ArticlePDF Available

Modularity-In-Design: An Analysis Based On the Theory Of Real Options

August 1998

Authors:

Kim Clark

Church of Jesus Christ of Latter-day Saints

Carliss Y Baldwin

Harvard University

This paper investigates the economic role of modularity in the design of new products and processes. We construct a formal model of the design process based on the financial theory of options, and analyze the benefits and costs of breaking apart a complex design into independent modules linked by pre-established interfaces. The model provides a rigorous framework for dealing with the effects of uncertainty, and shows how modular designs, independent experiments and testing technologies interact to create economic value. The model also shows how modularity rests on and contributes to the knowledge and organizational structure of a company.

Modular Partitions

…

shows what a heavy burden system-level testing imposes on modular designs. If system-level test costs are on the order of 10% of V(1,1) as shown in the first panel, the option value created by modularity is completely swamped by the combinatorial cost of testing. The optimal configuration of the development program as a whole is to run nine experiments on the Figure 4

…

Figures - uploaded by Carliss Y Baldwin

Content may be subject to copyright.

Content uploaded by Carliss Y Baldwin

Content may be subject to copyright.

Modularity-in-Design:

An Analysis Based on the Theory of Real Options

Carliss Y. Baldwin and Kim B. Clark

First Draft

March 1992

Revised Draft

December 1992

This Draft

July 1994

We thank Robert Hayes and Robert C. Merton for stimulating discussions that contributed to the

ideas found in this work. We also thank Kent Bowen, Sugato Bhattacharyya, Alfred Chandler,

Clayton Christensen, Dana Clyman, Benjamin Gomes-Casseres, Marco Iansiti, Michael Jensen,

Tarun Khanna, Bruce Kogut, Joshua Lerner, Daniel Levinthal, Michael Long, Anita McGahan,

Andre Perold, Dan Raff, Richard Tedlow, Karen Wruck and seminar participants at Harvard

Business School, Boston College, Boston University, and the TIMS/ORSA and Financial

Management Association annual meetings for insights that led to us to clarify and extend the

model. We especially thank the Editor, Abraham Seidmann, the Associate Editor (name if

permitted) and three referees (name if permitted) for detailed and constructive comments on a

previous draft. The finanicial support of the Harvard Business School Division of Research is

gratefully acknowledged. Any mistakes or omissions are ours alone.

Modularity-in-Design: An Analysis Based on the Theory of Real Options

Carliss Y. Baldwin and Kim. B. Clark

Abstract

We investigate the economic role of modularity in the design of new products and

processes. “Modularity-in-design” is an intermediate stage in a spectrum that ranges from

“modularity-in-production” to “modularity-in-use.” In this paper, we describe each stage of

modularity, give examples of products in each stage, and show how designs can -- but do not

have to -- move from one stage to the next. Although the three stages are linked in an

evolutionary sense, each type of modularity involves a unique set of benefits and costs, and thus

requires a separate economic analysis.

The remainder of the paper focuses on modularity-in-design. We construct a formal model

of the design process based on the financial theory of options. We analyze the benefits, costs and

managerial implications of breaking apart a complex design into independent modules linked by

pre-established interfaces. The model provides a rigorous framework for dealing with the

economic effects of uncertainty, and shows how modular designs, independent experiments and

testing technologies interact to create economic value. The model also highlights specific ways in

which modularity rests on and contributes to the knowledge base and organizational structure of a

company. The structure of the model and insights gleaned from the analysis may be useful in

formulating design strategy, organizing development programs, or supervising investments in

design capabilities and knowledge.

Modularity-in-Design: An Analysis Based on the Theory of Real Options

Carliss Y. Baldwin and Kim. B. Clark

1. Introduction

The design of a product -- its form, its function, its aesthetics, the details of its

engineering -- has a powerful effect on its economic value. Its effect on customer perception and

experience is, of course, direct. Design can also have a dramatic effect on how a product is made

and maintained, and thus on its cost, its quality and its reliability. Therefore, in the last decade,

the concept of “design performance” has been expanded beyond what the customer sees to

include the product’s ease of manufacturability, assembly, and servicing (Nevins and Whitney,

1989). But the impact of design goes even deeper. As increasing numbers of firms face

competition centered on rapid product development, a number of researchers have turned their

attention to the effects of design on the development process itself.1 From these studies,

“modularity” has emerged as a pervasive theme and recommended best practice.

A “modular design approach” breaks apart a complex system into components that

function independently. The components are linked in ways that minimize incidental

interactions, often using standard interfaces. For example, the familiar modular phone plug is a

standard mechanical interface for connecting devices to the telephone system. Devices must also

conform to electrical and informational interfaces (current and code requirements) in order to

send or receive intelligible signals over the wires.

The economic benefits of modularity are numerous and complex. In manufacturing,

modularity contributes to scale and scope economies in production. In software, it makes code

more readable and hence easy to maintain. In product development, it makes pieces of an overall

design reusable resulting in greater efficiency and higher speed in the development process. For

users, modularity makes possible customized designs, lower maintenance costs and more

1See for example, Von Hippel (1990), Clark and Fujimoto (1991), Eppinger (1991) and Ulrich (1992).

efficient upgrading over the lifetime of the product.

However, a single design approach cannot obtain all of these benefits all the time. Indeed,

the existence of such diverse benefits suggests that there are in fact different kinds of modularity,

with different sources of value, and different implications in the market. Furthermore,

numerous studies of engineering design show clearly that modularity is not an inherent feature of

the product, but must be created through specific design and engineering choices, and through

specific investments. And while the benefits of modularity seem, in theory, to be quite

significant, there is wide variation in firms' ability to realize those benefits in practice.

Part of the challenge of realizing the value in modularity lies in the fact that an effective

modular design requires several kinds of knowledge -- component knowledge (how the pieces of

a system work); architectural knowledge (how components may be brought together to make a

functioning system); interface knowledge (the details of interconnections and how they affect

performance); and testing knowledge (how to calibrate performance at both component and the

system level). Prior research has shown that it is quite easy for certain types of knowledge to

“fall between the cracks” of an organization. For example, it is not uncommon for organizations

to pursue component knowledge, but to neglect knowledge of architecture or interfaces.

(Henderson and Clark, 1990; Christensen, 1994). Such firms may have modular designs, but

system performance will suffer and potential value will be lost. Moreover, firms that fall into

this trap may be vulnerable to competitive attack by firms that are able to combine existing

elements of a design in new ways. The upshot is that unless a firm invests in a coherent set of

supporting capabilities, a modular design will have far less impact than it could.

Modularity is thus attractive, but complicated. Managers charged with evaluating

potential investments in modularity must ask: What is the purpose of this investment? Will it

reduce manufacturing cost? Will it make the design process more efficient? Will it be valued by

consumers? What else must be done to make modularity effective? These questions in turn

establish a broad agenda for research.

The goal of this paper is to understand the economic benefits and costs of modularity in

the design of new products and processes. Drawing on many lines of research in engineering,

technology and design management, we construct a formal model of the design process based on

the financial theory of options. We use the model to analyze the effect of modularity on the

expected performance, speed and efficiency of the development process. The model provides a

rigorous framework for dealing with the economic effects of uncertainty, and shows how

modular designs, independent experiments and testing technologies interact to create economic

value. The structure of the model, and quantitative insights gleaned from the analysis, should be

useful to managers who are responsible for formulating design strategy, organizing development

programs, or supervising investments in design capabilities and knowledge.

The plan of the rest of the paper is as follows. In section 2, we review the literature,

showing how three types of modularity with different economic implications arise at different

stages in the rationalization of a product design. We then focus on modularity as a design

practice that affects new product and process development. Section 3 develops the basic model,

and considers the effect of different degrees of modularity on economic value. Section 4

analyzes the interaction of modularity with other investments to create additional value through

experimentation, variety and speed. Section 5 examines the cost of modularity, independent

experiments and testing technologies. Section 6 relates modularity to the evolving knowledge

base of a firm and describes organizational implications. Section 7 concludes.

2. Stages of Modularity

Discussions of modularity can be found in technical journals and textbooks in a number

of fields.2 Because of the diversity of applications, it is difficult to integrate the many separate

views of modularity into a single over-arching concept. However, much of the diversity in the

literature can be explained by the fact that different types of modularity arise at different stages

of a product’s evolution. For our purposes it is useful to distinguish three stages: modularity-in-

2A partial list would include: in architecture, Alexander (1964); in software, Parnas (1972), Stevens, Myers and

Constantine (1974), Myers (1975) and Kernighan and Plauger (1976); in mechanical engineering, Rinderle and Suh

(1982); in industrial design, Ulrich (1992); in manufacturing, Nevins and Whitney (1989), Han, Hitami and Yoshida

(1985); in automotive design, Clark and Fujimoto (1991) and Eppinger et al (1994); and for general product

development Rothwell and Gardiner (1988) and Von Hippel (1990).

production, modularity-in-design, and modularity-in-use. As shown in Figure 1, these stages

follow a logical sequence: modularity-in-design implies some degree of modularity-in-

production and modularity-in-use presupposes some degree of modularity-in-design. However,

each type of modularity offers a distinct set of economic benefits.

Modularity-in-production rationalizes a product into components and allows parts to be

standardized (e.g.. all screws the same size). Modularity-in-design goes a step further: the

product is decomposed into a set of independent subunits, which can be mixed and matched to

design a complete system. Finally, a product becomes “modular-in-use” if consumers can mix

and match components to arrive at a functioning whole. A significant degree of control over the

design is thereby transferred from the firm to its customers.

Although each stage of modularity builds on the one before, progression from one stage

___________________________________________________________________________

Figure 1

Stages of modularity and associated economic benefits

Stages Modular-in-Production

Modular-in-Design

Modular-in-Use

Economic Manufacturing

Benefits Efficiencies

Innovation

Efficiencies

Consumer

Options

Economies of scale

and specialization,

"Variety reduction"

through component

swapping and

sharing; Inventory

and flow line

economies

Mix and match of

design components;

Independent

experimentation;

Testing and

selection at the

component level

Mix and match at or

after purchase;

Component selection

by the user; Flexible

upgrading

the next is not inevitable. Products whose designs are difficult to improve, such as some foods3

or simple tools,4 may benefit from modularity-in-production, but gain very little from the mix

and match flexibility of modularity-in-design. In other cases, elements that are modular in the

design phase become irreversibly fused during production. For example, a jet aircraft or a ship

may be designed as a set of modular units, with a high degree of flexibility and customization.5

But once the parts have been welded together, the finished products cannot be reconfigured

except at great cost. Ships and aircraft must sacrifice modularity-in-use to structural integrity,

and in this respect differ from computer systems or living rooms, whose elements can be

reconfigured and upgraded at the user’s whim.6

But in many products the nature of modularity does develop over time, moving in an

evolutionary pattern from manufacturing to design, and from design to use. However, each

degree of modularity must be “purchased” through an investment in knowledge and supporting

systems. The returns to those investments -- the benefits of different degrees of modularity -- are

different in different stages.

In the earliest stages, manufacturing efficiencies are the primary source of gain. The

focus of design efforts is often to reduce the total number of different parts used in a product

family. Benefits of this type of modularization include economies of scale in parts manufacture;

inventory savings; and a reduction of complexity in materials flow and scheduling. In addition,

because of the reduction in complexity, a given facility may be able to handle much higher levels

of product variety than before. Group technology, variety reduction programs and design-for-

3An Oreo cookie has built-in modularity but, many would argue, no room for improvement. In contrast, mix-and-

match “designer pizzas” are a staple at many restaurants.

