New Directions in Research on Dominant Design, Technological Innovations, and Industrial Change

Transcription

1 New Directions in Research on Dominant Design, Technological Innovations, and Industrial Change Johann Peter Murmann Kellogg Graduate School of Management Northwestern University Evanston, IL Phone: Fax: Version 2.0 1

2 I. Introduction Organization theorists, strategy scholars, economists, and historians of technology have all highlighted the powerful role of technology in shaping industrial dynamics and firm performance. Mastering the "black box of technology" represents a crucial organizational capability for succeeding in competitive markets. It is by now a well established proposition that technological change is one of the prime triggers of organizational change (Nelson and Winter, 1982; Pavitt, 1984; Tushman and Anderson, 1986; Metcalfe and Gibbons, 1989; Henderson and Clark, 1990; Freeman and Soete, 1997). Schumpeter (1950) long ago highlighted that technological change is a doubleedged sword. One the one hand, technological change creates vast opportunities for firms to improve existing products or to design entirely new products. On the other hand, technological change can destroy the usefulness of existing products and thereby threaten the livelihood of individuals and firms tied to the old technology. Scholars following the inspiration of Schumpeter have tried to understand in greater detail how this process of "creative destruction" inherent in technical change shapes the fate of firms, populations of firms, and entire nations (Rosenbloom and Christensen, 1994; Hannan and Freeman, 1989; Porter, 1990). Time and again studies tracing the evolution of technologies over long periods of time have shown that technical change is a highly unpredictable process (Rosenberg, 1982). Notwithstanding, scholars working in different academic disciplines have documented characteristic patterns of innovations and formulated illuminating models of technical evolution. All models essentially have in common that they attempt to couple characteristic patterns of technical evolution with changing levels of uncertainties, exploiting the fact that the degree of uncertainty inherent in the process is not uniform over time. In organization theory, the technology cycle model and its concept of a dominant design has received a great deal of attention and has stimulated important empirical research over the last decade. As the concept of a dominant design has taken the center stage in much empirical work linking technological and organizational change (Anderson and Tushman, 1990; Utterback and Suárez, 1993; Utterback, 1994; Suárez and Utterback, 1995), a number conceptual puzzles and difficult empirical issues have surfaced in the literature. While in some writings the term "dominant design" is applied to a total technological system (Abernathy, 1978; Van de Ven and Garud, 1993; Suárez and Utterback, 1995; Iansiti and Khanna, 1995), in other writings the term is applied to 2

3 components of systems (Rosenkopf and Tushman, 1994; Khazam and Mowery, 1994; Miller et al., 1995). Some authors even apply the term "dominant design" in the same paper to total systems and components of systems without explaining whether the concept applies to a total system and its components in the same way (Anderson & Tushman, 1990; Utterback and Suarez, 1993). A second difficulty in interpreting the body of research findings on dominant designs arises as authors frequently appear to investigate dominant designs at the level of a total system but then describe a dominant design in terms of the characteristics located at a particular component (Anderson and Tushman, 1990; Utterback and Suarez, 1993; Baum, Korn and Kotha, 1995). Trying to compare research findings across studies is also made difficult because various writers differ in regard to the level of detail at which they define dominant designs. Some studies define dominant designs in terms of general technological principles (see for example, Miller et al., 1995; Rosenkopf and Tushman, 1994). Other studies, in contrast, define dominant designs in terms of specific product names (Abernathy and Utterback, 1978; Teece, 1986; Anderson and Tushman, 1990; Utterback and Suarez, 1993). 1 There also seems to be disagreement about how often a dominant design can emerge during the evolution of a product class. In their analysis of the evolution of six products classes, Suarez and Utterback identify only one dominant design for the entire life-span of each product class. Anderson and Tushman (1990), on the other hand, find a periodic emergence of new dominant designs during the life-span of four other product classes. Finally, writers on dominant design disagree about the range of products to which the dominant design theory applies. Authors like Anderson and Tushman (1990) suggest that the theory applies to all technologies that evolve without interference of patent rights. Other writers take a much narrower view and see the theory only applicable to assembled products (Abernathy and Utterback, 1978; Teece, 1986; Nelson, 1995; Suarez and Utterback, 1995). Many other authors are simply silent on the boundary conditions of the theory (Van de Ven and Garud, 1993; Sanderson and Uzumeri, 1995; Baum, Korn and Kotha, 1995). Academic researchers, R&D managers, and public policy makers who want to use the dominant design theory currently face a great variety of definitions and analytical approaches that make it difficult to build new research efforts on the existing literature. In the absence of some uniform definitions of the major concepts in theory, empirical 1 The Productivity Dilemma (1978) has only Abernathy on the cover page. However, the chapter that sets forth the theoretical model of dominant designs was coauthored by Abernathy and Utterback. Similarly, Utterback helped to clarify the model in the implications chapter. This is why we cite The Productivity Dilemma when we refer to the Abernathy-Utterback model. 3

