant R&D efforts enabling substantial new downstream economic activity, then rewards are likely to be calibrated too low to support appropriate levels of future innovation. 1. Introduction across the upstream and downstream levels of competition in an in- dustrial system. In this paper, I look anew at the Profiting from Innovation (PFI) This paper considers the impact of digital convergence, the growing framework laid out in Teece (1986, 1988a, 2006). The questions ad- importance of platforms and ecosystems, and the amplified problems dressed in the earlier treatments—what determines which firms profit associated with enabling technologies. While adding some complexity, from an innovation, and which firms earn only meager (and possibly this wider aperture of inquiry turns out to reinforce the key elements of negative) returns—have enduring relevance for both management and PFI. Intellectual property, the nature of knowledge, complementary public policy.1 If anything, the importance of the issues is amplified as assets, standards, and timing all remain center stage. What is brought “digital convergence” and “digital disruption” gain pace. In particular, into sharper focus and requires additional granularity are the different the mega-convergence of certain industries that is being driven by the types of complementary assets and the ways they impact the capture of merging of wireless and Internet technologies requires that one open value from innovation when digital platforms are at issue. Attention is the aperture of business and economic inquiry from the innovation of also provided to how general-purpose technologies and a collaborative individual products and processes to innovation within ecosystems and standards development process enable downstream innovation. E-mail address: [email protected]. 1 Before PFI, capturing value was often considered just a matter of new-product pricing strategy. Setting prices for innovative products and services remains important, but value capture depends on more fundamental considerations. Empirical support for the PFI framework is well-established. Cohen et al. (2000) cite survey evidence showing that, while trade secrets and patents help support appropriability, complementary assets and capabilities are of comparable importance (See their figures 1–4). The framework is also frequently used in applied research (e.g., Desyllas and Sako, 2013) and continues to be taught in business schools (e.g., Tietz and Parker, 2010). https://doi.org/10.1016/j.respol.2017.01.015 Received 23 August 2016; Received in revised form 18 January 2017; Accepted 20 January 2017 Available online 21 April 2018 0048-7333/ © 2018 Published by Elsevier B.V. D.J. Teece Research Policy 47 (2018) 1367–1387 Business model issues with respect to value capture via licensing are even when the innovator “wins” (i.e., takes the largest piece of the explored. The particular challenges of capturing value from enabling available private returns), spillovers are still considerable (Mansfield technology are recognized, and policy implications are highlighted. et al., 1977; Griliches, 1992). Lichtenberg (1992) found that the na- In short, this essay considers the value capture impacts of changes tional rate of return (in terms of productivity) from private R&D in- wrought by the digital revolution, the activity of standards organiza- vestment was about seven times as large as the return from investment tions, the presence of enabling technologies, and the growing im- in plant and equipment. A survey of previous studies showed that the portance of complementary assets and technologies in the information social rate of return to private R&D was usually found to be about twice and communication technology (ICT) sectors and beyond. These phe- that of the private return (Hall et al., 2010). A more recent study de- nomena have enhanced salience in the wake of the digital revolution termined that, even taking rent capture (what the authors call “negative and the associated convergence of industries that is described in business stealing effects from product market rivals”) into account, Appendix A. social returns to R&D are at least twice as high as private returns (Bloom et al., 2013). A pioneering scholar of R&D spillovers summed it up this way: “there has been a significant number of reasonably well 2. PFI revisited done studies all pointing in the same direction: R&D spillovers are present, their magnitude may be quite large, and social rates of return The Profiting from Innovation (PFI) framework (Teece, 1986) was remain significantly above private rates” (Griliches, 1992, p. S43). launched thirty years ago in a very different technological and business The fundamental imperative for profiting from an innovation is environment than most companies face today. It is worth revisiting that, unless the inventor/innovator moves down an improvement path periodically in order to see if improvements are possible. I first did so and enjoys strong natural protection against imitation and/or has twelve years ago (Teece, 2006), when I sketched a number of ela- strong intellectual property protection, then potential future streams of borations and extensions in response to shifts in the environment since income are at risk. The relevant appropriability regime is thus critical to 1986. These included further development of the multi-invention con- shaping the possible outcomes. text for innovation, the incorporation of a richer understanding of Appropriability regimes, while partly endogenous (Pisano and Teece, network effects, and the consideration of business models engendered 2007), can, theoretically, be “weak” (innovations are difficult to protect by the launch of the Internet. I also discussed the growing importance of because they can be easily codified and legal protection of intellectual complementary technologies, network effects, and supporting infra- property is ineffective) or “strong” (innovations are easy to protect be- structure. Others have extended the framework in various ways, in- cause knowledge about them is tacit and/or they are well protected cluding how to take into account industry architecture (Jacobides et al., legally). The fact that empirical studies establish that the social returns to 2006). innovation are generally considerably greater than private returns is In the intervening decade, as elaborated in Appendix A, the techno- prima facie evidence that appropriability is almost always difficult. The business environment has shifted still further. The Internet is no longer challenge is larger if the innovations in question are basic research results a utility consulted just from user desktops. It is increasingly pervasive, or general-purpose/enabling technologies. While appropriability regimes accessed interactively by users on the go and extended to sensor- for some downstream digital businesses (e.g., Facebook) are strong,3 equipped terminals anywhere and everywhere. Means of communica- those for upstream providers of enabling technology are often quite tion have also evolved, from phone and email toward messaging apps weak. Business models cannot rely heavily on intellectual property (IP) that also serve as portals for shopping and a host of other services. to capture value because IP is generally not self-enforcing; patent in- PFl addressed a puzzle that had not been well explained in the fringement and trade secret misappropriation must be identified, then previous literature, namely: why do highly creative, pioneering firms negotiated or litigated, often at great expense. The net result is that free often fail to capture much of the economic returns from innovation? riding is common and patent licenses have to be negotiated under the Apple’s iPod was not the first standalone MP3 player, but