4Most hand tools have simple and hence very stable designs. Cooper Industries made a virtue of simplicity: its

strategy was to purchase hand tool companies, and invest in process technology, advertising and distribution, while

cutting back on variety and new product R&D. For these products, the potential benefits of design improvement were

judged not worth the effort of R&D. (Stuart and Collis, 1991)

5Rothwell and Gardiner(1988); Nevins and Whitney (1989).

6Reality is naturally quite a bit more complicated than Table 1 or the above examples indicate. In any complex

design, different parts will exhibit different degrees of modularity. For example, most automobile engines are

modular-in-production, that is, their designs have been rationalized to take advantage of economies of scale in casting

and machining, and they exploit economies of scope by sharing components within product families. However, the

tires of a car are relatively easy to remove and replace, and hence are modular-in-use. Despite this complexity, we

still think it is useful to describe a design as having an “average” degree of modularity.

manufacturing all use modularity-in-production as a technique for reducing manufacturing

costs.7

In the course of rationalizing parts for production, firms must collect and codify

information about the product’s basic structure.8 At this point the product develops an

architecture, i.e., a hierarchical organization of elements that defines the product and its variants.

Modules (or components) and the interfaces that connect them are a central feature of the

product's architecture and make possible innovation through variation of specific modules within

the overall architecture. Such modularity-in-design -- coupled with experimentation and testing -

- can improve the efficiency and speed of the development process. The nature of these

“innovation efficiencies” and investments needed to realize them are the primary focus of this

paper.

Once a firm has achieved modularity-in-design, moving to modularity-in-use transfers

choices that were once the designer's to the user. This transfer of “decision rights” is very

valuable to consumers who prize variety (What will I want to wear tomorrow?), or who face

uncertainty about future needs (What are the information needs of my company?).9 Therefore

consumers may be willing to pay a premium for products that are modular-in-use. However there

are risks in offering such products for once the consumer has control over what goes into the

system, he or she may add or subtract from it in ways not contemplated by the original designers.

Thus a firm introducing a product that is modular-in-use must be prepared to face competition

focused on subelements of the system.10

7For details on these techniques, see Han, Hitami and Yoshida (1985); Nevins and Whitney (1989); Suzue and

Kohdate (1990). For models and additional examples, see Rothwell and Gardiner (1988), Shirley (1990) and

Sanderson (1991).

8For example, Shirley (1990) describes how product analysis aimed at reducing production costs led to the creation of

a “library” of designs organized by module.

9On the value of decision rights generally, see Jensen and Meckling (1992). Wruck and Jensen (1994) describe the

efficiency gains from decentralizing decision rights within an integrated chemical plant. In effect, they describe the

modularization of a production process (as opposed to a product design).

10The classic example of competitive entry focused on a subset of a modular system is the entry of plug-compatible

peripheral manufacturers in the wake of IBM’s System/360. (Dorfman, 1986; Pugh et al, 1991; Christensen, 1994)

Other studies that focus on the effects of modularity-in-use on the development of particular markets include:

Langlois and Robertson, 1992 (stereos and microcomputers); Langlois, 1992 (cluster tools); Garud and

Kumaraswamy, 1993, (workstations); Gomes-Casseres, 1994 (microprocessors). In industrial economics, several

recent papers examine theoretical equilibria in markets where consumers can “mix and match” components of a

Overall the evidence in the literature points to important advantages in modularity, but

the creation of modularity and particularly the evolution from one degree to another is not

inherent in the technology, nor is it inevitable. The pursuit of modularity depends on choices the

firm makes, and yet the literature leaves open many questions about how modularity creates

economic value and how much modularity to pursue in production, design or use. On the one

hand, the economic benefits of modularity-in-production are fairly well understood. By

permitting parts sharing and swapping, modularity delivers classic scale and scope economies;

by reducing the number of basic building blocks, modularity also reduces the costs of

complexity.

On the other hand, less is known about the economics of modularity-in-design. Design

theorists in a number of fields have stressed the advantages of breaking apart designs into

flexible and reusable elements.11 Writing about development processes in general, Von Hippel

(1990) argued that the way in which tasks are subdivided can have a large impact on the speed,

efficiency and final outcome of the design effort. He drew on the early contributions of Simon

(1969), Marples (1961) and Alexander (1964) in framing his arguments, but, although he

discussed the conditions that ought to govern task partitioning, he did not attempt to develop a

modeling framework to analyze development choices.

In a recent seminal paper, Ulrich (1992) defined the concept of a product architecture and

argued that the choice of a “modular architecture” over an “integrated architecture” will affect

the performance, variety and efficiency of development efforts.12 Complementing Ulrich’s

taxonomic approach, Eppinger and his colleagues have mapped the actual tasks involved in

particular design processes, and shown how task ordering and partitioning can enhance or

complete system. See, for example, Matutes and Regibeau (1991); Economides and Salop (1991) and Farrell, Monroe

and Saloner (1993).

11Alexander(1964); Simon(1969); Myers (1975); Kernighan and Plauger (1976); Suh (1990).

12Ulrich proposes that there are three basic modular architectures: (1) slot architecture (each component has its own

slot); bus architecture (every thing connects to a common central component); and (3) sectional architecture (all

interfaces the same; no single central element). Ulrich’s taxonomy focuses on interfaces, while others have stressed

the hierarchical relationships among components of the system (see Marples, 1961; Parnas, Clements and Weiss,

1985; Henderson and Clark, 1990; and Christensen and Rosenbloom, 1994). As yet, there is no robust, yet

parsimonious taxonomy of modular systems.

impede the overall efficiency of development.13 Meanwhile, a large body of economic research

has shown that product development strategy and performance -- including development

efficiency, design quality and time-to-market -- dramatically affect competitive outcomes and

long-term industry structure.14 However, the economic studies tend to view modularity as a

technological “given,” not as a choice made by engineers and managers.15 (This is a major

difference between the economic and engineering literatures on modularity.)

For companies competing on the basis of timely and innovative products, modularity is

an important strategic choice that demands the attention of managers overseeing research, design

and development processes. Technology creates possibilities for modularity, but where to split a

design, how to assign tasks, how to link components, and, especially, how much to spend on

modularity are choices that managers must make. Their decisions in turn should be based on

informed judgment about the efficiency and other competitive advantages that may be obtained

via a modular design.

3. The Model

In practice, creating a modular design that delivers value in the market requires both

technical creativity and careful consideration of the financial costs and benefits associated with

the design. Although our model is not couched within a specific technology or engineering

discipline, it does attempt to capture the technical context in which modularity choices are made,

and to link the technical choices to financial costs and benefits. As a result, the model draws on

multiple disciplines and attempts to communicate results of interest to several different kinds of

scholars and practitioners. Our challenge in this section is to communicate what the model is

doing, and to build a common language and basis for analysis and evaluation. To meet that

13Eppinger (1991); Eppinger and McCord (1993); Eppinger et al (1994). For a synthesis of Ulrich’s and Eppinger’s

work on product design and architecture, see Ulrich and Eppinger (1994).

14A very partial list of economic studies would include: Foster (1986), Henderson and Clark (1990), Trajtenberg

(1990), Cusumano (1991), Clark and Fujimoto (1991), Ferguson and Morris (1993) Christensen (1994), Utterback

(1994) and Gomes-Casseres (1994).

15Richard Langlois’ work on the microcomputer industry is a noteworthy exception (Langlois, 1992; Langlois and

Robertson, 1992).

challenge we have found it useful to lay out our basic assumptions, and to discuss them in some

detail before presenting the model's formal results.

We take as our basic starting point a design team (who act as one) faced with the problem

of designing a complex, technologically uncertain product for a market of identical customers.

Customers make purchase decisions based on product performance. The product’s performance

is in turn the result of a large number of individual decisions made by the design team, but the

precise relationship between individual design decisions and ultimate performance cannot be

predicted in advance.16

The detailed assumptions used to construct the model fall into three basic categories: (1)

the organization of the design process; (2) the impact of design decisions on performance and

value; and (3) the nature of uncertainty. For each category, we list our formal assumptions first,

and then discuss the reasoning behind our choices.

3.1 The Organization of the Design Process

We assume:

(1.1) The design process consists of a set of tasks, numbered 1 to N. Each task requires a

design decision and results in the specification of a design parameter. The design

parameters in turn affect the performance of the product (as viewed by consumers), but

the precise impact of a particular decision on performance is uncertain. There is a

“current model” of the product which represents a feasible combination of the N design

parameters.

(1.2) At the outset of the design process, tasks are partitioned, i.e., divided up among different

design groups. There are obviously a very large number of ways to partition N tasks. A

16In the interest of transparency, we have had to suppress many important and interesting features of the real world.

For example, in reality, design processes typically involve many people with different (and sometimes conflicting)

objectives. Customers have widely varying tastes. Price is not the same as willingness to pay, but is set through the

interaction of demand, cost, volume and competitive behavior. All of these important concerns are omitted from the

model so that we may focus on the linkages between design and value. Fortunately, the model’s results are not very

sensitive to the precise specification, and thus insights gleaned from it can be used to guide strategy and evaluate

investments in development capabilities in a wide range of contexts.

modular partition divides the tasks into two or more subsets so that the detailed decisions

within each subset can be made without reference to those in other subsets.

(1.3) Purely as an analytical device, we assume that the total number of tasks, N, is divisible by

whole numbers up to j: N = j! A symmetric modular partition divides the N tasks into

independent subsystems each with an equal number of decisions. Below we will evaluate

the effects of symmetric and asymmetric modular partitions on the speed, efficiency and

performance of the design process.

Discussion. The characterization of a design process as a set of tasks and/or decisions is common

in the literature and not controversial.17 However, we focus on a situation where the design

team confronts both complexity and uncertainty, so that the design process is unstructured. In

practice, designers approach unstructured design processes in a hierarchical fashion. Initially, the

overall task is broken up into large blocks, and then smaller tasks are defined and assigned

within each block. This procedure continues and results in a multi-layered hierarchy of design

(Marples, 1961; Clark, 1985). Then as individual design decisions are made they are linked up

and tested for mutual consistency.18

A modular design approach is essentially a strategy for attacking an unstructured design

problem. The basic idea is to create “modules” within the overall design, which can be

segregated and addressed as independent subproblems. Designers divide their tasks so as to

maximize the interaction among elements within a module and minimize the interaction of

elements across modules (Alexander, 1964; Myers, 1975; Suh, 1990; Ulrich, 1992). But an

effective system solution to the overall design problem requires that designers also develop ways

17Eppinger and his colleagues have actually managed to list the tasks involved in several highly structured design

efforts. However, a question arises as to whether comparable lists can be constructed when the underlying processes

are not well understood (Eppinger et al 1994). In framing our assumptions we rely on the fact that practitioners and

design theorists across a range of fields are united in saying that modular partitions can be found or constructed for

most complex designs.

18The process of integrating individual design choices is complex and costly. Repeated cycles of problem-solving

may be necessary to get the design to work, and the end product may be an inelegant compromise. Over the last ten

years, a huge literature has developed around the theme of integrating design processes. See, for example, Nevins and

Whitney (1989); Clark and Fujimoto (1991); Iansiti, (1993).

to link modules together. Therefore, to make modularity work, designers must find a way to

achieve module independence and module integration at the same time. Software development

illustrates the dilemma and its solution.

A key development in the application of modularity to software was the principle of

“information hiding” initially put forward by Parnas (1972a,b). Parnas reasoned that if the details

of a particular software module’s design were hidden from other modules, then changes to that

module could be made without changing the rest of the system. The idea of separating function

(what the module does) and structure (how it accomplishes its function) was then adopted by

Constantine, Myers and Stevens (1974) as a tenet of the “structured” or “composite” approach to

software development.