4 studies are unlikely to lead to some integrated body of findings. The purpose of this paper is the help remedy this situation. Our strategy is twofold: We first critically review the literature on standardization not only in organization theory but also in neighboring disciplines to show the importance of dominant designs in writings on technical evolution across many fields. From this broad review we synthesize a conceptual framework that can help integrate the existing literature on dominant designs in organization theory. The goal of this essay is to make some progress in removing potential confusion and to solve some of the analytical puzzles that make it difficult for academic researchers to design more cumulative studies. By studying the literature around the dominant design concept, we also want to provide R&D managers and public policy makers with an example of how to evaluate the adequacy of their tool box of concepts for analyzing technological dynamics. The essay proposes that technological evolution proceeds in the form of a nested hierarchy of technology cycles. We argue that a hierarchical model of technical evolution gains tremendous analytical power when linkages between subsystems are viewed as subsystems in their own right. The model clarifies why innovations leading to order-ofmagnitude changes in performance can originate at any level of the physical hierarchy, and helps remove the confusions in the existing literature. II. An Illustration of the Phenomenon of Technical Change To appreciate the phenomenon of dominant designs, consider the evolutionary history of passenger airplanes. The dream of flying is as old as humanity itself but the first successful flying machine was constructed only in 1903 by the Wright brothers. Human flight was finally made possible by two developments: Abandoning the idea of imitating birds in a flying machine, designers switched to a principle of operation where a stationary wing would provide upward thrust and a propeller would provide forward thrust (Vincenti, 1990). When engineers were also able to make the internal combustion engine, originally developed for automobiles, light enough for an airplane, the Wright brothers were at last in the position to keep their flying machine in the air for more then 10 seconds and travel 120 feet. Two years later the Wright brothers designed an improved airplane that carried one 1 person (the pilot) for 24 miles at a speed of 35 miles per hour. Over the last ninety years airplanes advanced dramatically along the three crucial performance dimensions: speed, operating cost per passenger, and flying safety. Today s largest commercial airplane, the Boeing 747 SP 400, can cruise at over 600 miles per hour, carry 455 passengers over more than 7000 miles, and has a failure rate that 4

5 is lower than any other transportation technology. Tracing new features from one design to the next reveals that aeronautical engineers continuously introduced design changes ranging from improved nuts and bolts to a very different shape and organization of the entire artifact. Sometimes these changes were radical (both in terms of the technological principles used and the additional performance levels achieved) but more often they were incremental in nature. For analytical purposes it is useful to view an airplane as the union of a number of essential functions: a propulsion, lifting, landing, and control subsystem, a passenger compartment, and finally a mechanism for linking all these different functional domains to produce a well-designed overall flying system. During the infancy stage of the airplane, designers experimented with configuring the major components of an airplane in different ways. The purpose of changing the arrangement of fuselage, wings, elevating and steering rudder, engine, propeller, and landing gear was to bring about stable, controllable and efficient flight. The first successful airplane of the Wright brothers had two wings and was powered by a twelve horsepower 4 cylinder internal combustion fuel engine driving two propellers. With the exception of the engine, the airplane was made out of wood and fabric. The evolution that transformed this early design into a contemporary jet airliner proceeded through cycles of variation, selection and retention at all levels of the physical artifact. Here is a selection of innovation episodes that reveal the emergence of dominant designs at the system, subsystem, and basic component levels: Early variation of systems architecture: At the highest level of the physical artifact, the architecture of the overall body, aeronautical engineers experimented with a great many design alternatives before a particular configuration emerged as the dominant design for a period of time. Trying to make airplanes more controllable (so that pilots could fly curves and travel over longer distances), the Wright brothers and other designers experimented with changing the size of the various main components and placing them in different relations to one another. During this process, designers not only built monoplanes with either low or high-mounted wings but also created double and triple wing airplanes. In some designs, the propeller was placed in front of the wings facing forward, in others it was placed behind the wings facing backwards. To achieve more stability and greater distance some designers built airplanes with two propellers powered by independent engines; others tried to perfect the airplane configuration with single engine motor power. After experimenting and learning about the advantages and disadvantages of various configurations, the engine-forward, tail-aft biplane by WW I had become the dominant design, which designers typically took as the starting point in their efforts to create better airplane designs (Vincenti, 1990). As engines became more powerful, designers switched from the biplane to the single wing configuration. Figure 1 shows how, after the emergence of the engine-forward, tail-aft monoplane as the dominant design for the overall configuration, engineers focused on experimenting with 5