it has shadow of continued infringement. Patents rarely, if ever, confer strong dominated the category for more than a decade (Cole, 2013). Merck appropriability, outside of special cases such as new drugs, chemical was a pioneer in cholesterol-lowering drugs (Zocor), but Pfizer, a late products, and rather simple mechanical inventions (Levin et al., 1987). entrant, secured a superior market position with Lipitor (Hilzenrath, Patents can also, in some cases, be “invented around” at modest costs 1998). At first glance, it is tempting to say that these examples reflect (Mansfield, 1985; Mansfield et al., 1981). the result of Schumpeterian gales of creative destruction where winners Often patents provide little protection because the legal and fi- are constantly challenged and overturned by entrants. But the cited nancial requirements for upholding their validity or for proving their cases and countless others involve mostly incremental/imitative en- infringement are high, or because, in many countries, law enforcement trants rather than the radical breakthroughs associated with enabling for intellectual property is weak or nonexistent. In terms of preventing and general-purpose technologies. imitation or bringing infringers into licensing arrangements, a patent is The focus of the 1986 PFI article was on a single, autonomous in- merely a passport to a journey down the road to enforcement and po- novation that was commercially viable (i.e., technological uncertainty tential royalty streams. Of course, a large portfolio of patents can prove (Rosenberg, 1982) was assumed to be low). The paper thus side-stepped valuable for cross-licensing deals with rivals that help reduce the like- one question—why inventions so often fail to succeed in (or even reach) lihood of costly litigation. To help with appropriability, the inventor of the market—and focused instead on how the spoils are divided once a core technology can also seek complementary patents on new features positive net value is within sight. The paper also focused on product and/or processes, and, in some cases, on designs. innovations rather than process improvements, creative output, or A further complication in recent times is the emergence of cybertheft other valuable intellectual capital. These and other simplifications and other cybersecurity problems. Being secure to market must often made the analysis tractable. In this paper, I maintain the focus on value take precedence over being first to market. The adoption of secure coding capture rather than value creation, but I now consider how the pro- practices can reduce the number of exploitable vulnerabilities in software blems for the innovator differ in the case of enabling technologies and products and in hardware products with embedded software. The advent in the presence of multi-level platforms and ecosystems.2 The PFI framework provides an explanation as to why some in- novators win in the marketplace while others lose out—often to tech- 3 Facebook sells advertising space on its site, which it offers to personalize to help nologically weak imitators. The framework also made it apparent that, advertisers target specific groups of users based on their stated and observed character- istics. Its ad revenues grew rapidly after it developed a mobile app. Like most consumer- facing tech firms, its business relies completely on upstream digital and wireless enabling 2 See Michel (2014) for additional discussion of value-capture strategies for innovators. technologies. 1368 D.J. Teece Research Policy 47 (2018) 1367–1387 of the massively interconnected “Internet of Things” is blurring distinc- The aggregate effect of a GPT occurs over a protracted length of tions between critical infrastructure, regulated devices, and consumer time. Its invention is often the result of collaboration between in- products in a way that increases susceptibility to cyber threats. dividuals with disparate skills. Rosenberg’s (1963) study of the machine The additions I will sketch out in this paper do not invalidate the tool industry makes these mechanisms apparent. The inventions may logic of the PFI model. The core independent variables in the PFI not be immediately identifiable as GPTs; they might be enabling tech- model—the strength of the appropriability regime, complementary as- nologies or less until improvements occur and new uses are discovered. sets, complementary technologies, standards (and associated installed For example, when the laser was invented around 1960, it was of sci- base effects), and timing—are more relevant than ever for a world entific interest but had no obvious application. Today, lasers are ubi- where firms increasingly offer platforms to which complementors can quitous, being implemented in applications ranging from supermarket add incremental value with their own innovations. Giving due con- checkout stands to military battlefields, with other uses in surveying, sideration to platforms, standards, and ecosystems complicates the si- medicine, telecommunications, manufacturing, entertainment, and tuation somewhat, but hopefully not overly so. Peter Drucker once more. wrote that “Truly unique events are rare … All events but the truly The widespread applicability of a GPT will often require a decade or unique require a generic solution” (Drucker, 1967, p. 93). It can be more to become evident, but over time the cumulative effects can be argued that digital convergence and multi-level ecosystems and plat- significant. There have been pioneering cliometric studies on the social forms are creating complex strategic decision environments that can benefits of GPTs such as steam engines (von Tunzelmann, 1978), British generate unique, emergent, and rare problems. While the new land- railways (Hawke, 1970), and American railroads (Fogel, 1964; Fishlow, scape to be navigated is indeed daunting and exceptional, the appli- 1965). Hawke (1970) showed that the railroads saved between 7% and cation of the (expanded) PFI framework can nevertheless still yield 11% of British national income in 1865. valuable insights. 3.1.2. Enabling technologies 3. Expanding the PFI framework The threshold for a GPT is very high, but there’s a similar category, enabling technologies (present but not well defined in the literature), The brief description advanced earlier (and expanded in Appendix that can be thought of as junior GPTs, meeting criteria (2) and (3), A) of the digital revolution has highlighted elements of PFI that need above, but not necessarily having measurable economy-wide impacts.6 additional elaboration. It also suggests the need to look further at While the list of GPTs is relatively short, there are hundreds, if not general-purpose/enabling technologies and the level of industry at thousands, of enabling technologies. Each one might not be thought of which innovation is taking place. The framework naturally speaks to as “radical” or a “game changer” by economic historians endeavoring to the special difficulty associated with appropriability when general- understand economic growth, but they are nevertheless important to purpose and/or enabling technologies are being developed and de- particular firms and industries. They can often be disruptive to the ployed. A more granular view of standards and complementary assets is status quo and generate very considerable economic benefit and social also presented, along with the analysis of related ecosystem and busi- surplus. ness model issues. The European Commission has identified six “key enabling tech- nologies” that are non-software research fields (micro and nanoelec- tronics, nanotechnology, industrial biotechnology, advanced materials, 3.1. General-purpose technologies, enabling technologies, & the photonics, and advanced manufacturing) said to underpin innovation in appropriability challenge products across many industries and to be important for addressing societal challenges (Commission of the European Communities, 2009). 