The concept of information hiding implicitly requires that well-defined interfaces exist

between different subelements of a design. In software engineering the ideal interface consists

of a functional name, and a minimal set of data transferred to and from the module. The details

of how the module accomplishes its function should not be “visible” to the program that “calls”

it. Equally important, the module should operate on a general data set, and not have to "know” in

detail how the data on which it operates were stored, organized and addressed (Myers, 1975;

Schach, 1993).

More generally an interface defines the rules that a given subsystem design team needs to

follow to ensure that the results of their work can be integrated effectively with the work of other

teams. The interface thus embodies knowledge about the interactions between subsystems, and is

critical to achieving overall system performance.19 In the final section of this paper, we discuss

the knowledge that is needed to arrive at a modular partition, and describe how such knowledge

19Eppinger et al (1994) observed that “decoupling” tasks in a large design (which is the same as modularization) can

be accomplished by inserting another set of tasks earlier in the design process. The earlier tasks, in addition to

dividing the effort in a sensible way, also establish the interfaces that subsystems must adhere to. The effort of

designing interfaces is thus an investment that must be recouped later in the process. In practice, interfaces are

usually not perfect. Effective integration then requires modifications and mutual adjustments in components to

respond to unanticipated interactions. This is particularly true when the new design extends beyond established

practice; integration then requires joint problem solving and close interaction between specialized components groups

(Clark and Fujimoto, 1991; Bacon et al, 1993; Iansiti, 1993) All of this underscores the fact that modularization is an

evolutionary process that is based on and in turn supports knowledge building in an enterprise.

is likely to emerge and evolve over time. At this point, however, we simply posit the existence of

modular partitions and related interfaces that allow the design to be divided up into independent

subsystems.

3.2 The Effect of Design Decisions on Value

We assume:

(2.1) All consumers evaluate products in the same way: faced with two versions, all will agree

as to which is better. The “total performance” of the product, denoted X, reflects the

consumers’ willingness to pay: higher X translates into higher revenues and a higher

return to the development effort. Without loss of generality, the performance of the

“current model” is set to zero.

(2.2) The product has many attributes. Fundamentally the design parameters (hence the

designers’ collective decisions) determine the product’s attributes and the consumers’

willingness to pay. For simplicity, we assume that each design parameter contributes to

total performance in a specific and separable way, and thus X may be decomposed into a

sum of the shadow values corresponding to individual design parameters:

X = Σ xi ;

i=1

(2.3) The shadow values of specific design parameters are unknown to both the consumers and

the designers. Designers do not know the exact effect of any single design decision on the

product’s overall performance and value. Consumers in turn can only evaluate realized

products. Given two versions of the product, consumers can express their preferences, but

they cannot explain why a given set of design parameters has a particular value.

(2.4) At a specific point in time, T, the design team fixes the N design parameters and builds a

prototype. The prototype’s overall performance is measurable: the design team can

observe X, and compare its performance to that of the current model. In addition, if the

design was initially partitioned into modules, the design team can evaluate modular

combinations, that is, it can mix and match modules of the prototype with complementary

modules of the current product.

(2.5) Building and evaluating a prototype involves a cost, which we call “the cost of testing.”

Discussion. Compared to the rich set of descriptions of the design process, much less has been

written about how individual design decisions get translated into economic value. For a complex

product, the chain that connects design to value is long. At one end, technical factors may restrict

performance and slow the rate of problem solving. At the other end, consumer perceptions

(which may be evolving along with the product), pricing and competition all affect the value

captured by the company.20 In general, technical hurdles are hard to measure, consumer desires

are hard to understand and competitors’ actions are hard to predict, and yet the value of a new

product depends on the interaction of all three!21

Faced with this complexity, we have elected to map design decisions onto value in a very

simple way that still preserves the essential structure of design processes. At the core of our

assumptions are three basic observations: (1) design affects product performance, which in turn

affects consumers’ willingness to pay; and (2) products have multiple attributes (many

dimensions of performance), which consumers weigh in some fashion to arrive at a final

valuation; and (3) the design process takes time and candidate designs must be tested. We think

these observations are self-evident for many design processes, and hence should not be

controversial.

We go on to assume that (1) product attributes have a one-to-one correspondence with

design parameters; (2) consumers’ perceptions can be decomposed into several dimensions, with

overall value being a simple sum of the “shadow value” of each dimension; (3) the design

20On the evolution of consumer perceptions in response to new designs see Clark (1985) on the “hierarchy of

concept.” Evidence of changing preferences based on developing knowledge about a product can be found in

Trajtenberg’s (1990) study of the market for medical imaging technology.

21Recently there have been a few attempts to map technical, consumer and competitive parameters into financial

value using modern management science and financial modeling techniques: see, for example, Cohen et al, 1993

(new food products) or Nichols, 1993 (drug development at Merck). Mostly, however, research, design and

development efforts have not been subjected to rigorous finanacial valuation (Baldwin and Clark, 1992). This may

change as a result increased capital market oversight (Hall, 1990; Jensen 1993).

process takes a fixed amount of time and testing involves a fixed cost. These specific additional

assumptions were made to simplify the model. They should not be taken as literally true, but

viewed as approximations of a more complex reality.22

3.3 The Nature of Uncertainty

We assume:

(3.1) A change in any design parameter will have an uncertain impact on the value of the

system as a whole.

(3.2) The probability distribution of aggregate value is unchanged by modularization.

(3.3) (To simplify the analysis) (a) every modularization results in a normal distribution of

value for each module; (b) the expectation of value for each module is zero; and (c) the

variance, denoted σn2, for a module comprising n design parameters, increases linearly

with time (T) and the complexity (n) of the design problem: σn2=σ2nT where σ2 is a

parameter of the design process.

Discussion. The observation that design choices have uncertain effects on the performance of

complex systems pervades the literature and is a prime rationale given for decomposing designs

into modular -- independent -- subunits.23 Uncertainty in the relationship between parameters of

design and system value complicates the design problem, but in so doing creates new sources of

value. Uncertainty arises because of incomplete knowledge, either about technical relationships

(i.e. the design parameter-system performance connection), or customer response (i.e. the system

22In general, the qualitative results of the model carry over to more complex specifications. For example, the

mapping from design parameters to value can be many-to-many (many design decisions affecting many functions);

can involve intermediate variables (functions the consumer does not perceive directly); and can be nonlinear. The

time horizon and the cost of testing may be uncertain; or triggered by an internal event (like attaining a certain

performance goal). Finally, the development process may have several well-defined stages and include multiple

rounds of designing, screening and testing. Any of these extensions can be incorporated within the basic structure of

the model, although notational transparency would be lost with each additional complexity.

23The research of Eppinger and colleagues (1991; 1993; 1994) into the fine structure of design processes has shown

that there is indeed great uncertainty about the impact of individual design parameters on even well-understood

designs. Numerous empirical studies in finance and new product development show the larger effects of uncertainty

in high failure rates for new products and new ventures.

performance-system value connection).

Choices designers make get translated into value once the product is introduced in the

market. But before introduction, the value of a particular design is uncertain (and could even be

worse than the current product). Thus, anything the designers can do to create and act on

information in advance of the market will have significant value. Said another way, given the

uncertainty we have assumed, building options into the design process and acting on them as

information develops can have a huge impact on technological outcomes and the value of the

effort.

Of course, one can imagine a world of no uncertainty. In that world designers would

know exactly how their choices mapped into value and there would be no possibility of

introducing a design of less value than the current product. Designers would simply restrict their

choices to the set of value-increasing parameters, and would not need to build options into their

designs. That is a world of no surprises, no setbacks, no failures, no breakthroughs. It is not the

world we model. What we are after is a model that sheds light on the halting, groping, and

searching that characterizes research, design and development processes for complex products

ranging from new medical procedures to the latest in consumer electronics.24 A modular design

is essentially an options-based approach to a complex, uncertain design problem, and thus we use

an options-based model to investigate the value of modularity and how best to exploit it.

Assumptions 3.2 and 3.3 put some additional structure on the problem in order to

simplify and focus the model. The first of these deals with the effect of modularization on the

distribution of outcomes. The question is this: Aside from option value, are designs likely to be

better or worse as a result of a modularization? Design theorists and engineers are divided on

this issue, and a full answer clearly depends on the physical properties of the object being

24From the outside, these halting, groping processes may be indistinguishable from progress under certainty, because

the sequence of products actually introduced may still be marked by continuous improvement. “Bad draws” will be

evident only in terms of slippage in the schedule, and if managers have built extra time into the schedule, even these

delays would not be apparent to anyone outside the process. But even though the two processes might look the same

to an external observer, they must be managed very differently. In the first instance, there is no need to worry about

creating or exploiting options. In the second, options built into the process and exercised by managers create

significant economic value.

designed, as well as the designers’ skills both in perceiving opportunities to modularize, and in

designing interfaces.25 In our analysis, we assume that the distribution of value is unchanged by

modularization. This assumption is made for transparency and analytical convenience -- in

reality, modularization may make design outcomes better or worse, but the direction of the effect

will vary from case to case. But assuming modularity is neutral allows us to separate the value of

options created by modularity from its other effects on the design.26

It is also helpful if the distributions have a common shape and location under alternate

modularizations. Thus we assume a normal, zero-mean distribution for the value of the overall

system and all modules. In essence, this assumption suppresses what is going on within modules

(by assuming all modules are intrinsically alike), and allows us to focus on different ways of

dividing and combining modules.27

If module values are independent, additive, and normally distributed (assumptions 1.2,

2.2 and 3.3), and if modularization leaves the overall distribution unchanged (assumption 3.2),

then it follows that the variance of a module is an increasing function of the number of design

decisions in that module. We impose a slightly more restrictive functional form, assuming that a

module’s variance is linear in both the number of design decisions (n) and time (T). In essence,

we are saying that complex projects are more risky, and that to arrive at a dramatically different

design requires more time than to make a few superficial changes.28

25For balanced discussion of this issue, see Von Hippel (1990); Ulrich (1992); Schach (1993); and Eppinger et al

(1994). Alexander (1964), Myers (1975) and Suh (1982) are ardent proponents of modularity. On the other side,

Brooks (1975) takes the position that a “great” design must be the product of a single mind, i.e., any division of tasks

detracts from quality.

26For example, Eppinger and McCord (1993) hypothesize that modularity reduces the number of iterations needed to

complete a design. This efficiency gain would be represented in our model as a reduction in the time needed to

complete the project, and is not an option value. In real world applications, the value of higher quality and/or reduced

cycle time can be added to the value of mix and match flexibility in estimating the total economic impact of a

modular design approach.

27If outcomes (at both the system and module level) can be viewed as the sum of a large number of independent

actions, their distribution will be asymptotically normal by the Central Limit Theorem.

28Many readers will recognize that our assumptions correspond to the definition of classical Brownian motion.

Indeed, the development of a new design can be modeled as a stochastic search process. Diffusion-type searches (for

which Brownian motion is the paradigmatic case) commence with a known design, and then proceed in small steps;

at each step there is uncertainty about where the next step will lead. Diffusion processes have a wide range of

application in the physical sciences, operations research and finance, and their properties have been intensively

investigated. See Harrison (1985) or Merton (1990) on the characteristics of these processes and for references to

related work.

Taken as a whole, our assumptions about uncertainty work best for complex products or

systems, where learning takes place at many points, and where the final design is the result of a

large number of incremental modifications.29 The assumptions are not a good characterization of

processes where the design is simple, or where improvement rests on a radical

reconceptualization of all elements of the design. Thus our model may provide insight into

development processes for computers, automobiles, aircraft and large software systems, but will

not shed much light on how best to design shoes, apparel or furniture.

3.4 The Value of a Modular System

Given the assumptions listed above, we can very quickly arrive at an options-based value

of a modular system.30 We begin by defining a standard normal variate for X and all modular

partitions of X. Let <xi> be the values of the design decisions in a module of size αN, where α is

an integer divided by N, and αN is the number of design decisions in the module. It follows from

the assumptions that

αN

∑ xi

zα = i = 1_____ ;

σ (αNT)1/2

is distributed N(0,1).