6 small variations of the dominant design to find the most aerodynamic airplane shape as well as on improving the individual functional domains. Evolution of propulsion function: The propulsion subsystem of the Wright brothers airplane consisted of an internal combustion engine and two wood propellers. To improve the bite of propellers, a large variety of shapes shown in figure 2 were tested. The historical record does not indicate whether a particular shape became the dominant design. However, metal replaced wood as the dominant material for the propeller blades by the 1930s. Trying to scale up propulsive power, engineers examined the effects of mounting up to eight engines on the airplane. Around , the three engine approach as embodied in the Ford Tri-Motor became the dominant design for commercial airplanes until the appearance of the DC-3 ushered in a period of two motor designs. A crucial innovation in raising the performance of individual combustion engines occurred at a low level component. After testing a number of different materials, engineers converged on the sodium-cooled exhaust valves as the dominant design for this component of the motor cylinders in the early 1930s. A dramatic increase of engine performance was also achieved by adding lead to the engine fuel. The outcome of much experimentation with different lead levels was a 90 octane standard that remained in existence until 1945 when it was replaced by a 100 octane standard (Hanieski, 1973). When theoretical work in aeronautics in the 1920s predicted that it was possible to travel at least twice as fast as previously assumed, designers looked for a propulsion technology that would not have the speed limitations of propellers. The German design community experimented with a number of different propulsion concepts: rockets, controlled bomb explosion, pure jets and turbine jets (Jewkes, Sawers and Stillerman, 1961; Constant, 1980). From these variations, the turbine jet (turbojet) engine emerged as the most viable option. Initially, turbojets had many fewer parts than traditional piston engines, but over the course of fifty years jet designers have added so many parts that jet engines have again become very complex subsystems. Within turbojet technology a large number of alternative architectures were tried out until the ducted fan type axial flow turbojet became the dominant design, largely because of its fuel efficiency (Constant, 1980). For the turbojet to become a viable technology, material scientists had to mix hundreds of different alloys to find an alloy that could withstand the enormous heat of a jet engine. Engineers found the Nimonic 80 alloy to be the most effective heat resistant material for constructing a gas turbine and it became the dominant material. A similar process of search, development and experimentation with a large variety of alloys led to the selection of alloy G. 18B for the rotor and rim of the turbine (Hanieski, 1973). Without solving these "material" bottlenecks, the turbojet would not have replaced the piston-engine. To successfully incorporate jet-engines into airplane technology, it was necessary to make a number of complementary changes in other subsystems. For example, airframes had to be made much stronger in order withstand the higher levels of stress created by jet-engines. Furthermore, when jet engines finally became 6

7 commercially viable in the mid 1950s after a twenty year development period, the runways of airports had to be extended because jets need to accelerate longer before taking off. The introduction of jet engines could not be accomplished in a modular way by simply mounting them in the space allocated for the traditional piston engines. Before jet-engines could become the dominant design for commercial airplanes, systemic innovation both in the airplane as a whole and its larger technological context (runways) had to be accomplished. Evolution of lifting function: The shape of the wings has important aerodynamics consequences for the performance of the airplane as a whole. Ideally wings create large lifting forces without causing much drag that slows down the airplane. Figures 3 and 4 give evidence of the many different wing shapes. It is estimated that in pursuing a formula for optimal wing shapes designers had tested over 2000 different airfoils by 1936 (Vincenti, 1990). These 2000 different designs embody modification of the shape of the cross section, the thickness, and the curvature of the top and bottom side of the wing. Because the wing shape best suited for a particular airplane depends on its size and a number of other parameters, wing shapes are typically custom made for every airplane model and thus no dominant shape design emerged. However, a number of standard design choices emerged with regard to other parameters of wing design. The internal architecture of wings also underwent a great deal of variation. In the early 1930s, however, most wing designers adopted Wagner s invention of a latticework frame which was very efficient in shifting stress to the covering sheet. Partly because of this advantage feature, metal wings became the dominant design. (For alternative lattice designs, see Figure 5.) To increase the lifting, steering, and breaking functions of wings, today s commercial jet airliners include variable sweep back wings with multiple slots, replacing the earlier fixed wing dominant design. Figure 6 shows a number of alternative slot designs that were tried before the multiple slot approach became the dominant design. Evolution of fuselage: Every component used to assemble the overall airplane is made from a particular material. Structural materials used for the construction of the fuselage have two chief performance requirements: to be as light as possible, yet very resistant to stresses. Partly due to cultural values associated at the time with metals, engineers started to use metal materials in more and more airplane parts instead of wood and fabric materials (Schatzberg, 1994). By 1919, the first allmetal airplane (the Junkers F 13) was designed. This marked the beginning of a trajectory that would eventually displace wood entirely from fuselage (and wing) construction. At the level of basic materials, then, metal replaced wood as the dominant material for structural parts. Recently engineers have challenged the dominance of pure metal alloys by starting to employ fiber reinforced composite materials for airplane structures (Schatzberg, 1994). 7