3.1.1. General purpose technologies (GPTs) They have been prioritized for investments as part of Europe’s in- Bresnahan and Trajtenberg (1995; see also Bresnahan, 2012) in- dustrial policy (European Commission, 2017). troduced the important concept of general-purpose technologies In some cases, enabling technologies can have quite massive effects. (GPTs), which they identify by three characteristics. GPTs (1) are per- Containerization of cargo shipping, which greatly reduced the trans- vasive, i.e., in wide use; (2) are capable of ongoing technical im- portation cost of many types of products, had a large economic impact, provement; and (3) enable complementary innovations in application even though container technology has not advanced much technologi- sectors. In other words, GPTs have economy-wide effects, get even cally over the years, which possibly disqualifies it as a GPT but not better over time, and spawn other innovations because invention in one necessarily as an enabling technology. An econometric study by area triggers discoveries and creates opportunities elsewhere. For in- Bernhofen et al. (2016) found, after controlling for the effect of trade stance, the first commercially available microprocessor was developed liberalization agreements, that the adoption of container shipping over at Intel to reduce the number of components in a Japanese desktop the period 1967–1982 by a group of 22 OECD countries raised the value calculator. of trade by 1240%, swamping the impact from tariff reductions. An- While some non-GPT technologies may possess each of the identi- other significant enabling technology is 3G/4G wireless, which has fied characteristics to some extent, a GPT will be distinguished pri- enabled mobility for Facebook, location-sensitive mapping, and marily by its cumulative economic impact (Jovanovic and Rousseau, streaming media, among numerous other applications. 2005). It can be either endogenous or exogenous to the economic An enabling technology can be used to drive technological change system (Lipsey et al., 1998). A GPT can be a product, a process, or an in an industry. The utility of the enabling technology is not exhausted organizational system. Examples that Lipsey et al. (1998) consider by embedding it in a single product or even a single system; it can be “clear and dramatic” are limited to printing, bronze, made-to-order used by a range of downstream customers for their own products and materials, the waterwheel, steam power,4 electricity, the internal services. Enabling technologies and GPTs were not the focus of PFI, combustion engine, railways, motor vehicles, lasers, and the Internet. CRISPR, the metonymic name for a technology that can change genes within living organisms, can no doubt be added to the list.5 6 Enabling technology is closely related to “generic technology,” which, according to the preferred definition in Martin (1993: 51-2), is “technology the exploitation of which will yield benefits for a wide range of sectors of the economy and/or society.” This is 4 Rosenberg and Trajtenberg (2004) present the Corliss steam engine as an important distinct from “generic knowledge” (Nelson, 1989) that comes from hands-on experience nineteenth-century GPT. and is often a byproduct, rather than a goal, of formal innovation. Generic knowledge 5 See Cohen (2017) for a brief description of the early scientific and commercial history generally cannot be protected as intellectual property, while the opposite is true for of CRISPR. generic (or enabling) technology. 1369 D.J. Teece Research Policy 47 (2018) 1367–1387 which looked at commercially viable product innovations and im- As noted, a probable consequence is dramatic underinvestment in ac- plicitly assumed that only a narrow range of inventions was in- tivities likely to spawn enabling technologies and GPTs. corporated into a single product. Bresnahan and Trajtenberg (1995) identified two ways in which the One enabling technology (which might eventually become a GPT) is social benefits of a GPT may be curtailed by appropriability issues. First, artificial intelligence (AI). AI encompasses a range of software techni- the capture of a GPT’s value from downstream sectors may be limited ques employed to teach computers to sense, reason, interpret, com- by business or regulatory barriers. Second, the uptake of the GPT by municate, and make decisions in a human-like manner. It gained cur- multiple downstream sectors may be too low, in the sense that if these rency in the 1950s, but needed cheaper and faster computing power, sectors were coordinating their activity, they would see that faster especially developments in the complementary hardware technologies uptake of the technology would support its more rapid improvement by of processors and memory, to become commercially interesting. AI- the upstream innovator.8 The practical significance of this second point based technologies can already recognize faces, understand speech, and is that GPTs develop faster when there’s a large, demanding, and in- drive vehicles. Regulators use AI to sort Internet chatter and zero in on come generating application sector, such as the U.S. Defense Depart- fraud, illegal activity, and terrorist plots. The applications for AI are ment’s purchases of early transistors and microprocessors, which avoids limited only by economics and imagination, and its cost continues to the need to coordinate and serve many smaller user groups. fall as it becomes available as part of the capabilities offered for hire by Helpman (1998, p. 4) likewise noted that a lack of (inherently dif- the providers of “cloud” (online) computing. ficult) coordination can lead to undersupply of GPTs and improve- A closely associated technology is machine learning, a category of ments: software tools with a relatively narrow goal of learning known prop- GPTs introduce two-types of externalities: one between the GPT and erties from data with a minimum of pre-programmed rules. One sub- the application sectors; another across the application sectors. The type, deep learning, performs abstract tasks such as recognizing images, former stems from the difficulties that a GPT inventor may have in using neural networks and other tools that try to mimic the operation of appropriating the fruits of her invention. When institutional condi- the human brain. Deep learning helps Google improve the search results tions prevent full appropriation, the GPT is effectively underpriced it provides users and drives the recommendation engines used by many and therefore, undersupplied. The latter stems from the fact that, online retailers. The most viable business models for capturing value since the application sectors are not coordinated, each one condi- from machine learning are likely to be firm-specific because an algo- tions its expansion on the available general-purpose technology. But rithm is likely to learn faster and produce more accurate outputs when if they coordinated a joint expansion, they would raise the profit- it is narrowly focused (Norton, 2016). ability of the GPT and encourage its improvement. A better GPT Enabling technologies vary not only in type but in ownership. They benefits them all. may be owned or controlled (in part or in whole) by private companies, universities, or consortia. Many software-based enabling technologies, A key implication is that companies which invent or improve GPTs such as deep learning, are often available as open-source software. The have great difficulty appropriating the value created by their invest- use of technology from the open source community, where