Let V1 denote the present value of a process with only one module containing all N

decisions. The final outcome of this process, X, is the sum of the individual xi, and is normally

29Processes exhibiting both incremental and radical change can be modeled through the theoretical device of a

“jump-diffusion” process. Fundamental results of our analysis would be similar under such a process, although the

exact solutions would be different.

30Options theory, a field pioneered by Black and Scholes (1973) and Merton (1973), explicitly deals with

contingencies, that is, optimizing actions taken in response to future events. Classical option theory also provides for

the valuation of complex payoffs via dynamic replication. However, the options that we model may not have

corresponding traded securities, and for this reason, we do not invoke arbitrage in valuing a modular system’s

expected performance. Instead, we assume that future levels of performance are priced using a known discount rate.

This assumption suppresses many complexties of real product markets, which, if incorporated, would change the

details of the model but not the basic results. For a detailed discussion of dynamic replication vs. dynamic

programming approaches to real options valuation, see Dixit and Pindyck (1994).

distributed with mean zero and variance σ2NT. The present value of X, conditional on its being

superior to the current model, is

∞

V1 = e-rT σ (NT)1/2 ∫ z n(z)dz ; (1)

∞

where e-rT is a discount factor applied at time 0 to values obtained at time T, and ∫ z n(z)dz is the

expectation of the right tail of a standard normal distribution and equals .399.

Intuitively, equation (1) says that the development process may result in a product that is

better (X>0) or worse (X<0) than the current model. But on seeing the outcome (i.e., after testing

the design to determine X), the design team has the option to introduce the new version or

continue with the current one. Clearly it will replace the current model if and only if X is

positive. The integral is the conditional expectation of the upper half of a standard normal

distribution; σ (NT)1/2 scales that expectation for the risk of the particular development process;

and e-rT discounts future cash flows to arrive at a present value.31

The essence of modularity lies in the firm’s ability to value subsets of the system and

select for introduction the best of any subset. For a one-module development process, the

company has only two choices: introduce the new product or stick with the current model. If the

product is split into two modules, with a well-defined interface, then the company has four

choices: introduce an entirely new system, keep the old system, or combine one old module with

one new one. Similarly, three modules generates eight candidates, four modules sixteen, etc.32

The combinatorial properties of modules clearly present a new set of challenges for the

design team, since measuring the performance of subsystems requires additional resources. We

shall return to this issue in the sections on testing and knowledge-building below. For now,

31Strictly speaking, if the xi are instantaneously uncorrelated with any traded security, and outcomes are small

relative to the economy as a whole, then the appropriate discount rate is the risk-free rate (Mason and Merton, 1985).

32Implicitly we are assuming that the act of modularization partitions the design of both the old and the new product.

This assumption is most likely to be true when the older product’s design has been partially rationalized for

production (cf. Shirley, 1990). However, there are cases in which the existing product design cannot be partitioned:

for example, software engineers speak of the impossibility of breaking apart “spaghetti code” (Korson, 1986). The

company must then either fund two modular efforts (see the discussion of experimentation below), or wait a

generation to realize the option value inherent in a modular design.

though, we only want to know what the option to select the best subsystem does to the value of

the outcomes, not what it costs to find the best. Basically, as the design team adds modules, it

adds options -- one for each module. But the value of each option falls in proportion to the

decline in its standard deviation.33

Now let the process be partitioned into a finite number of independent modules of sizes

(αN, βN ...), where α, β, etc. are integers divided by N, and (α+β...)=1. Figure 2 shows how a

design task of complexity N may be partitioned into both symmetric and asymmetric modules.

By the same reasoning used in equation (1), the value of this effort, denoted Vα, is:

∞

Vα = e-rT σ (NT)1/2 ( α1/2 + β1/2 ...) [ ∫ z n(z) dz ] ; (2)

We may compare equations (1) and (2), and summarize their relationship via a simple rule of

thumb:

Figure 2

Modular Partitions

No. of

Modules Decisions per Module

1 N

Asymmetric Modular Partitions

2αΝ βΝ

3αΝ γΝ βΝ

Symmetric Modular Partitions

2 N/2 N/2

3 N/3 N/3 N/3

33It may seem counter-intuitive that value declines as risk goes down, but such a relationship is a standard result in

options theory.

Proposition 1. Let a design problem of complexity N be partitioned into independent modules of

sizes (αN, βN ...) as defined above. Then under the assumptions of symmetry of distribution,

normality and the conservation of value, the modular partition has value:

Vα = ( α1/2 + β1/2 ...) V1 ; (3)

relative to an unmodular design effort.

Proof. Immediate from equations (1) and (2).

From the fact that <α, β,...> are fractions that sum to one, it follows that the sum of their

square roots is greater than one. Thus, if we ignore the cost of designing interfaces and testing

outcomes (which may be substantial), a modular design is always more valuable than a

monolithic design of the same degree of complexity. Moreover, further modularizations within

an existing design increase its value: if the α-module is split into sub-modules α1 and α2, then

its contribution to overall value will rise because α11/2 + α21/2 > α1/2.

In general, there is a square root rule at work here. To see this, define VN/j as the value of

a symmetric modular partition of a design into j independent modules of size N/j (see Figure 2).

Substituting 1/j for <α,β,...> in equation (3), and collecting terms, we see that in theory,

symmetric modularizations obey a simple square root rule:

VN/j = √j V1 (4)

Equations (3) and (4) are special cases of Merton's (1973) result, that for any distribution

of underlying value, a portfolio of options is more valuable than an option on a portfolio. In

terms of value, complex products may be likened to a portfolio of stocks. In effect, a modular

design creates options that exploit the underlying independence of subsystems within the

product, just as options within a stock portfolio exploit the independence of individual securities.

The options permit the design team to combine the best of the old and the new design, thereby

enhancing the value of the overall effort.34

4. Exploiting Modularity

Modularity creates value by expanding the set of options related to a design. Under simplified

assumptions Proposition 1 offers a simple rule of thumb for estimating how much value might be

captured by modularizing a planned development program. However, much of the power of

modularity lies in its impact on other design investments. Thus exploiting the full potential of a

modular design often means that the whole development program must be reconceptualized.

New design capabilities may suddenly take on value and resources may need to be directed into

entirely new channels. In this section we explore the interactions of modular designs with other

investments in development capability.

4.1 Experimentation.

In the preceding section, we assumed that the design team would create only one new

design per module. In principle, however, the team could decide to run several experiments in

parallel. There is great diversity in practice and corresponding disagreement about the value of

parallel experimental programs. On the one hand, parallel programs are redundant, and much

cost cutting is aimed at eliminating such redundancy. On the other hand, researchers have

pointed to the presence of redundant programs at successful innovative firms, and recommended

the practice to companies seeking to emulate their success (Quinn, 1985).

In general, the value of "redundant" efforts hinges on the uncertainty of the process and

the degree of independence among the experimental results. Loosely speaking, the value of

34With a change in the definition of certain variables, our model can be extended to options that arise “in the field.”

For example, performance in the field below a threshold x* may constitute a malfunction. “Field service modularity”

allows such malfunctions to be found and fixed without tearing down and rebuilding the whole system. “Modular

upgradability” allows a similar degree of flexibility with respect to subsystem improvements, which the customer

may not want (or which may not be available) at the time the system is purchased. However, field options are the first

step towards modularity-in-use, and (as was indicated in section 2) the original system designer may not be able to

maintain control once modularity-in-use is widespead. For example, modularizing a part that breaks frequently will

enhance the efficiency of field service personnel, but may also stimulate entry by competitors into the replacement

market.

redundancy increases with both uncertainty and independence. Rene Stulz (1982) provided a

general analysis of redundancy in his valuation of the option to select the maximum of two risky

assets. Ron Sanchez (1991) later applied Stulz's framework to the special case of redundant

R&D.

With a symmetric, joint normal distribution, such as we have assumed, redundant

experiments affect value in a particularly transparent way. Assume that on a given module, the

design team pursues k parallel, independent design efforts, and selects the best of these for the

final design. The distribution of the maximum of the k experiments equals the distribution of the

highest component of the order statistic of a sample of k independent draws from a standard

normal distribution. Its expectation conditional on being greater than zero, denoted Q(k), is:

∞

Q(k) = k ∫ z [N(z)]k-1 n(z) dz (5)

where N(z) and n(z) are respectively the standard normal distribution and density functions

(Lindgren, 1968).

Q(k) replaces the integral terms in equations (1) and (2), and the previous results now

appear as special cases. We can immediately generalize Proposition 1 to allow for different

levels of experimentation within each module:

Proposition 2. Let a design problem of complexity N be partitioned into independent modules

numbered 1 to j. Let αi denote the size (degree of complexity) of the ith module. (As in

Proposition 1, the αi’s are integers divided by N, and their sum equals one.) Let ki denote the

number of independent experiments with respect to the ith module (the ki’s are positive integers).

Under the assumptions of symmetry of distribution, normality and the conservation of value, the

value of the overall program is:

V(α1...αj, k1...kj) = e-rT Σ σ(αiNT)1/2 Q(ki) (6)

i=1

Proof. Immediate by substitution of (5) into (2) and rearranging terms.

Equation (6) condenses many different ways of organizing the design process into a

simple valuation equation, It focuses attention first, on the overall complexity of the design task,

and second, on the independence of modules and of experiments within modules. If such

independence is a good approximation of reality (a question that can be answered by looking at

the diversity of thinking across different design groups), then the value of the whole can be

modeled as the sum of the values of the individual modules. The value of each module in turn

depends on uncertainty, which we posited was a function of time and per-module complexity,

and on the number of independent experiments.

Although it should not be taken literally, equation (6) can guide decisions as to where and

how much to experiment within a given design. It shows that experiments are most valuable

where module uncertainty is greatest, and when experimental outcomes are independent (i.e..

when there are several quite different lines to pursue). Modularity also enhances the value of

each experiment. Intuitively, it decouples the results of subsystem experiments and allows the

designers to select the best solution for each subsystem. The overall design then has the

properties of the "best of the best."

The interactions of modularity and experimentation can be visualized for the special case

of a symmetric modular partition.35 If all modules have the same degree of uncertainty and

complexity, then, ignoring costs, it will be optimal to run the same number of experiments on

each module. The 2j arguments of equation (6) collapse to two, and the value of the overall

effort, denoted V(j,k) is:

V(j,k) = e-rT σ(TNj)1/2 Q(k) (7)

35Any modularization must be guided by the internal structure of the design problem itself. To achieve independence

across modules, interfaces must be defined and tasks partitioned according to the natural “fault lines” of the design

and the technologies behind it. Except in very special cases (like software), the natural modules embedded in a design

are unlikely to be symmetric. But symmetric modular partitions are still a useful construct that will permit us to delve

into the interactions between modularity, experimentation and testing in sections below.

Figure 3 graphs this function in three dimensions. The x and y axes are respectively the number

of modules and the number of independent experiments per module, while the z axis is the value

of the overall development program from equation (7). Entries have been scaled so that the

value of one development program for an unmodular design is one: V(1,1) = 1. Throughout the

rest of the paper, all magnitudes will be scaled to this value, which we label “the benchmark

program.”

Figure 3 shows the high degree of interaction between modularity and the design team’s

experimental strategy. On each axis, as the number of modules or experiments rises (holding the

other variables constant), overall value increases at a decreasing rate. For example, one

experiment on each of ten modules is worth 3.16 times the value of the benchmark program; ten

experiments on an unmodularized program is worth 3.86 times the benchmark program. The

highest rate of increase, however, is on the 45o line: ten experiments on each of ten modules is

worth 12.20 times the value of the benchmark program. Thus combining modularity with a

multiple experiments turns out to be far more valuable than single-mindedly pursuing modularity

alone or experimentation alone.36 This means that companies which are in the process of

changing their approach to design need to be aware of the possibilities along both dimensions.