8 Evolution of landing function: The pioneer airplanes, for example the 1910 Nieport model, typically were equipped with a four-wheel fixed gear, resembling a little cart. Efforts to make landing gears more robust led to the tripoid design with two big wheels mounted below the fuselage and a very small wheel at the bottom of the tail, giving the entire fuselage a downward slope toward the rear. The tripoid configuration became the dominant design for commercial airplanes until the tri-cycle undercarriage was introduced in 1938 by Douglas s model 4E (Miller and Sawers, 1968). By introducing a third leg of equal length, an airplane would be less inclined to flip onto its face (i.e. the front end of the fuselage) during landing. Landing gears that put the commercial airplane in a fully horizontal position became the dominant design until today. In the 1930s engineers started to explore a number of different design ideas for making the landing gear more aerodynamic (See Figure 7). These attempts can be classified into two broad design approaches, the enclosing of wheels and the construction of retractable landing gears (Vincenti, 1994). Trying out a number of different methods of enclosure, designers put airplanes into service that either had their wheels enclosed (a design called wheel pants or spats) or had the entire landing gear enclosed (a design called trouser pants). Similarly, a number of different retraction mechanisms were devised. Retractable landing gears promised to deliver the greatest aerodynamic gains, but these devices were much more complicated than wheel and trouser pants, leading designers initially to focus on enclosing the landing gear. In the end, however, it was the laterally retracting landing gear that became the dominant design for all commercial airplanes going faster than 250 mph, winning the design competition not only against wheel and trouser pants but also against mechanisms where wheels would retract backwards or into the sides of the fuselages. Evolution of linking function: All components at some level have to be joined to compose an integrated artifact. As the case of rivets that join metal sheets to the lattice of the airplane demonstrate, dominant designs emerge at the most trivial linking mechanisms. To make additional aerodynamic gains after streamlining the fuselage and the wings, engineers started to explore different methods for making rivets even with the skin of the fuselage and the wings. This process is called flush riveting. In the early phase of flush riveting designers used a wide variety of had angles, ranging from 78 to 130 degrees. When in December 1941 the aeronautical board of the army and navy made the 100 degree angle head mandatory for all military aircraft, it quickly spread throughout the entire industry. Thus the 100 degree angle head became the dominant design, eliminating other options from the airplane construction practice (Vincenti, 1990). Evolution of control subsystem: 8

9 While early airplanes were only controlled by the steering skills of the pilot, today s large commercial planes can in principle be flown by a collection of instruments without a human pilot. To provide maximum safety, human pilots are still sitting in the cockpit. Automatic flight was made possible because engineers over the last 90 years developed a large number of control instruments from the gyrostabilizer that relieved the pilot from constant steering action to automatic navigation instruments. The strings and pulleys that provided the mechanical connection between the pilot s wheel and the rudders were gradually transformed into power operated controls, first introduced in the Douglas DC 4E airplane. In a fly-by-wire control system, now guiding such airplanes as the Airbus 320 and the Boeing 777, the pilot no longer has a direct tactile connection to the pressures that impinge upon the various rudders and flaps of the plane. These fly-by-wire systems are expected to become the dominant design in the future. Many standards in the design of the control instruments emerged, partly because the airline industry is so heavily regulated. While analog displays were the dominant design in the early days of instrumentation, digital displays have in recent years replaced analog displays as the dominant design, as the cockpit is becoming ever more computerized. Innovation and industrial dynamics: Besides the dramatic fluctuations in demand due to the two world wars and the Great Depression, it was the innovations in the design of commercial airplanes that had powerful effects on industry dynamics from the very beginning of the technology (Rae, 1968). During the airframe revolution between 1925 and 1935 the introduction of such important innovations as the all-metal, lowwing monoplane, the controllable-pitch propeller, the retractable landing gear, and wing flaps led to significant entry of new firms, exit of incumbents, mergers and dramatic reconfigurations of relative market shares. The former leaders in the airframe industry like the Curtiss-Wright corporation were overtaken by firms like Boeing, Douglas, Lockheed, and Martin which pioneered these important innovations. When in 1936 Douglas integrated this set of innovations into its DC- 3, the firm achieved so economical an airplane that it very quickly became the largest manufacturer of commercial airplanes in the world until the jet era in the 1950s. As other firms tried to imitate Douglas s design formula, the all-metal, low-wing monoplane, the controllable-pitch propeller, the retractable landing gear, and wing became standard design features for the next 20 years, largely because the DC-3 dominated the commercial market. By 1941, almost 8 out of 10 airliners were DC-3s (Klein, 1977). When jet engines became commercially viable in the mid-1950s, Boeing was quicker, however, to respond to the technological discontinuity in the engine subsystem of airplane. Boeing tested a prototype, its 707, a full year before Douglas even began developing its DC-8 jet airliner. Boeing captured a leading position in the beginning of the jet era and has succeeded in remaining the largest 9