appro- ment, because management is challenged in finding workable business priability is inherently limited, places a premium on the ability of the models to capture value. Owners of enabling technologies face similar licensee to apply PFI principles, such as by assembling valuable com- challenges. As a result, underinvestment in GPTs and enabling tech- plements. In the case of deep learning, for example, this requires access nologies is a near certainty. to large quantities of high-quality data for training along with a distinct Several factors amplify the inability of owners of GPTs and enabling product or service. In the remainder of this section, however, I will technologies to capture value from downstream implementers. First, focus primarily on cases where enabling technology is owned, at least in designing a business model and/or controlling the necessary com- part, by the focal organization. plementary assets and technologies to internalize more of the spillovers A characteristic of enabling technologies and GPTs is that there are is typically beyond the scope of what a single firm can do. A case in large positive spillover effects of two kinds: static and dynamic (Carlaw point is Pilkington Glass and its invention of float glass, which was, and Lipsey, 2002). Static spillovers are standard externalities that don’t arguably, an important enabling technology. The float process, in- lead to any change in behavior by other economic agents, either at the troduced in 1959, had reduced production cost versus the dominant time or in the future. Dynamic externalities occur when the innovation plate glass process by more than a third by the time the first license was alters the current and future value of existing technologies and also granted in 1962, and by about 75% by 1974 (Teece, 2000, Appendix enables further technological opportunities for other agents. These Table B.2). It has also facilitated downstream innovations such as flat circumstances render profiting from innovation complex and difficult. panel displays. Despite Pilkington’s strong intellectual property position and the evident and quite substantial value of its technology, Pilkington licensed the float process on an exclusive territorial basis to all comers 3.1.3. Appropriability at a mere 6% of revenues. This reflects the fact that the invention was so Large (positive) spillovers signal that appropriability is especially universal in its attractiveness, and family-owned Pilkington so un- challenging. This is particularly true for enabling technologies and prepared and unable (or unwilling) to find the capital for the large-scale GPTs. Their economic contribution is very high, and the pioneers can investment required to implement the technology by itself on a typically extract only a tiny fraction of the value they create. This im- worldwide basis, that widespread licensing seemed the best alternative. plies that private enterprise will tend to underinvest in creating them, As a result, a modest royalty for Pilkington effectively passed most of absent better PFI strategies and/or government support. the benefits to manufacturers and consumers; Pilkington was able to Given that government support for the required research is gen- garner only a tiny share of the total returns for itself. erally limited or non-existent, it is important for policymakers and the Second, the future value of a GPT or enabling technology may not judiciary to be sympathetic to the appropriability challenges faced by be clear at the outset; and even if it is, there may be regulatory con- developers and owners of enabling technologies and GPTs. Despite the straints that limit value capture. The transistor, developed by AT&T’s flagging of this problem by Bresnahan and Trajtenberg (1995), there Bell Labs, was protected by patents, but AT&T licensed the technology has been little effort to understand it or develop policies to mitigate it.7 widely and almost royalty free. AT&T’s logic appeared to be that, as a 7 More recently, Thoma has noted that “studies are silent on the financial viability and 8 sustainability of the business models of the licensor of a general technology in the long George Richardson (1972) laid out a somewhat less sophisticated version of this ar- run” (Thoma, 2008, p. 110). gument 25 years earlier. 1370 D.J. Teece Research Policy 47 (2018) 1367–1387 large-scale buyer of telecom equipment, it would benefit from the metallurgy, machine tools, and more will be needed. For most in- competition among its licensees, a strategic perspective that, at novations, relevant capabilities are often already available externally. minimum, missed capturing value from the technology beyond the Outsourcing can shorten the path to successful commercialization but telecom industry (Levin, 1982). AT&T was also highly regulated, which will also increase the loss of value for the innovator. A start-up in- raised additional barriers to commercialization and value capture. The novator will almost certainly need to rely on partners, and even most spillovers were world-wide but AT&T’s value-capture footprint was incumbent firms will lack some of the relevant capabilities to derive full almost entirely domestic.9 value from the enabling technology. Third, enabling technologies, by their nature, are intermediate in- In short, capturing value from enabling technologies is more chal- puts in the value chain. The technology at issue is transferred and/or lenging, from a business model standpoint, than capturing value from a licensed to downstream firms. Designing a business model to capture modest discrete innovation. Maine and Garnsey (2006) recognize that, value more completely is far easier in consumer final product markets while what they call “generic, radical technology” (such as advanced than in such interfirm intermediate product markets. Although there is materials) create value across a broad range of industries, they none- theoretically only one monopoly rent to capture in the value chain, the theless “may face very high barriers to commercialization” (p. 375). chance of the innovator being able to design and implement an airtight They conclude that “advanced materials ventures are most likely to business model to do so is low. For example, suppose there is a patented achieve success if they develop an IP claim on a long-term, emerging technology that when placed on a communications chip in a mobile market application with major potential while focusing most of their phone will create $100 of incremental value to the implementer. The time and resources on substitution applications” (Maine and Garnsey, ability to extract a reasonable portion of this surplus from the seller of 2006, p. 392). Teece (1986) was likewise clear that the challenges were the enhanced component will be compromised if the patent owner is considerable and that if the innovator has to rely on partners to com- confined to a licensing model, especially if there is an inability to ef- mercialize the technology, the profits from innovation would need to be fectively price discriminate. The problem is amplified because, if the shared, most likely lowering the return as a result. This is quintessen- only intellectual property protection available is patents, these are not tially the case with enabling technologies and GPTs. These issues are self-enforcing. The value that can be captured by a license is affected by discussed further below, once complementarity has been addressed in the costs associated with launching a licensing program, including the greater detail. litigation costs associated with challenging patent infringers and trade secret misappropriators. Allowing the larger slice of economic value to 3.1.4. Enabling technologies in mobile communication ecosystems flow to the licensee may help launch