These values and

36Technically V(j,k) is a "super modular" function in the sense of Milgrom and Roberts (1990).

Figure 3

13579

No. of Modules

No. of

Experiments

Value

The Impact of Modularity and Experimentation

on Value -- 3d View

__________________________________________________________________________

the costs of attaining them (discussed below) should be part of managers’ comprehensive picture

of their company's design capabilities.

4.2 Speed

Increased speed is another important benefit of pursuing a modular, multi-experiment

design strategy. One measure of speed is the expected performance of the final product divided

by the time period of development.37 This value, denoted S, is simply the undiscounted value of

the options summed in equation (6), divided by T. For a symmetric modular partition:

S(j,k) = σ (Nj/T)1/2 Q(k) (8)

37 This is an unusual formulation of speed compared to most empirical studies of lead time in product development,

which use pure calendar time. However, most studies recognize the importance of controlling for differences in

product complexity and project content in evaluating project lead time. We have taken that notion one step further by

controlling for the quality of the outcome. Thus, in terms of the literature on product development, our measure of

speed is the inverse of a quality adjusted time-to-market.

If we scale this value relative to S(1,1), the speed of the benchmark program, the result is

identical to the function graphed in figure 3. Thus the figure can be viewed in two ways. If the

firm’s market power allows it to capture all of the consumer surplus, it shows the direct impact

of modularity and experimentation on firm value. Alternatively, if the firm is competing for

transient rents in a technology race, it shows the impact of modularity and experimentation on a

critical strategic variable, the speed of product improvement.

According to figure 3, the impact of design strategy variables on speed can be dramatic:

running ten experiments on each of ten modules results in an expected rate of product

improvement that is 12 times higher than that of the benchmark one-module, one-experiment

effort. In reality, the cost of designing modules, running experiments and testing outcomes will

make such high levels of modularity and experimentation uneconomic. Furthermore, many

variables besides those modeled here, will generally affect time-to-market. But in technology

races, being even a little bit better -- say one-and-a-half times faster in terms of speed -- can be a

dramatic competitive advantage (Wheelwright and Clark, 1992). Our analysis shows that a

design strategy that takes advantage of implicit modules in a complex system, and mounts

multiple experiments on modules with high uncertainty and independence, can serve as the

cornerstone of a competitive strategy based on constant innovation and rapid product

improvement.

4.3 Variety

In addition to increasing speed, modularity greatly increases the potential variety of a

product line. Variety is most often measured by the number of versions of the product that are

available. If a product is divided into j modules, and ki variants (in addition to the existing

version) are developed for module i, then the total number of products in the line, denoted L, is:

L(j, k1...kj) = ∏ (ki + 1) ; (9)

i=1

This number explodes as the number of modules increases: for example five variants on each of

five modules results in 55 = 3125 products; five variants on each of ten modules obtains 510 =

9.8 million products.

These numbers are so staggering that they must be questioned. First, it is clear even

without a model that modularity is the only way to attain high variety at a reasonable cost. In

fact, modularity-in-production makes variety attainable in the context of high-speed, mass

production: that is why modularity is a key element in design-for-manufacturing and group

technology programs aimed at reducing production costs (Han Hitami and Yoshida, 1985;

Nevins and Whitney, 1989; Shirley, 1990; Sanderson, 1991). By the same token, modularity

creates economies of scale in product design: if the goal is to achieve the highest number of

different options at the lowest cost, then the combinatorial efficiencies of modularity-in-design

are indeed compelling.

However, simply counting the number of versions of a product obscures the fact that

sheer numbers can be increased by adding versions that are practically identical. Just as

significant as the impact of modularity on the number of products, is its effect on the range of

products that can be offered. Modeling the “degree of difference” among products is a

challenging task that goes beyond the scope of this paper. However, the model does provide

ways to begin to think about variety that are deeper than the sheer number of products.

Fundamentally, the purpose of variety is to allow for differences in consumer

preferences: different versions of the product can then come closer to each consumer’s ideal.

Differences in consumer taste are often modeled by locating groups of consumers in a subspace

of product attributes (Lancaster, 1966; Matutes and Regibeau, 1992). In this model each version

of the product is a point in an N-dimensional space of attributes. The design process itself

generates new points, representing new feasible versions of the product.

With a modular design, the problem is not how to generate new versions of the product,

for each experiment on a single module generates thousands or millions of “new” versions of the

product. The problem is to discover which versions are economically significant, that is, which

are “close” enough to a subset of consumers (and far enough away from competitors) to have

value. Thus management of product variety in the context of a modular design involves

collecting, collating, organizing and codifying knowledge about the range of feasible products

and the “location” of potential customers.38

5. The Costs of a Modular Design

Of course, the benefits of modularity -- performance, speed and variety -- are not free.

Modular designs require an investment in architecture and interfaces, and experiments are costly

to run. Most importantly, the experimental results and their modular combinations must be

evaluated to arrive at the final design. Thus in deciding how much modularity to pursue, or how

many experiments to run, the values in equation (6) or (7) need to be compared to the relevant

costs.

As a first approximation, it is reasonable to assume that interface design costs will be

roughly proportional to the number of modules, j, and that the cost of experimentation will be

proportional to the number of experiments, k. Define J as the design cost per module and K as

the cost per experiment, and T(j,k) as the cost of testing j modules and k experiments per

module. The cost C(j,k) of a modular, multi-experiment development program is then:

C(j,k) = kK - jJ - T(j,k) (10)

The cost of testing depends on whether tests are conducted at the module or the system level, and

is the main focus of the next section.

Although they appear symmetrically in equation (8), there is an important difference

between the cost of modularizing (J) and the cost of experiments (K). If a particular

modularization and the associated interface designs survive for more than one generation, the

38Investigating new product configurations and discovering new sets of customers is fundamentally different from

improving a design for an existing set of customers. Christensen and Rosenbloom (1994) argue that “being too close”

to existing customers led disk drive manufacturers to reject internal proposals to make smaller drives. Smaller drives

were then introduced by startup ventures, and sold to entirely new groups of customers.

partitions and interfaces do not have to be designed again. At that point J has the character of a

sunk cost: it disappears from value calculations from that time forward. Because their cost is

sunk, if the partition and interfaces designs are proprietary, they can serve as a barrier to entry to

the development process (Sutton, 1991; Ferguson and Morris, 1993). In contrast, experimental

costs (K) are incurred each time a different experiment is run, and thus K has the character of a

variable cost.

5.1 Testing

Our analysis shows that the impact of modularity-plus-experimentation on the value,

speed and range of product improvement can be substantial. However, these economic gains

will not materialize if the design team lacks adequate testing technology. The combinatorial

explosion that results when multiple experiments on many modules must all be tested can easily

overcome the benefits of mix-and-match flexibility.

Testing technologies have made giant strides over the last twenty years. One of the

greatest effects of computer-aided-design (CAD), manufacturing (CAM) and software

engineering (CASE) has been to reduce the cost of evaluating trial solutions at many points in

the design process. Where once it was necessary to construct a physical prototype in order to test

the behavior of a module or a system, emulators and simulators now provide virtual prototypes at

a fraction of the cost. Design changes that once required costly handwork or rewiring, can now

be coded in a few keystokes.39 This drop in the cost and time required for testing constitutes a

profound change in design technology, whose economic implications are only beginning to be

understood.40

39Bauer, Collar and Tang (1992) vividly describe the difference between debugging a physical prototype, and using a

sophisticated simulation package to test circuit design:

[I]n the past, the back of a debugged board would look like a plate of spaghetti, with 300 or more wires

strteching to and fro. But when we debugged the [computer]-tested board, there were only six -- just six --

wires on the back.... In the past, debugging a system in its entirety usually took us as long as a whole year.

But, thanks to [the simulation], we had a prototype ... up and running just six weeks after debugging began.

(pp. 105-106; Italics in original)

40Of course, the new technology requires investment (a sunk cost) in tools and models that represent the objects

being tested. Sanderson (1991) models the manufacturing efficiency gains from such investments but does not

consider their effect on experimentation.

Testing -- whether by hand or by computer -- can take place at the level of the whole

system or at the level of individual modules. In general, testing at the module level requires

greater depth of knowledge than testing at the system level.41 To predict whether a particular

module design will work without imbedding it in a whole system means that the module's

relationship to the rest of the system must be understood in some detail. Thus the development of

system-level tests usually precedes the development of module-level tests in the evolution of a

product.

5.1.1 System Tests

If all testing takes place at the system level, and the cost per test is c, then the overall cost

T(j,k) of testing j modules and k experiments per module is:

T(j,k) = [(k+1)j -1]c (11)

This number is proportional to the number of potential products (see equation 9), and increases

very rapidly as j, the number of modules, goes up. Therefore unless system-level test costs are

very low indeed (on the order of one-millionth the value of the benchmark program), the value-

maximizing design strategy is to pursue relatively unmodularized development with limited

amounts of experimentation.

Figure 4 graphs the net present value of the overall program (obtained by subtracting

equation 11 from equation 7) for two levels of testing cost. In the first instance, a test costs 10%

of the value of the benchmark program. In the second instance, a test costs 1% of the value of the

benchmark program. To complete the specifications, we assume that the costs of designing

41Depth of knowledge should not be confused with cost. Modules are (by definition) smaller and simpler than the

systems in which they function, and thus it is usually cheaper to construct and carry out module-level tests than

system-level tests. But to calibrate the performance of a module , it is necessary to know what the system-as-a-whole

is supposed to do, and to have some idea of how the module contributes to its overall function. And often an

imperfectly-understood module can be placed in a prototype system to “see if it runs.” Of course this approach entails

cycling, rework and debugging after the fact, all of which are extremely costly. But the costs can be avoided only

through greater knowledge about the system and components (Clark, 1985).

modules and running experiments are each equal to 5% of the value of the benchmark program: J

= K=.05 V(1,1).

Figure 4 shows what a heavy burden system-level testing imposes on modular designs. If

system-level test costs are on the order of 10% of V(1,1) as shown in the first panel, the option

value created by modularity is completely swamped by the combinatorial cost of testing. The

optimal configuration of the development program as a whole is to run nine experiments on the

Figure 4

13579

No. of Modules

No. of

Experiments

Value

Profit with System-Level Testing

Cost per test = .1 V(1,1)

Optimum:

1 Module

9 Experiments

13579

No. of Modules

No. of

Experiments

Value

Cost per Test = .01 V(1,1)

Optimum:

2 Modules

8 Experiments

unmodularized system. In fact, in this highly stylized example, for any level of testing cost

higher than 2% of V(1,1), the best economic solution is not to modularize at all. A company

might invest in modularity for other reasons, e.g., to obtain production economies of scale, but as

long as testing costs remained high and system testing was the norm, modularity would not make

the development process more efficient.

The picture changes only slightly when costs drop by an order of magnitude. As the

second panel shows, if the cost per test is 1% of V(1,1), a larger number of modules or

experiments becomes economically feasible. But when one combines modules and experiments

(moving out along the 45o line in the XY plane), the combinatorial explosion of testing costs

once again swamps the value of mix-and-match flexibility. Under our stylized assumptions, the

optimal configuration is to design two modules and run eight experiments.

5.1.2 Module Tests

The fact that testing costs foreclose a set of valuable options creates incentives to invest

in better testing technologies. However, as we have seen, simply lowering the cost per test even

by several orders of magnitude is not enough, for the number of combinations to be tested

increases exponentially with the number of modules. Another approach is to design tests so that

the best module can be selected "at its own level."