10 producer of commercial airplanes to the present day, while Douglas was reduced to very small player in the market only to be taken over by Boeing recently, and Lockheed altogether abandoned the commercial jet market. Radical innovations in individual subsystems have also led to a large amount of entry and exit among the populations of firms associated with the production of individual components and subsystems (Rae, 1968). For instance, the leading manufacturers of water-cooled aircraft engines in the early 1920s Curtiss Aeroplane and Motor Corporation, the Wright Aeronautical Corporation, and the Packard Motor Car company were challenged by dynamic competitors like Lawrence and Pratt & Whitney who entered the industry to pioneer the development of air-cooled engines. While Wright was able make the transition to air-cooled engines and remain a major producer in the 1930s, the other leading firms were overtaken by Pratt & Whitney, and many exited the industry (Klein, 1977). Although this case study of innovations only covers a small number of the innovations that propelled airplane technology to current performance levels, it gives a good overview of why scholars of innovation have found it useful to conceptualize technological evolution as a process of variation, selection, and retention. This short case history also suggests that dominant designs can occur at all levels of the physical artifact, from the small component to the total system, or to put it in other words, at different levels of resolution. We will now review more systematically the evidence on dominant designs. After conducting a critical review of the literature on dominant designs in organization theory proper, we will examine the evidence on dominant designs uncovered by scholars outside the field of organization theory and canvass these literatures for ideas that will help us to formulate a refined model of dominant designs. 10

11 III. Literature Review 1. Survey of Writings on Dominant Designs in Organization Theory and Strategy. Since Abernathy and Utterback (1978) first developed the concept of a dominant design from a study of the automobile industry, many writers in the field of organization theory and strategy have found the concept to be extremely useful tool for studying the evolution of technological products. At the heart of dominant design thinking lies the empirical observation that technology evolves by trial and error and thus entails enormous risks for the population of firms engaged in its development. When a new product class appears, it is very unclear what kind of inherent potential a technology possesses and what kind of needs users have. The only way to reduce the uncertainty about technological potential and user needs is to create different designs and wait for feedback from users. Over time, merely one or a few designs from the large number of design trials eventually succeed. The firms that happen to be the producers of the winning designs will flourish while the firms that invested in the failing designs will incur great economic losses and more often than not go out of business. The dynamics that lead to dominant designs are of central importance to firms that have a stake in the way technology evolves because the emergence of a dominant designs produces winners and losers. As organization theorists and strategy scholars have developed a greater concern about the role of technology in shaping the fate of firms, dominant design thinking has become a major intellectual focus of both theoretical and empirical work. Following the lead of Abernathy and Utterback, scholars have applied dominant design ideas to a wide variety of products. Dominant designs have been found in such diverse industries as cement production machinery, flat and container glass production systems, minicomputers (Anderson and Tushman, 1990), video recorders (Rosenbloom and Cusumano, 1987), typewriters, TV sets, TV tubes, transistors, electronic calculators (Utterback and Suarez, 1993), radio transmitters (Rosenkopf and Tushman, 1994), hearing aids (Van de Ven and Garud, 1993), computer work stations (Khazam and Mowery, 1994), disc drives (Rosenbloom and Christensen, 1994), facsimile transmission 11

12 devices (Baum, Korn and Kotha, 1995), mainframe computers (Iansiti & Khana, 1995), personal stereos (Sanderson and Uzumeri, 1995), flight simulators (Miller, Hobday, Leroux-Demers, Olleros, 1995) and microprocessors (Wade, 1995). (See Table 1 for an overview of the different studies, their units of analysis and key findings, etc.) However, researchers have also found that not all technological discontinuities lead to dominant designs, suggesting that the emergence of a dominant design may not be a universal phenomenon. Anderson and Tushman (1990) revealed that dominant designs do emerge after most but not after every single discontinuity in a given product class. Utterback and Suarez (1993) failed to uncover evidence that a dominant design appeared in the case of integrated circuits and supercomputers. Although Henderson and Clark (1990) devote considerable attention to dominant design concepts in the theoretical section of their paper on the failure of established firms in the photolithographic aligner industry, they never state whether or not they believe that a dominant design emerged in this particular technology. This makes the photolithographic aligner industry a case that can neither be interpreted as evidence for nor against dominant design theory. The weight of the existing evidence presented by the aforementioned authors suggests that the process of standardization leading to a dominant design takes place in a great variety of industries. But if the present evidence is correct, there are clearly instances where dominant designs do not emerge. To refine dominant design theory, it would undoubtedly be a great step forward to uncover the factors that explain under what conditions dominant designs emerge. Our review of the existing empirical studies, however, suggests that it is currently impossible to isolate from the available evidence with any degree of certainty a few factors that can predict under what circumstances dominant designs emerge. A close reading of the existing theoretical and empirical literature reveals that researchers have worked with a great variety of definitions and empirical methodologies to determine dominant designs. The absence of a shared set of definitions makes it very difficult to directly compare the results reported in the various industry studies. Before it is possible to isolate a small set of factors that predict the emergence of dominant designs, it will be necessary to achieve some convergence in the way researchers use dominant design concepts and collect evidence. As a first step towards 12