a licensing program; but this, in The wireless communications sector affords numerous examples turn, will reduce the innovator’s share of the profits, which reduces the that illuminate the issues of value capture for many types of partici- incentives of private firms to invest in the development of enabling pants, especially the suppliers of enabling technologies. The key ad- technologies. vances behind the digital communications revolution began as pro- Fourth, the bargaining position of the innovator is undermined prietary technologies that were subsequently codified in a series of when it lacks the relevant assets and capabilities to pose a credible wireless standards, each of which provides a step change improvement threat that it will exclusively develop and commercialize the tech- in communication performance. The improvements in digital mobile nology and practice the patent(s) itself. This can be true for any tech- data across “generations” of the standard—from 2G in the early 1990s nology but is especially true for enabling technology with applications (with a top speed of 0.064 megabits per second) through the current 4G in several different industries. Hence, the broader the applicability of an (up to 200 Mbps)—have been more than incremental. Each new gen- enabling technology across industries, the less credible the threat and eration provided dramatic improvements not only in transmission the less complete the rent appropriation by the upstream innovator. speed but also in service quality, congestion management, cell hand- The original PFI framework applied to a discrete innovation that over, and signal quality, which has reduced download times for content was assumed to be commercially viable. There was little uncertainty by orders of magnitude (Fig. 1). 5G is on a path to be rolled out in 2020 specified with respect to the likely efficacy of the product or service. with yet further improvements in bandwidth, speed, and latency that Licensing was the recommended strategy if there was strong intellectual will enable the emergence of new “Internet of Things” business models property protection, although this was presented as an unlikely cir- involving massive quantities of data or mission-critical processing. cumstance. With weak appropriability, the basic idea was to attempt to Autonomous vehicles and healthcare services will be among the bene- capture value by acquiring or controlling the required complementary ficiaries (Teece, 2017a, 2017b). Wireless service is also posing a com- assets for in-house commercialization of the innovation. Otherwise, the petitive challenge to fixed wireline service, which has historically been innovator would have to either (1) build them, which is slow and ex- a cash cow for phone companies. pensive, or (2) form partnerships. Either strategy would drain some of If enabling technology fails to advance, or if new enabling tech- the value away from the innovator. nologies are not developed, then new products and applications are less Furthermore, the very nature of enabling technologies in the digital likely to be developed, resulting in a loss of potential benefits. Indeed, economy is that they are likely to be embedded in multi-invention in- the most significant technological innovation undergirding the “grand novations for which other firms also hold relevant patents, so appro- convergence” is the dramatic improvement in mobile broadband ca- priability is inherently challenging. Capturing more value requires not pacity, which has helped feed growing demand as more people have only applying the technology but also driving the technology’s path gone mobile and new, bandwidth-hungry services like live video forward into derivative applications. This inherently involves engaging streaming are developed. with partners because, as discussed in the four points above, the pursuit A key enabling technology supporting 3G was “CDMA,” a 2G of full vertical integration and/or horizontal diversification is unlikely technology that was pioneered and promoted initially by only a single to be viable given existing capabilities and resources. For example, to firm, Qualcomm. After spirited competition, engineers (from many make 3D printing (also called “additive manufacturing”) viable for in- companies) who were participating in establishing a 3G standard at dustrial manufacturing, complementary advances in design software, SDOs selected CDMA-based technology because it offered a far higher increase in bandwidth efficiency than was possible with the successor to GSM, which was the dominant 2G technology. In a similar vein, 9 The examples of float glass and the transistor reflect returns to enabling technologies “OFDMA” (a newer, non-Qualcomm technology) was selected by stan- that one way or another received private funding. It is entirely possible there are many other opportunities for enabling technology that are never developed (or significantly dard-setting bodies as one part of the 4G standard, allowing further delayed) because the private sector cannot see a business model adequate to provide increases in speed and bandwidth for mobile communications. These compensation for the required investment. technical selections involve a great deal of discussion, testing and 1371 D.J. Teece Research Policy 47 (2018) 1367–1387 small number of developers to make patented technology available to hundreds of implementers worldwide. The complexity and interdependence of the communication tech- nologies underlying the mobile data revolution have caused multiple, connected business ecosystems to emerge (Teece, 2012). Specialized firms and even vertically integrated companies are no longer islands (if they ever were) connected to others only by market transactions. Each is now part of one or more constellations of business entities and or- ganizations whose fates are tied technologically and/or competitively. Ecosystems have sprung up for a range of new services including streaming media, cloud computing, the Internet of Things, and mobile payment systems. Each of these areas has mobile communications at its core, enabling the parts of the system to work together and enhancing its availability for end users. In some cases, technologies specific to a particular function are designed into the wireless standard at an early stage. This requires that companies contemplating major new services form alliances and par- ticipate in standards development. But each generation of the wireless standard also serves as an enabling technology for services that have yet to be devised. Today the mobile industry includes (1) dedicated technology and component firms like Qualcomm; (2) implementers of standards-com- pliant infrastructure equipment like Ericsson; (3) network service pro- viders, like AT&T; (4) downstream implementers of standards-com- pliant mobile devices like Samsung; (5) software vendors like Apple and Google developing operating systems and apps; and (6) content provi- ders, from individuals to media conglomerates. Many companies span two or more of these links in the value chain. The total size of the mobile ecosystem in 2014 was, according to the Boston Consulting Group, around $3.3 trillion. Table 1 shows how this revenue breaks down across eight categories, including technology, Fig. 1. Download Time For One 12-megapixel Photo (seconds). networks, hardware, software, and services. It is more difficult, how- Notes: ever, to determine how profits are distributed among these groups, and Values calculated using midpoint download speeds from information for the T- the factors driving those returns. Mobile network, https://www.t-mobile.com/company/company-info/ The PFI framework has to be up to the task of addressing the dis- consumer/internet-services.html. (Accessed 12 March 2018). 