To evaluate a subsystem without embedding it in a working prototype requires detailed

knowledge about what the subsystem contributes to the whole, as well as how different

subsystems interact. Potential dysfunctional effects of the subsystem must also be understood,

so that potentially harmful interactions can be identified ahead of time. Thus module testing

demands functional knowledge, pre-set standards and ways of measuring a module’s

performance against the standards. Knowledge of how to test goes beyond the knowledge needed

to partition tasks, design interfaces or conduct experiments, although knowledge gleaned from

these other activities may contribute to the development of testing technology.

Testing is thus another dimension on which a company may usefully invest in knowledge

about a complex system. The benefits of such an investment are twofold. First is a reduction in

direct testing costs. If module testing replaces system testing, the overall cost T(j,k) falls

precipitously for higher levels of j and k. The cost of testing j modules and k experiments per

module becomes:

T (j, k) = j k c; (11)

where c again denotes the cost of one test. It is significant that in moving from system-level

testing to module-level testing, the number of modules, j, now enters as a multiplicative factor

and not as an exponent.

Second and even more important is the “unleashing” of option values implicit in a

modular, multi-experiment design strategy. What module-level testing can mean for management

of a development program is shown in Figure 5, which graphs net present values for the same

single-test costs as before: c =.1V(1,1) and .01 V(1,1).42 With “high cost” testing, the optimal

configuration is now to divide the design into 7 modules and run 5 experiments on each. But

truly dramatic changes occur when module testing is combined with low single-test costs. With

test costs on the order of 1% of the value of a program, the optimal program configuration -- 170

modules and 3 experiments per module -- is no longer on the chart. Moreover, the value of the

entire development effort increases substantially. Organizing ten experiments on each of ten

modules now creates ten times as much value as the benchmark unmodularized program. (In

contrast, the highest value achievable with only system-level testing was less than four times the

value of the benchmark program.)

5.1.3 Testing Costs and Product Evolution.

The dichotomy between system and module testing shown in the figures is too stark to

represent reality. In practice, designers are not restricted to only system or only module tests and

the optimal strategy is to use whatever testing tools and protocols are available. Firms that design

42For purposes of comparison, we are assuming that it costs the same amount to test a module as a system, despite the

fact that modules are less complex than systems. In reality, module tests may be cheaper than system tests.

complex systems must therefore invest in both types of testing technologies.

Figure 5

13579

No. of Modules

No. of

Experiments

Value

Profit with Module-Level Testing

Cost per Test = .1 V(1,1)

Optimum:

7 Modules

5 Experiments

13579

No. of Modules

No. of

Experiments

Value

Cost per Test = .01 V(1,1)

Optimum:

Off the Chart

At any point in time, however, the balance between system and module testing will be

determined by the company’s recent investments. Specifically, has it organized its development

process around better and cheaper system-level tests and invested in the latest tools of virtual

system modeling? Or, has it placed its bets on module testing and screening off-the-shelf

components? Research conducted in the automobile, computer workstation, and software

industries shows that companies differ substantially on these dimensions of investment

strategy.43 Some differences are dictated by technology and industry circumstances, but among

close competitors, different approaches to testing may arise because of different views about the

product’s likely evolution.

Large investments in system-level testing are appropriate when the product’s boundaries

(the activities that must take place “under one roof”) have stabilized, and the potential payoff to

further modularization is low. Essentially this means that the interdependencies among the

components of the product are both tight and not very well understood. In this case the only valid

test of a system is to “make” it, where “making” it may mean constructing a virtual image. The

ability to construct, test and revise the design rapidly and at low cost can then constitute a major

competitive advantage.

However, large expenditures on elaborate system-level testing tools may prove

unproductive if system modularity itelf is changing over time. Remodularizing the system

requires testing tools that make the definition and selection of components easier. Some of these

are similar to those used in system-level testing: for example, a large-scale virtual image of the

system is useful in screening off-the-shelf components as well as in testing inhouse designs. But

there is one critical difference: module-level testing rests on the principle of information hiding.

By definition, a module-level test does not require full knowledge of the structure of the object

being tested, but only knowledge of a subset of relevant parameters. Knowing what facts about

the underlying components do not need to be communicated to the larger system (and hence can

be left out of the system’s virtual image) is the true test of the designers’ understanding of how

43Clark and Fujimoto (1991); Iansiti (1994); Cusumano (1994).

the system really works.

Where the cost of testing is very low (as it is now in many industries), our distinction

between system and module-level testing may appear to be hair-splitting. But product designs

may evolve differently depending on which testing strategy is emphasized, and low test costs

tend to magnify the incentives to pursue different routes. Figure 6, which plots values for testing

costs on the order of one-millionth of the value of the benchmark program -- .000001 V(1,1),

illustrates how testing technology may influence the evolution of a product’s design over several

generations.

The upper panel of figure 6 shows the payoffs to modularity as perceived by a firm that is

focused on system-level tests. In effect, its testing technology sets limits on where its designs can

go: this is what is meant by the sharp dropoff in the function that occurs as more modules are

combined with more experiments. Even with extremely low costs per test, at some point the

combinatorial explosion takes hold and it becomes uneconomic to test all possible variants of the

system. As long as all industry participants perceive the design in the same way, the architecture

of the product will stabilize in this configuration.44 At this point, those companies that were first

to invest in system testing and other design capabilities for this configuration will enjoy a

relatively long-term competitive advantage.

But suppose some industry participants perceive that further modularization is possible

through investment in task partitioning, interfaces and, importantly, module-level tests. The

second panel suggests a very different evolutionary pattern. Here the testing technology does not

limit the amount of modularity that is “manageable,” and as a result, for any given level of

modularity, there are incentives to invest in still more modularity. This means that the design

may not stabilize very quickly nor very close to its current configuration. In these circumstances,

investments in inhouse system testing capabilities that do not envision further modularization

will be a risky and possibly misguided allocation of resources.

44This stabilization corresponds to the emergence of a dominant design. See Abernathy and Utterback (1978) and

Utterback (1994).

Figure 6

15913 17

No. of Modules

No. of

Experiments

Value

Profit with System-level Testing

Cost per test = .000001 V(1,1)

Optimum:

4 Modules

20 Experiments

913 17

No. of Modules

No. of

Experiments

Value

Profit with Module-Level Testing

Cost per test = .000001 V(1,1)

6. Modularity, Knowledge and Organization

Beyond its impact on value, speed or variety of an individual development program, the

pursuit of modularity-in-design changes the way in which an enterprise creates, transmits and

stores knowledge. Therefore, a modular approach to product design is both a strategic choice,

and a commitment to a different organizational structure and pattern of evolution.

To see how modularity affects an organization through its knowledge base, consider what

happens in a firm where design is not modular. Generally, a design team taking a “monolithic”

approach will not attempt to break down the product to see what principles govern its

performance. As a result, such teams achieve only a relatively superficial understanding of the

product and its production system. Moreover, within the product and the production system

nothing will be standard: everything will be unique and idiosyncratic to the particular design.

Monolithic products and production systems may be well-adapted to a particular set of external

conditions, but in general they will not support rapid change or high variety. Nevertheless, most

complex products and production systems begin with fairly monolithic designs, because to

decompose a task requires knowledge.

A modular design approach breaks down a product into discrete subsystems and

components, organizes these via an architecture, and links the individual modules using standard

interfaces. The goal of modularization is always to achieve functionality of the whole, but

independence across the modules and subsystems. To achieve this goal, the design team must

have three types of knowledge: (1) “architectural knowledge” of how components can be

assembled into a working system; (2) “interface knowledge” of how to connect modules and how

modules affect one another; and (3) “component knowledge” of how to design and build specific

components. These three types of knowledge must be “in place” prior to the modularization

itself. To the extent that there are gaps in architectural, interface or component knowledge, the

initial modularization may fall short: either the system-as-a-whole will not function, or, as is

more common, the modules will not be fully independent and significant effort will be needed to

integrate the system after the fact.

In addition, to achieve the economic benefits inherent in modularity-in-design, the

designers must have testing technology that permits modules to be selected “at their own level.”

Knowledge of what to test and how to test is different (and may be somewhat deeper) than

architectural, interface or component knowledge, but is just as critical to the goal of efficient

design improvement. Testing technology also needs to be in place early in the design process:

ideally test standards should be set at the time of the initial task partition so that subsystem

design groups will know what targets to shoot for.

Thus in very specific ways -- in terms of architecture, interfaces, components and tests --

a modular design rests on greater knowledge of how the product works than a monolithic

design. Attempts to modularize beyond the boundaries of existing knowledge (ie, splitting

components whose functional interdependencies are not well understood) will lead to costly

iterations when the time comes to integrate the modules into a functioning system. Therefore, at

the outset of a development effort, a key managerial question is: how much modularity will the

enterprise’s knowledge base support? To make an informed judgment on this issue, managers

must understand the architectural and technological tradeoffs that underlie the basic design, and

appreciate the potential benefits and costs of achieving a given degree of modularity.

Knowledge is prerequisite for implementing a modular design approach, but further

knowledge is gained through the process. First, designers will be asked to partition their

information, with each group focusing on one specific component of the whole system. At the

same time, the designers must agree among themselves on the specific architecture, interfaces

and test procedures that will govern the system. Co-ordinating the task partition and overseeing

what can be complex negotiations over interfaces or tests, consumes time and effort early in the

process. But the result is a design that makes tacit knowledge explicit, and that provides for the

acquisition of more knowledge later on. Knowledge acquired late in the process can be

accommodated into the design precisely because it has been “hidden” within specific modules.45

Of course, a design team pursuing a non-modular strategy must also create some

breakdown of the effort into specific tasks. However, the motivation of task partitions in such

45Indeed a modular design can accomodate new knowledge (involving hidden structures) even after the end of the

development process. This is the principle that underlies field upgradability. Modular upgradability tends to blur the

time boundaries of the development process (Iansiti, 1994; Cusumano, 1994). It can also be of great value to users,

but, as with any kind of modularity-in-use, may stimulate entry and competition.

cases is merely to establish a workable division of labor, and thus assignments may follow a

simple “equalize the workload” rule or some other natural division (e.g., by number of parts). In

contrast to an ad hoc partition, a modular partition creates a framework that illuminates the

product’s internal functions and clarifies the relationship of the parts to the whole. Because it

corresponds to the way the product works, a modular partition is not only a useful way to divide

tasks, but is also a useful way to organize and store knowledge obtained in the design process

itself.46 This new knowledge in turn makes possible further refinements (and perhaps higher

degrees of modularity) in subsequent rounds of development.

Greater depth of knowledge will make possible improved system design, better module

performance, expanded variety and more rapid technical change, over and above what is

attainable through pure mix-and-match flexibility. Thus if a firm uses modularity not just as a

one-time design approach, but as template for its evolving knowledge base, modularity will not

be neutral (as was assumed) but will in fact result in better designs in later rounds of

development.

The link between modularity and the firm’s knowledge base highlights the connection

between a modular design strategy and the organization of the firm itself. The way a firm defines

subsystems and modules within a product not only sets up targets of opportunity within a design

program, but also determines a pattern of focus and specialization for the company’s human

resources. Moreover, as Henderson and Clark (1990) have argued, the interfaces that govern the

relationship between modules have counterparts in the organizational routines and processes that

link the people engaged in subsystem and module development work to one another. Thus the

modular structure of the product comes to be reflected in the organizational structure of the firm

making it.

But there is a risk. When modularization occurs as a result of “natural” product evolution

(the process described in section 2), much of the framework may be tacit and embedded in

organizational routines (Nelson and Winter, 1982). There will then be a tendency to “see” and

46Recall that the design library described in Shirley (1990) was organized by module.