13 facilitating such a convergence, we review how and why researchers differ in the way they use dominant design concepts. Our goal is to bring into view the sources of disagreement and thereby enable an informed debate on how these differences can be overcome without forcing researchers into a straightjacket that makes it impossible to emphasize certain aspects of dominant designs over others. We will organize our discussion in terms of the most important dimensions along which researchers differ in their work on dominant designs. They concern 1) the definition of a dominant design, 2) the chosen unit and level of analysis, 3) the underlying causal mechanisms, and 4) the boundary conditions of the theory. These dimensions are, of course, not entirely independent from one another since the general conception of dominant designs will have direct implications on how a particular researcher will think about how to take various analytical steps required in theorizing and conducting empirical research on dominant designs. Before discussing the differences among authors, let us get clear about areas where researchers generally are in broad agreement. All researchers on dominant designs share the view that technology is an important factor in shaping the evolution of industries and the performance of firms that are affected by a particular technology. There is no controversy about the proposition that a powerful process of standardization (i. e. a reduction in design variety) accompanies the development of a technology. Broad agreement also exists about the notion that the emergence of dominant designs often represents a defining event that marks the transition from one competitive regime to another one. All scholars hold that the birth of a new product class is marked by a great degree of uncertainty as to what a technology can do and what kind of performance characteristics would be most beneficial to users. Furthermore, scholars who have studied the actual course of product evolution all have come to appreciate the central fact that the evolutionary path is replete with designs that have failed in the marketplace. Although scholars start from this common ground, they arrive at very different understandings of what dominant designs are and how one goes about researching when they occur. 13

14 a. Disagreements about Definitions Important differences already begin with the ways in which scholars have defined the idea of a dominant design. Not all of them do provide an explicit definition of dominant designs in their writings. However, as one analyzes how scholars have used the concept of a dominant design, it becomes apparent that even when scholars do not commit themselves to an explicit definition, they often attach a different meaning to the concept. It would take us too far to review all the different meanings that have been associated with this concept. We will limit our discussion to those authors who have engaged in efforts to develop dominant design theory itself, leaving out those who have merely applied existing dominant design ideas to a particular research concern without trying to make a contribution to the development of dominant design theory itself. Since Abernathy and Utterback pioneered the concept of a dominant design, we begin with their definition to provide a convenient reference point for evaluating later definitions. They identify several dimensions that characterize a dominant design. Writing a book on the dynamic relationship between product and process innovation, these authors see a dominant design as the turning point that leads the industry to move from a made-to-order to a standardized-product manufacturing system. According to Abernathy and Utterback, this transition from flexible to specialized production processes is marked by a series of steps. The first one is the development of a model that has sufficiently broad appeal in contrast to the design of earlier product variants that focused on performance dimensions valued only by a small number of users. This design that can satisfy the needs of a broad class of users is not a radical innovation but rather a creative synthesis of innovations that were introduced independently in earlier products. The second and decisive step is the achievement of a dominant product design, one that attracts significant market share and forces imitative competition design reaction (p.147). In the third step, competitors are forced to imitate this broadly appealing design, inducing product standardization throughout the industry. The emergence of a dominant design closes a period where firms compete by introducing radically different product designs into the market. After the dominant design is in place, innovations focus on incrementally 14

15 changing the product from year to year, they become more cumulative, and competition centers more on price than on product differences. It is rather unfortunate that Abernathy and Utterback use the term dominant design already for the second step where one design gains significant market share without having necessarily reached anywhere close to 50% of the market. Abernathy clearly stipulates that a dominant design is one that diffuses almost completely through the industry (p.61-2). Diffusion throughout the industry is precisely what makes it the dominant design. Employing the term for the period before and after introduces considerable ambiguities. To summarize Abernathy s and Utterback s definition, a dominant design meets the needs of most users, diffuses almost completely throughout the industry, is a synthesis of previous independently introduced innovations, and ushers in a period of incremental innovations exploiting latent performance potentials. Abernathy and Utterback illustrate their concept of a dominant design by referring to the Ford Model T and the DC-3 as powerful examples of dominant designs that shaped the automobile and the airplane industries. The manner in which Abernathy and Utterback describe the emergence of a dominant design conveys the strong notion that the design that will eventually emerge as the dominant one is pre-ordained to achieve this status. The only factor that can prevent this process from taking place are overly segmented product markets. These scholars appear to suggest that it is simply a matter of time until designers will have tried enough variants of a product to achieve a synthesis that is slated to dominate all others. While this design may not be the best along all performance dimensions, Abernathy and Utterback seem to regard it as the best compromise that will then force all other industry participants to imitate the design and make it the standard for the entire industry. In his recent writings with Suarez (Utterback and Suarez, 1993), Utterback continues to underscore the notion that a dominant design is successful because it brings together the most useful features that were previously scattered across different designs. Utterback and Suarez s (1995) insistence that some authors have used measures which are tautological to determine a dominant design, such as market share (Anderson and Tushman, 1990) (p.426), indicates that these authors use the term dominant design not so much to refer to the design that has achieved at least 50% of the market, but rather to 15