2G is a GSM tribution of profits within complex ecosystems if it is to be a general network; 3G is a High Speed Packet Access (HSPA) network; 4G is a Long-Term framework.12 While PFI tells us that profits will go to the bottleneck Evolution (“4G LTE”) network. assets, these are not always easy to identify, and they may shift over The average size of a picture from the 12-megapixel camera of an iPhone 6S time. For example, an increase in data rates with a new generation of Plus is 2.93 MB; (Hollister, S. “The iPhone 6S camera is a huge storage hog (but it might be worth it)”, CNET, 6 October 2015, https://www.cnet.com/uk/ technology developed by upstream developers such as Qualcomm and news/iphone-6s-camera-filesizes-4k-live-photos-hdr/. (Accessed 12 March Ericsson may increase profits to providers of video content (e.g., Net- 2018). flix) and to advertising vendors while imposing greater capital and depreciation costs on network providers. And the fortunes of specific types of videos will face different appropriability regimes, such as negotiation. Each new generation of communication technology re- whether they are treated as copyright protected. quires billions of R&D dollars and over a decade of elapsed time to In the standard PFI model, the drivers are (1) the appropriability develop and formalize in standards. regime (i.e., strength of IP protection and difficulty of imitation), (2) The technological advances across the generations of these wireless complementary assets and technologies and related business model is- enabling technologies have been historically enabled by armies of en- sues, (3) standards and installed base effects, and (4) timing. For the gineers and billions of dollars in R&D investment at numerous com- mobile sector, however, one must also include, in recognition of the panies, including AT&T, IBM, TI, Motorola, Siemens, Nokia, multiple ecosystems at work, (5) ecosystem strength. As a matter of Qualcomm, and Ericsson. The major wireless technology developers theory, and assuming that there are no artificial (regulatory) barriers to today rely on licensing as their primary value capture mechanism. They entry, then the industry level (i.e., the location along the value chain, include Qualcomm, which also sells chips using its technology; Nokia, such as components, equipment, etc.) need not be critical and the above which is now almost exclusively a telecom equipment supplier; and drivers remain the most powerful. The fates of ecosystems will vary InterDigital, a pure licensing company.10 Each generation of the tech- with their nurturing of the relevant complements, the nature of the nology is included in standards developed under the auspices of the complementarities, and the success of the ecosystem leader’s (en- European Telecommunications Standards Institute (ETSI). ETSI requires trepreneurial) efforts to coordinate and strategize. contributors of technology to make licenses available on “FRAND” Within ecosystems, value captured by individual firms will depend terms.11 Thus, downstream implementers are third-party beneficiaries on firm-level dynamic capabilities, the scarcity of each firm’s resources, of umbrella licensing contracts between ETSI and the technology de- the nature of complementarities, and the business models they adopt. A velopers. ETSI has orchestrated quite remarkable cooperation among a well-functioning ecosystem will allow multiple avenues for earning 10 The primary consumer-side implementers of cellular technology, Apple and Samsung, have not historically been major technology contributors to wireless standards. 12 These issues will be addressed more fully in an article under development by 11 FRAND is a legal term that stands for “fair, reasonable, and non-discriminatory”. Jacobides et al. (2018). 1372 D.J. Teece Research Policy 47 (2018) 1367–1387 Table 1 preoccupation of literary and mathematical economists. The sim- The Value of the Mobile Ecosystem, 2014. plest things are often the most complicated to understand fully. Source: BCG (2015, p. 11). (Samuelson, 1974, p. 1255) Category 2014 Revenue The literature on complements is both confused and complex, and Core Comms.Technologies $40 billion this section will make an effort to bring incremental clarity. It should be Mobile Infrastructure Eqpmt. $220 billion noted at the outset that it is common for two or more technologies to Mobile Site Operations $230 billion produce much more when practiced together.13 The first steam trains Mobile Operators $1010 billion emerged when high-pressure steam engines were yoked to coal cars Components for Devices $260 billion Device Manufacturers $450 billion running on coal-mining hand cart rails. The lawnmower came from Device Retail $520 billion coupling a small gasoline engine to a miniature mechanical reaper. The Mobile Content and Apps $530 billion laser and the computer were capable of much together (e.g., CD Total $3260 billion players); and of still more when optical fiber was added to the mix. Economists tend to think of complementarity in terms of its effect on factor prices or on value from use (Carlaw and Lipsey, 2002). Innova- tion studies (e.g., Rosenberg and Frischtak, 1983) look instead at profits. In Google’s ecosystem, Google subsidizes the ecosystem’s technological relatedness and the impact of new combinations of ex- technology but profits handsomely from advertising linked to user isting technologies. However, the full economic significance of com- searches. In Apple’s ecosystem, Apple profits from selling attractive and plements lies not just with their ability to improve and support ap- easy-to-use devices and from taking a share of the revenue from apps propriability. Absent complementary technologies, many products and content sold by their providers. simply won’t work and won’t get developed and launched. This was the case, for example, in the U.S. electrical supply industry at the end of the 3.2. Complementary assets 19th century. The industry had a killer app—lighting—but was mired in a “war of the currents” between alternating and direct current, each of While complements were highlighted in the original PFI formula- which had certain deficiencies. It was only with the development of tion, the analysis was broad brush. For instance, time was ignored. With rotary converters that one system (alternating current) was able to the introduction of time, co-evolution becomes important. PFI re- develop a dominant position and spur rapid deployment (David, 1992). cognized “rents” (profits) as flowing to bottleneck assets, and there Next I briefly describe multiple (and complementary) types of com- could be different complementary assets that are relevant at different plementarity, summarized in Table 2: points in time. While bottleneck assets loom large in PFI as a magnet for profits, (1) Hicksian (Production) Complementarity: Factors of production rents also arise from combining different technologies in unique, value- are Hicksian complements when a decrease in the price of one enhancing ways that lead to gains exceeding the additive nature of factor leads to an increase in the quantity used of its complements standard complements, where the demand for one product (e.g., auto- in production. An innovation that reduces the cost of a factor is mobiles) can drive demand for another (e.g., gasoline). Even though the equivalent to a decline in the factor’s price. This of course pre- combination in a gasoline automobile engine involves complex tech- supposes an existing process utilizing the two factors, which makes nical interactions, it does not involve complex economic interactions. In it somewhat irrelevant for studying innovation. Yet, it generalizes many instances, technologies that were not obviously complementary, to the insight included in