“store” new information about components (including ways to test them) in the specific

organizational units charged with responsibility for those components. In contrast, the

architecture and interfaces may not be the responsibility of any particular person or group. In

this case there will be no natural collection point or repository for information about how to

combine existing components in new ways or how to streamline interactions among components.

Therefore unless managers take explicit steps to define and preserve knowledge about

architectures and interfaces, such knowledge can easily “fall through the cracks” of the

organization.47

Its connections with knowledge and organizational structure give modularity the

character of an organizational capability. It is a capability with great potential, but it requires

investment and is difficult to execute. Indeed, without major investment of time, resources and

creativity in testing technology, and without skill and ingenuity in its application, the power of

modularity will be held captive to the tyranny of combinatorial testing costs. However, the firm

that understands how to create effective modules, how to set up effective interfaces, how to

design clever parallel experiments, how to do module level testing, and how to do low cost

system level testing, may reap substantial rewards. But, if it is to be successful, modularity must

become deeply embedded in what the firm knows, how it is organized, and how it gets work

done.

7. Conclusion

In this paper we have tried to build a framework for understanding modularity as a

strategy of design. In the technical realm, modularity has been implemented in many industry

settings, and, accordingly, it has been the focus of much discussion and analysis in the

engineering literature. But our understanding of the economic incentives that lie behind

modularity, and its consequences in different competitve settings is limited. In developing our

47The failure of otherwise innovative firms to recognize the potential benefits of certain “architectural” product

reconfigurations is a persistent theme in the literature on innovation (Tushman and Anderson, 1986; Henderson and

Clark, 1990; Christensen, 1994; Langlois, 1994).

framework we have tried to put modularity-in-design in context by comparing it to modularity-

in-production and modualrity-in-use. In addition, we constructed a model that captures the

option value inherent in modularity, as well as the costs of pursuing this approach. Although the

model is relatively simple and stylized, it does suggest some important connections between

modularity-in-design and the management and organization of the firm. From an academic

perspective, it begins the process of understanding the economic reasons to pursue modularity-

in-design, and its implications for practice and for future research.

From a managerial perspective, our analysis of modularity takes choices of a technical

nature - product structure, interfaces, testing technology, and associated investments - and puts

them squarely on the agenda of a firm’s senior management. It shows how design decisions,

many of which may be buried deep within the organization, can have a major impact on value

created for consumers and on a firm’s ability to compete. It also shows how seemingly unrelated

technical decisions -- task assignments and testing technologies, for example -- are intimately

connected in the way they affect value, speed and variety. Because they are inter-related and have

a large potential impact on value and competitiveness, such issues ought not to be decided in

isolation, nor by rote application of precedent. Our analysis provides a way to frame these

decisions in business terms, and highlights the central issues to be considered in resolving them.

REFERENCES

Abernathy, William J. and James M. Utterback (1978) “Patterns of Industrial Innovation,”

Technology Review, 80(7):40-47.

Alexander, Christopher (1964) Notes on the Synthesis of Form, Harvard University Press,

Cambridge, MA.

Baldwin, Carliss Y. and Kim B. Clark (1992) “Capabilities and Capital Investment: New

Perspectives on Capital Budgeting,” Journal of Applied Corporate Finance, 5(2):67-82.

Bauer, Roy A., Emilio Collar and Victor Tang (1992) The Silverlake Project, Oxford University

Press, New York, NY.

Black, Fischer and Myron Scholes (1973), "The Pricing of Options and Corporate Liabilities,"

Journal of Political Economy, 81: 637-654.

Brooks, Frederick W. (1975) The Mythical Man Month, Addison Wesley, New York, NY.

Christensen, Clayton M. (1994) “The Rigid Disk Drive Industry: A History of Commercial and

Economic Turbulence,” Business History Review, 67(Winter):531-588.

Christensen, Clayton M. (1994) “Industry Maturity and the Vanishing Rationale for Industrial

R&D,” Harvard Business School Working Paper 94-059, Boston, MA, (March).

Christensen, Clayton M. and Richard S. Rosenbloom (1994) “Explaining the Attacker’s

Advantage: Technological Paradigms, Organizational Dynamics and the Value Network,”

Research Policy 22(forthcoming).

Clark, Kim B. (1985), "The Interaction of Design Hierarchies and Market Concepts in

Technological Evolution," Research Policy, 14, 5: 235-251.

Clark, Kim B. (1988), "Knowledge, Problem Solving and Innovation in the Evolutionary Firm,"

Harvard Business School Working Paper.

Clark, Kim B. and Takahiro Fujimoto (1991), Product Development Performance: Strategy,

Organization and Management in the World Auto Industry, Harvard Business School

Press, Boston, MA.

Cohen, Morris A., Jehosua Eliasburg and Teck H. Ho, “New Product Development: Performance,

Timing and the Marketing-Manufacturing Interface,” Working Paper 92-04-01, Wharton

School, University of Pennsylvania, Rev. April 1993.

Cusumano, Michael A. (1991) Japan’s Software Factories: A Challenge to U.S. Management,

Oxford University Press, Oxford, UK.

Cusumano, Michael A. and Stanley a. Smith (1994) “Beyond the Waterfall: A Comparison of

“Classic” versus PC Software Development, mimeo, Massachusetts Institute of

Technology (July).

Dixit, Avinash K. and Robert S. Pindyck, (1994) Investment Under Uncertainty, Princeton

University Press, Princeton, NJ.

Dorfman, Nancy S. (1986) Innovation and Market Structure: Lessons from the Computer and

Semiconductor Industries, Ballinger, Cambridge, MA.

Economides, Nicholas and Steven C. Salop (1992) “Competition and Integration among

Complements and Network Market Structure,” Journal of Industrial Economics,

40(1):105-123.

Eppinger, Steven D. (1991), "Model-based Approaches to Managing Concurrent Engineering,”

Journal of Engineering Design, 2 (4).

Eppinger, Steven D. and Kent R. McCord (1993) “Managing the Iteration Problem in Concurrent

Engineering,” MIT Working Paper 3594-93-MSA, August.

Eppinger, Steven D., Daniel E. Whitney, Robert P. Smith and David Gebala (1994) “A Model-

Based Method for Organizing Tasks in Product Development,” Research in Engineering

Design, forthcoming.

Farrell, Joseph, Hunter K. Monroe and Garth Saloner (1993) “Order Statistics, Interface

Standards and Open Systems,” mimeo, University of California, Berkeley, CA.

Ferguson, Charles H. and Charles R. Morris (1993) Computer Wars: How the West Can Win in a

Post-IBM World, Times Books, New York, NY.

Freeman, Christopher (1982) The Economics of Innovation, 2nd Ed. MIT Press, Cambridge, MA.

Foster, Richard (1986) Innovation: The Attackers Advantage, Summit Books, New York, NY.

Garud, Raghu and Arun Kumaraswamy (1993) “Changing Competitive Dynamics in Network

Industries: An Exploration of Sun Microsystems’ Open Systems Strategy,” Strategic

Management Journal, 14:351-369.

Gomes-Casseres, Benjamin (1994) Collective Competition: International Alliances in High

Technology, Harvard University Press, Cambridge, MA.

Hall, Bronwyn H. (1990) “The Impact of Corporate Restructuring on Industrial Research and

Development,” Brookings Papers on Economic Activity, 1990(1):85-133.

Han, I., K. Hitami, and T. Yoshida (1985), Group Technology Applications to Product

Management, Kluwer-Nijhoff Publishers, Boston, MA.

Harrison, J.M. (1985) Brownian Motion and Stochastic Flow Systems, John Wiley and Sons, New

York.

Henderson, Rebecca M. and Kim B. Clark (1990), "Generational Innovation: The

Reconfiguration of Existing Systems and the Failure of Established Firms," Administrative

Sciences Quarterly 35: 9-30.

Iansiti, Marco (1994) “Technology Integration: Managing Technological Evolution in a Complex

Environment,” Research Policy (forthcoming).

Iansiti, Marco (1994) “Shooting the Rapids: System Focused Product Development in the

Computer and Multimedia Environment,” mimeo, Harvard Business School, Boston, MA

(July).

Jensen, Michael C. and William H. Meckling (1992) “Specific and General Knowledge, and

Organizational Structure, Contract Economics, Lars Werin and Hans Wijkander, eds.,

Basil Blackwell, Ltd., Oxford, UK, pp. 251-274.

Jensen, Michael C. (1993) “The Modern Industrial Revolution, Exit and the Failure of Internal

Control Systems,” Journal of Finance, 48(3):831-880.

Kernighan, Brian and P.J. Plauger (1976) Software Tools, Addison-Wesley, Reading, MA.

Korson, Timothy D. (1986) “An Empirical Study of the Effect of Modularity on Program

Modifiability,” unpublished PhD dissertation, Georgia State University, Atlanta, GA.

Lancaster, K.J. (1966), "A New Approach to Consumer Theory," Journal of Political Economy,

74: 132-157

Langlois, Richard N. (1992a) “External Economies and Economic Progress: The Case of the

Microcomputer Industry,” Business History Review, 66:---.

Langlois, Richard N. (1992b) “Capabilities and Vertical Disintegration in Process Technology:

The Case of Semiconductor Fabrication Equipment,” mimeo, Univeristy of Connecticut,

Storrs, CT (November).

Langlois, Richard N. (1994) “Cognition and Capabilities: Opportunities Seized and Missed In the

History of the Computer Industry,” mimeo, University of Connecticut, Storrs, CT (March)

Langlois, Richard N. and Paul L. Robertson (1992) "Networks and Innovation in a Modular

System: Lessons from the Microcomputer and Stereo Component Industries," Research

Policy, 21: 297-313.

Lindgren, B.W. (1968), Statistical Theory, MacMillan Publishing Co., New York, NY.

Matutes, Carmen and Pierre Regibeau (1992) “Compatibility and Bundling of Complementary

Goods in a Duopoly,” Journal of Industrial Economics, 40(1):37-54.

Marples, David L. (1961), "The Decisions of Engineering Design," IRE Transactions on

Engineering Management 2: 55-71.

Mason, Scott P. and Robert Merton (1985) “The Role of Contingent Claims Analysis in

Corporate Finance,” Recent Advances in Corporate Finance, E. Altman and M.

Subrahmanyam, Richard D. Irwin, Homewood IL.

Merton, Robert C. (1973), "Theory of Rational Option Pricing," Bell Journal of Economics and

Management Science, 4: 141-183

Merton, Robert C. (1990), Continuous Time Finance, Baoil Blackwell, Cambridge, MA.

Milgrom, Paul and John Roberts (1990), "The Economics of Manufacturing: Technology,

Strategy and Organization," American Economic Review, 80(3): 511-528.

Myers, Glenford J. (1975) Reliable Software through Composite Design, Van Nostrand Reinhold

Company, New York.

Nelson, Richard R., and Sidney G. Winter (1982), An Evolutionary Theory of Economic Change,

Harvard University Press, Cambridge, MA.

Nevins, James L. and Daniel E. Whitney (1989) Concurrent Design of Products and Processes,

McGraw-Hill Publishing Company, New York, NY.

Nichols, Nancy A. “Scientific Management at Merck: An Interview with CFO Judy Lewent,”

Harvard Business Review, 72(1):88-99 January-February 1994.

Parnas, D.L. (1972a) “A Technique for Software Module Specification with Examples,”

Communications of the ACM, 15:330-336, May.

Parnas, D.L. (1972b) “On the Criteria to Be Used in Decomposing Systems into Modules,”

Communications of the ACM, 15:1053-1058, December.

Parnas, D.L., P.C. Clements and D.M.Weiss (1985) “The Modular Structure of Complex

Systems,” IEEE Transactions on Software Engineering, SE-11:259-266, March.