16 identify the design that offers a collection of design features that will make it the irresistible choice in the market as soon as it appears. Utterback and Suarez s proposal that [a] better measure might be notable increase in licensing activity during several years by a given firm or by a group of firms with products based on the same core technology (p.426) gives even stronger evidence that for these scholars the defining for dominant design occurs well before the design captures over 50% of the market share. The interpretation that Utterback and Suarez use the term dominant design in the sense of the best one in the market the one that can dominate all others once it has been found as opposed to the most frequent one receives further support from the way in which Utterback and Suarez react to scholars who have identified economies of scales as an important force in bringing about dominant designs. Utterback and Suarez write that we think that economies of scale are of primary importance after a dominant design is in place (p.418). In their account, economies of scale are not a mechanism that helps bring about a dominant design, but rather the emergence of the best compromise makes it possible in the first place to sell a standardized product to many different users. While most of Utterback and Suarez s statements seem to suggest that a dominant design is dominant precisely because it is the best compromise, these authors are sometimes pulled away from such a position. This vacillation in perspective makes it rather difficult to give their account a straightforward reading. Utterback and Suarez s assertion that a dominant design may not be ideal choice in a broader context of optimality, but rather a design, such as the familiar QWERTY typewriter keyboard, that becomes an effective standard is difficult to reconcile with their notion of a dominant design as the best technological compromise. When the authors continue that dominant design is not necessarily the one which embodies the most extreme technical performance (p.418), it becomes even more difficult to see how they can maintain a position that identifies a dominant design as the best compromise. But if a dominant design is not dominant because it captures over 50% of the market share (we read Abernathy, unlike Utterback, as making this an defining requirement for a dominant design) and if now Utterback and Suarez retreat from the idea that a dominant design is the best technological compromise, we are left with the tantalizing question: What then is a dominant design? 16

17 Before moving on to the definitions of other scholars, we want to record for our later discussion of differences in unit of analysis that Utterback and Suarez elaborate their position by explaining that for complex products with many parts a dominant design embodies a collection of related standards (p.418). Utterback and Suarez try to clarify here the relationship between the concept of a dominant design and the concept of standards that was widely used in the economics literature in the 1980s. They affirm once again that for products that are made from parts, a dominant design amounts to a standard way of assembling the parts into a functional whole. In their discussion of dominant designs, Henderson and Clark (1990) adopt a more structural definition than the previous scholars. Henderson and Clark (1990) also use the concept of a dominant design to refer to standardization at the level of the overall product, but they are more explicit about the requirements that have to be fulfilled before a dominant design can be said to have emerged. A dominant design is characterized both by a set of core design concepts that correspond to the major functions performed by the product and that are embodied in components and by a product architecture that defines the ways in which these components are integrated. It is equivalent to the acceptance of a particular product architecture (p.14). For Henderson and Clark (1990) a dominant design manifests itself when designers converge on a common design approach for all major functions of the product and for the linkages that integrate the components into a functional whole. This definition of a dominant design is striking because of two features. Henderson and Clark require a standard design approach for both components as well as linkages. Second, these authors are silent on the question of the technological capabilities of the dominant design vis-àvis its competitors. The definition entirely lacks a sense that a dominant design is the dominant design because it represents the best technological approach. Tushman and Anderson (1986) start their work on dominant designs by adopting the Abernathy and Utterback (1978) definition of a dominant design as a synthesis of previously introduced design elements. In their later work, Anderson and Tushman (1990) move away from the synthesis idea and adopt a more structural view, as do Henderson and Clark (1990). By defining a dominant design as a single architecture that 17