the original PFI framework that the suc- such as microchips and biological materials, can be combined cessful commercialization of an innovation will affect demand for (“orchestrated”) to produce entirely new products that are unique and complementary goods and services. valuable to users. This is a value-creating form of complementarity; (2) Edgeworth/Pareto (Consumer) Complementarity: Two goods, X standard notions of complements and scope economies fall far short in and Y, are Edgeworth complements in consumption if the utility of the innovation context. The refinement of the PFI framework, and a consuming them together is greater than that of consuming them in better understanding of ecosystems, begins by more closely examining isolation. As a result, the quantity demanded of either good is af- the multiple types of complementarity. fected by a change in the quantity demanded of the other. If de- mand is downward sloping, then this is the consumption-side 3.2.1. Varieties of complementarity equivalent of the production-side Hicksian case. In Edgeworth’s Complements are pervasive throughout the economic system, and broader conception, complements can include non-priced items particularly in technology development and business transformation. such as government policies, or organizational structures. Milgrom Nevertheless, they are frequently ignored. However, they are central in and Roberts (1994) formalized this for a group of activities, labeling PFI. Perhaps their neglect can be blamed on Schumpeter (1942), who them as Edgeworth complements if doing more of any subset in- stressed that “new combinations” of artifacts organized by the en- creases the returns to doing more of any subset of the remaining trepreneur brought gales of creative destruction. While emphasizing the activities. substitution of new products for old, he did not stress that, with com- (3) Hirshleifer (Asset Price) Complementarity: This is a financial plements, a rising tide can lift many boats. In the PFI framework, perspective on the Hicksian insight that can, at least in theory, be complements need to be considered with more granularity in order to used to profit from an innovation. If innovation is likely to move illuminate value capture issues, particularly the ramifications of digital prices as described above, then the initial invention creates fore- convergence. knowledge about how asset prices might move in the future. An At the heart of economic notions of complementarity is the notion, economically rational inventor with financial means and this type due to Edgeworth, that the marginal value of a variable increases with of foreknowledge could gain additional value by going long in asset another variable. As simple as it sounds, there is much complexity in markets that would be positively impacted by the invention and/or the notion of complementarity, prompting Nobel Laureate Paul Samuelson to say in 1974 that: 13 The time is ripe for a fresh, modern look at the concept of com- If a piece of technology is independent of others and offers little scope for im- plementarity … the last word has not yet been said on this ancient provement, we have what Mokyr (2002, p. 19) calls a “singleton technique” that might be discovered by chance, with little understanding of the underlying mechanism. While this was common before 1800, it is rare in today’s complex knowledge domains. 1373 D.J. Teece Research Policy 47 (2018) 1367–1387 Table 2 Types of Complementarity: Summary. Type Representative Authors Description Production Hicks (1970) A decrease in price of X leads to an increase in the quantity of Y Consumptiona Edgeworth (1897/1925) An increase in the quantity demanded of X leads to increased demand for Y Asset Price Hirshleifer (1971) Financial arbitrage opportunities are created by foreknowledge of the probable impact of an innovation. Input Oligopoly Cournot (1838/1960) Inputs X and Y will be sold for less if the companies can collude to maximize profits. Technological Teece (1986, 1988b, 2006) Unlocking some or all of the value of an innovation requires additional innovation in one or more horizontal, lateral, or vertical complements; ownership of complements aids appropriability.b Innovational Bresnahan and Trajtenberg (1995) Improvements in a GPT increase the productivity of goods in downstream applications. a Edgeworth defined complements in a way that shifts the focus from consumption to “production” or supply side issues. Activities are Edgeworth complements “if doing (more of) any one of them increases the returns to doing (more of) the others” (Milgrom and Roberts, 1995, p. 181). A set of practices may work well together (e.g., lifetime tenure, low bonus, internal promotions, and internal job retention) whereas one of the activities without the others would be less effective (or even counter-productive). This has echoes in the organizational studies field, where Nadler and Tushman (1980) developed a “congruence model” in which an organi- zation functions well when its components are acting in harmony, fitting with each other and with the business environment. In strategy, Rumelt (2011) likewise highlighted the importance of “coherent action”. b Jacobides et al. (2018) label as “unique complementarity” what I call cospecialized assets (Teece, 1986). In this situation, one item or technology doesn’t do anything useful whatsoever without another. This is also the same as Hart and Moore’s “strict complementarity,” in which two assets “are unproductive unless they are used together” (Hart and Moore, 1990, p. 1135). short in markets that would be negatively impacted. This model of occurs when the value of an innovation depends on altering the anticipated complementarity was in fact invoked in the original PFI nature of one or more existing technologies and/or on creating new article by noting that the innovator could speculate, where futures ones. This type of complementarity has nothing to do with prices markets were available, on complementary assets that were likely and quantities. It applies when the full benefit (or even any benefit) to increase in value (Teece, 1986, 2006). Hirshleifer (1971) illus- of the innovation cannot be achieved until some other, com- trates this principle with an analysis of Eli Whitney, who received a plementary technology (which, on its own, has only lower value patent for the cotton gin in 1794 yet died poor 30 years later after uses) has been created or re-engineered. The complements can be dissipating his profits in litigation over patent infringement. related vertically, horizontally, or laterally. This is in fact typical of Hirshleifer points out that Whitney could have foreseen the nega- enabling technologies. For example, realizing the full value from tive effect his invention would have on the price of cotton and the the introduction of electricity required the creation of electric positive effect on the price of land and then speculated accordingly motors that could be attached to machines. The introduction of a on either or both. new generation of cellular network requires new microchips and (4) Cournot (Input Oligopoly) Complementarity: Two products are handsets to exploit it. Technological complementarity can pose a Cournot complements if they are used together but sold by separate PFI challenge for an innovator if the complement is created by a companies with monopoly power.14 Consider, for example, the case separate company and becomes a bottleneck asset. One solution is where each firm has a monopoly over one of two inputs used to to create the complement in-house, but the innovator may lack the make a product. If the companies are unable to collude, they will capabilities to do so. charge prices that maximize their individual profits considered in (6) Innovational Complementarity: Improvements in a general-pur- isolation yet fail to maximize their collective profit. This problem pose technology will increase the productivity in downstream sec- arises in PFI when there are two or more bottleneck assets needed tors. For instance, the improvement of a cellular network opens new to produce an innovative product, and they are owned by separate innovation opportunities for firms providing wireless data devices. parties. Whereas common ownership would lead to charging a This may be a special case of (5) above. monopoly price, the separate firms, absent collusion, might charge higher prices that lower their total profit. This is an interesting While complements were highlighted in the original PFI formula- theoretical puzzle but not likely to be of practical significance be- tion, the analysis was quite general, and time was ignored. With the cause markets are seldom in equilibrium long enough to determine introduction of time, co-evolution becomes important. There could also the “right” price. A real-world example would be Intel and Micro- be different complementary assets that become salient at different soft in the PC industry; their dominance of intermediate goods did points in time. Rents are generated not only from exploiting bottleneck not harm consumers because competition in the downstream assets; they also arise from combining different technologies in unique, market drove down end-user prices. The Cournot model has been value-enhancing ways that lead to system-wide gains exceeding the misapplied in the context of patents where economic theorists have additive nature of standard Edgeworth complements. With Edgeworth postulated that owners of standard essential patents will pursue complements, the demand for one product (e.g., automobiles) can drive self-interested pricing strategies that result in high cumulative demand for another (e.g., gasoline). But Edgeworth complements and royalties (Galetovic and Gupta, 2016) that are destructive of the Cournot complements are largely irrelevant to the innovation context. market. In reality, we tend to observe the opposite, in part because patents are not self-enforcing and in part because licensing as a business model has historically been weak in its ability to capture 3.2.2. PFI and complementarity reconsidered anything but a tiny fraction of the social benefits from enabling The lines between the various forms of complementarity are in some technology. FRAND licensing regimes have also limited the bar- cases blurred, since two goods can fit several of these categories at the gaining power of patent holders still further. same time. PFI (implicitly) embraced all of them except the Cournot (5) Technological Complementarity: Technological complementarity (oligopoly version). What all these types of complementarity have in common is that their presence raises complex coordination issues, which in turn renders 14 As pointed out to the author by Jacobides, Cournot complements are a special case appropriability more difficult. As noted in the context of GPT and en- of unique complementarities (where input A is needed in fixed proportion to input B) on abling technologies, an upstream innovator has no guarantee that the top of which separate monopoly supply of each input is assumed. downstream users of the technology (and the providers of 1374 D.J. Teece Research Policy 47 (2018) 1367–1387 complements) will make the investments needed to generate the largest scope and complexity of the economic applications of a significant value, and vice-versa. Pessimistic expectations can become self-ful- enabling technology. When the application opportunities are numerous filling, which reduces the innovator’s incentives for investing in future and diverse, joint activity places heavy demands on the management research and development. resources of the innovator. Coupled with contractual issues, this leads These points have been given only cursory treatment in the eco- the innovator to fall back on the licensing option. Licensing, however, is nomics literature. Malmgren (1961) and Richardson (1972) had noted often inadequate, especially in instances where courts are reluctant (or the issue in general, but other scholars have largely ignored it, perhaps unable) to issue injunctions. These issues are explored in more detail believing that expectations with respect to complementary investments below. (vertical & lateral) somehow converge naturally. In the original PFI paper’s discussion of business model issues, In most cases, however, it must be recognized that technological emphasis was given to questions around value chain issues, namely, and innovational complementarities impose severe coordination, whether the firm was ready and able to assemble the specific and market design, and control challenges that impair the alignment of generic assets needed to bring the innovation to market. Since then, I activities across firms in a market economy.15 After such issues were and others have expanded the notion of business models to encompass raised by Malmgren and Richardson, they were echoed by Williamson the architecture of the business, including customers, costs, integration, (1975), remarked upon by Teece (1984), explored empirically to a and likely competitor responses (Chesbrough and Rosenbloom, 2002; limited degree by Armour and Teece (1980) and Helfat and Teece Teece, 2010). Moreover, business model design is an iterative process. (1987), emphasized (as discussed above) in the vertical GPT context by This point is particularly relevant as more and more businesses are built Bresnahan and Trajtenberg (1995) and Helpman (1998), but never fully around software. Software-based business models are relatively easy to explored or developed by economists or management scholars. The modify in response to customer feedback, changes in the user base, or economics literature tends to assume that, in the main, upstream and other evidence of missed opportunities. downstream investment expectations will converge, which seems un- Subsequent work on PFI has emphasized technological as well as likely given the proprietary (and hence secret) nature of much of the value chain complementarity (Somaya and Teece, 2006; Teece, 2006; required innovation activity.16 Bresnahan (2012) is amongst the few Somaya et al., 2011). Whereas Hicksian and Edgeworth com- studies pointing out potential contractual and market failure issues. plementarities present the innovator with challenges and opportunities Jones notes that that can be mitigated or exploited by thinking ahead strategically, technological and innovational complementarities place a burden on [w]ith Bresnahan’s starting point, however, the nature of knowledge the innovator to coordinate with all owners of relevant intellectual distribution is such that one does not even know who to integrate property and with downstream implementers. In the original PFI fra- with. That is the key problem: the fact that you cannot identify the mework, the implied advice to the innovator was to own (or control) recombinant possibilities ex ante means that you cannot easily solve any bottleneck asset(s). In a multi-invention context, the bottleneck the bargaining problem in practice—you cannot integrate your way could be a technology rather than a conventional asset. The innovator around it. So innovation faces a serious market failure in the sense can acquire, license, or ally with the owner of relevant technologies in that socially profitable innovation does not occur. order to ensure a predictable path to follow both for the initial com- (Jones, 2012, p. 660) mercialization of the innovation and for its future development In short, there appears to be no market mechanism (and perhaps not (Chesbrough and Teece, 1996). even vertical