18 establishes dominance in a product class (1990, p.613), Tushman and Anderson, however, take a much less restrictive view than Henderson and Clark (1990). Anderson and Tushman use the term architecture in a very broad sense that leaves open how many of the components and linkages have to become standardized across designs in order to constitute a dominant design. By adopting such abstract definition of a dominant design, the authors do not commit themselves to a very concrete set of requirements that have to be fulfilled to find positive evidence that a dominant design has emerged in a product category. On this definition any design feature that becomes the standardized across different design approaches could in principle qualify as a dominant design. While this broadening of the definition of a dominant design has the advantage that it can accommodate researchers who study a particular component in a technological system, it comes at the expense of introducing even further ambiguities into the concept of a dominant design. In contrast to their imprecise qualitative account of dominant design, Anderson and Tushman (1990) are very clear on the numerical threshold a design has to overcome to qualify as a dominant design. For a dominant design to exist, they demand that a single configuration or a narrow range of configurations must account for over 50 per cent of new product sales or new product installations. Not only this requirement differentiates Anderson and Tushman (1990) from Utterback and Suarez (1993, 1995). The former authors, in stark contrast to the latter ones, contend that a dominant design can only be known in retrospect and not in real time. Putting the definitions of a dominant design that have been offered by different scholars side by side creates a canvas filled with ambiguities and questions. It is not surprising that researchers trying to build on the existing literature find it difficult to extract a consistent set of principles that can guide research on dominant designs. Some ambiguities can be removed by taking a closer look at the evidence already available. But others will require a great deal more empirical and theoretical research. For instance, the notion that a design becomes a dominant design because it is the best technological approach is not supported by the evidence that has been accumulated since The Productivity Dilemma was published in Scholars in organization theory, strategy, history of technology and particularly economics have shown both theoretically and empirically that is quite simple for an technologically inferior product to become the 18

19 dominant design (Anderson and Tushman, 1990; Cusumano, Mylonadis and Rosenbloom, 1992; Farrell and Saloner, 1985). Utterback and Suarez (1993 and 1995) began to recognize this evidence but they were not able incorporate it into their theory of dominant designs. Other ambiguities are much harder to remove. Abernathy and Utterback s (1978) insistence, for example, that dominant designs need to have sufficiently broad appeal is empirically very difficult to operationalize. Levinthal (1998) has emphasized that technologies frequently undergo changes in new user environments and later reinvade their original user environment. This makes it difficult to pinpoint the precise moment when a design has broad appeal. Furthermore, how should one go about identifying when users have very different requirements that cannot be met by the same design? Which users should be grouped together and which ones apart? Do small, medium, and large cars all constitute different user segments and require the researcher to look for three different dominant designs, or do they all fall into the same segment and thus require the researcher to search for one dominant design? If the latter is the case, one could also wonder whether small trucks should fall in the same segment, etc. These are questions every researcher has to address but they are difficult to make tractable in a theoretical way. In our view, the safest way to proceed currently is to consider a number of alternative definitions of relevant user segments and determine how sensitive the empirical results are to changes in classifications. Empirical researchers have handled this ambiguity actually quite well by mostly picking user segments that have a great deal of face validity because they follow widely shared definitions of markets. While researchers have been able to circumvent this theoretical problem raised by Abernathy and Utterback s definition with considerable skill, they have experienced much greater difficulties in forging an agreeing on how to analyze a given technological product class and properly identify dominant designs. b. Disagreements about Units and Levels of Analysis The Abernathy and Utterback model as articulated in chapter four of The Productivity Dilemma (1978) leaves the strong impression that dominant designs are a phenomenon that occurs at the level of the entire product. Because the authors emphasize 19

20 that a dominant design is a synthesis of previous independently introduced innovations, they appear to exclude the possibility that the concept of a dominant design could also apply to single components that constitute the Ford Model T automobile or the DC-3 airplane. Although the Abernathy-Utterback concept of a dominant design was formulated with regard to the entire product, researchers have often not followed the model with regard to the unit of analysis. Instead of assembling evidence that standardization has occurred in all functional domains and their linkages (to follow the more precise Henderson and Clark 1990 definition), other researchers have frequently focused on standardization in one or a few components. For instance, Anderson and Tushman (1990) examine standardization in kiln length and the heating subsystem rather than standardization in all functional domains of a cement production system. In their study of minicomputers, they similarly pick out two functional components the central processing and the memory unit to characterize dominant designs. Rosenbloom and Cusumano (1987) apply the concept of a dominant design to the scanning head of a video recorder, one out of many components that make up this technological system. In their recent papers, three of the eight products that Utterback and Suarez (1993, 1995) study are components of larger systems. Likewise when Baum, Korn, and Kotha (1995) examine the emergence of a dominant design in the facsimile industry, they focus on standardization in the interface component instead of studying standardization in the design of the overall facsimile technology. Focusing on one or a few components of a larger system by itself would not be a problem if all authors proceeded in the manner of Rosenbloom and Cusumano (1987). Given the evidence they possess, they draw the valid inference that a dominant design emerged with regard to how firms went about designing the scanning component of video recorders, but they resist making claims about dominant design at the level of the entire video recorder. Proceeding in the way of Rosenbloom and Cusumano would simply expand the Abernathy-Utterback model to components of larger technological systems. In the other parts of the book where he details the sequence of innovations in the automobile industry, Abernathy (1978) himself did not strictly follow the spirit of the model formulated in the chapters co-authored with Utterback. When he set himself to organize and describe the sequence of innovations in automobiles, he applied the concept 20