Pindyck, Robert S. (1991), "Irreversibility, Uncertainty and Investment," Journal of Economic

Literature, 29: 1110-1152.

Pugh, Emerson W., Lyle R. Johnson and John H. Palmer (1991) IBM’s 360 and Early 370

Systems, MIT Press, Cambridge, MA.

Quinn, James Brian (1985) “Managing Innovation: Controlled Chaos,” Harvard Business Review,

63(3):73-84.

Rinderle, J.R. and N.P. Suh, (1982) “Measures of Functional Coupling in Design,” ASME Journal

of Engineering for Industry, 104:383-388, November.

Rothwell, Roy and Paul Gardiner (1988) Journal of Marketing Management, 3(3):372-387.

Sanderson, Susan Walsh, (1991) “Cost Models for Evaluating Virtual Design Strategies in

Multicycle Product Families,” Journal of Engineering and Technology Management, 8:

339-358.

Sanchez, Ronald A. (1991), "Strategic Flexibility, Real Options and Product-based Strategy,

unpublished Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge, MA.

Schach, Stephen R. (1993) Software Engineering, 2nd Ed., Irwin, Burr Ridge, IL.

Simon, Herbert A. (1969)The Sciences of the Artificial, MIT Press, Cambridge, MA.

Shirley, Gordon V. (1990), "Models for Managing the Redesign and Manufacture of Product

Sets," Journal of Manufacturing and Operations Management, 3: 85-104.

Stevens, W.P., G.J. Myers and Constantine, L.L. (1974) “Structured Design,” IBM Systems

Journal, 13(2):115-139.

Stuart, Toby and David Collis, “Cooper Industries’ Corporate Strategy,” Harvard Business School

Publishing Division, Boston, MA.

Stulz, Rene M. (1982), "Options on the Minimum or the Maximum of Two Risky Assets,"

Journal of Financial Economics, 10: 161-185.

Suh, Nam P. (1990) The Principles of Design, Oxford University Press, Oxford, England.

Sutton, John (1991) Sunk Cost and Market Structure: Price Competition, Advertising and the

Evolution of Concentration, MIT Press, Cambridge, MA.

Suzue, Toshio and Akira Kohdate (1990) Variety Reduction Program: A Production Strategy for

Product Diversification, Productivity Press, Cambridge, MA.

Tushman and Anderson, “Technological Discontinuities and Organizational Environments,”

Administrative Sciences Quarterly 31:439-465, 1986.

Trajentenberg, Manuel (1990) Economic Analysis of Product Innovation: The Case of CT

Scanners, Harvard University Press, Cambridge, MA.

Ulrich, Karl T. (1992), "The Role of Product Architecture in the Manufacturing Firm," Working

Paper No. 3483-92-MSA, Sloan School of Management, MIT, Cambridge, MA.

Ulrich, Karl T. and Steven D.Eppinger (1994) Methodologies for Product Design and

Development, McGraw-Hill, New York, NY.

Utterback James (1994) Mastering the Dynamics of Innovation: How Companies Can Seize

Opportunities in the Face of Technological Change, Harvard Business School Press,

Boston, MA.

Von Hippel, Eric (1990), "Task Partitioning: An Innovation Process Variable," Research Policy,

19: 407-418

Wruck, Karen H. and Michael C. Jensen (1993) “Science, Specific Knowledge and Total Quality

Management,” mimeo, Harvard Business School (July).

Wheelwright and Clark (1992), Revolutionizing Product Development, Free Press, New York,

NY.

Modular Architectures in the Marketing Process

Article

Oct 1999
J MARKETING

Ron Sanchez

Modular product architectures are being adopted as the basis for new product designs in a growing number of markets. In this article, the author explains how adoption of modular product architectures makes possible new kinds of processes for creating and realizing products, as well as new approaches to identifying and managing organizational knowledge used in product creation and realization. When managed strategically, modular product, process, and knowledge architectures enable firms to create greater product variety, introduce technologically improved products more rapidly, bring new products to market more quickly, and lower costs of product creation and realization. This article investigates the ways in which the strategic use of modular architectures can affect the marketing process. It explains ways in which modular architectures change the technologically determined possibilities for product creation and realization in the marketing process, create new market dynamics, and lead to new objectives and methods for the marketing process.

Transformation of the Acoustic Amplifier into a Modular Structure

Article

Nov 2023

The modular concept of device construction is most often based on the results of economic analyses. It allows lowering both manufacturing and design costs. It also improves their operating characteristics, in particular their repairability. It also happens that the modular structure of a consumer product is used to strengthen its market position. The authors of the article faced such a situation. The management of a company producing audiophile equipment decided to offer its potential customers an acoustic amplifier in which the user would be able to replace the preamplifiers himself, depending on the requirements related to the input signal. This type of modularity is known as "component swapping." Before commencing the development of the structure, detailed analyses of the benefits and potential risks related to the individual stages of the product's life cycle were carried out: design, production and operation. On this basis, the "Divaldi" company, the manufacturer of the equipment, developed assumptions regarding the designed amplifier. In particular, it was considered that the modular system would be built by modifying the existing, successful design of the INT-02 amplifier. The modification will enable easy replacement of the two preamplifier sections. Users will have a choice of classic preamplifiers: line and phono, as well as modules integrated with an analogue-to-digital converter and a module with a Bluetooth receiver. A low-volume production was assumed from previously manufactured and stored components: the base and typical preamplifier sections. As for the exploitation stage, the possibility of self-replacement of modules by the user has been adopted. Based on these assumptions, the concept of an amplifier with two identical pockets was proposed, enabling the simultaneous installation and use of two preamplifier sect ions. Detailed requirements for interfaces between the modules and the base unit were formulated: mechanical and electrical. A review of the available types of signal connectors led to the selection of the PCIe connector as the one that best meets the technical and economic criteria. From the mechanical side, it was proposed to use sliding guides, guaranteeing proper positioning of the modules in the pockets of the base. After selecting and approving the solutions, constructions of the mechanical components of the system were developed, creating a new standard: frame, module pockets and preamplifier module. The successful implementation of the amplifier's modularity became the basis for formulating proposals for further modifications of the design to meet the growing requirements of users who expect more and more possibilities to adapt the structure and interfaces of devices to their own needs. Keywords: modularity, acoustic amplifier, customization

Modularisation as a competitive criterion in industries manufacturing machinery and equipment in Brazil

Article

Full-text available

May 2020

This research aimed at identifying the competitive criteria that, together with modularisation, impacted the differentiation strategy in industries manufacturing machinery and equipment in Brazil. This was survey-type research that used structural equation modelling (SEM-PLS) for data analysis. A total of 236 responses were obtained from participants in Brazilian machinery and equipment companies in the area of R&D, including directors, managers, coordinators, and supervisors. From this sample, the results show that applied modularisation positively impacted the competitive criteria related to innovation, quality, and flexibility, whereas, when applied to the delivery costs and performance criteria, it did not contribute to a competitive differential. Since it deals with a non-probabilistic sample by convenience, the present model does not represent the relationship between the constructs in a definitive way, for it concerns a model that address a sample of the machinery and equipment manufacturing industry, referring to ISIC D28, from Brazil.

Component suppliers in the commodity battle: Can digital technology in multi-tier supply chains help to transform liabilities into opportunities?

Article

Full-text available

Feb 2021

Tobias Herbst

This conceptual study analyzes how digital technologies can create opportunities for component suppliers in multi-tier supply chains facing the issue of commoditization and it develops a strategic framework based on prior studies. This study finds that modularity facilitates the understanding of the dynamics of commoditization and knowledge sharing. Furthermore, modularization drives the asymmetric distribution of information in a multi-tier supply chain. Finally, digital technologies can offer strategic opportunities for targeting different stages of the value chain. Component suppliers should evaluate the criticality of products in new business models and advanced services to reduce the complexity of the landscape of strategic options and, thus, reduce wrong decisions. Managers should encourage general information sharing and open data exchange in a multi-tier supply chain. They should focus on vertical and horizontal partnerships and collaborations to foster the benefits of digital technologies. Conclusively, component suppliers should cooperate to build a counterweight to original equipment manufacturers in the vertical partnership. This study is the first to integrate the benefits of commodities and digital technologies in a strategic framework. It contributes to the literature on competition in the background of commoditization by outlining the concept of information asymmetry within a multi-tier supply chain driven by modularity. Thus, it advances the literature on digital technology and commoditization.

The Institutional Social Role for Maritim Village’s Food Stability

Article

Full-text available

Mar 2023

This study aims to explore social networks and interactions between networks in strengthening food stability in maritime villages in coastal areas. The geographical condition of the region has become one of the triggers for food instability and food security problems in coastal areas. This study uses the method of Social Network Analysis (SNA). Data processing uses R to visualize the role of stakeholder centrality. The analysis used is the analysis of baseline network, degree, and betweenness centrality. The value of modularity is an indicator of increasing or decreasing community networks. The number of nodes and links identifies the actors in the network and the interactions between actors. The results showed that social networks have an essential role in food stability. The private sector is a key actor in food stability in maritime villages. Is indicated by the modularity value of 0.43 (>0), which connects to 30 social network nodes and 62 links to food stability in maritime villages in the coastal area of Bintan Regency.

TASARIM ÜRETİM VE KULLANIM SÜRECİNDE MODÜLER DONATI TASARIMI ÜZERİNE BİR İNCELEME

Conference Paper

Full-text available

Oct 2020

ÖZET Geçmişten günümüze teknolojinin gelişimi, tüketimin artması ve üretimin çeşitlenmesi ile toplumun istek ve ihtiyaçlarında değişiklikler meydana gelmiştir. Bu doğrultuda gereklilikleri karşılamak amacıyla yeni teknikler ve malzemeler kullanılarak yapım sistemleri geliştirilmektedir. Yapım sistemleri üretim ve kullanım aşamalarını etkilediğinden zamana karşı bir esneklik ve dönüşüm içerisindedir. İhtiyaca bağlı dönüşümler, gerekliliklere cevap verir nitelikte olan ürünün bileşenleri arasındaki koordinasyonu sağlaması, birim olarak standartlaşmayı kolaylaştırması bakımından ‘modüler sistemler’ büyük önem arz etmektedir. Sistem belirlenen temel bir modülün üretilmesi ya da kullanılması ile başlamaktadır. Bu modülün çoğaltılması, azaltılması; farklı şekillerde birleşimleri ya da ayrıştırılması ile oluşturulan donatı sistemleri kullanıcı-mekân ihtiyaçlarına en hızlı şekilde cevap vermektedir. Araştırmada amaçlanan modüler donatı tasarımlarının çok yönlü kazançlarını tasarımcıların ve mekân kullanıcılarının bilgilerine sunmaktır. Çalışmada konu üzerine literatür taraması yapılarak modüler donatı sistemleri hakkında kavramsal çerçevede tanımlamalara yer verilmiş, modüler sistemin aşamaları incelenmiş, avantajları, dezavantajları ve önemi sunulmuştur. Ayrıca araştırma yöntemleri dikkate alınarak, yapılmış olan donatı örnekleri üzerinde üretimde modülerlik, tasarımda modülerlik ve kullanımda modülerlik aşamaları analiz edilmiştir. Örneklem grubu üzerinde yapılan analizler tablo halinde verilmiştir. Araştırma sonucunda modüler donatıların kullanıldıkları hacimde maksimum fayda ve fonksiyonellik sağladıkları, kullanıcı ihtiyaçlarına cevap verdiği, karmaşıklığı ortadan kaldırdığı görülmektedir. Ayrıca mekân sınırlarına göre özelleştirilmiş ürün oluşumunda etkili olduğu, mekânın kontrol edilebilirliğini arttırdığı, sonucuna ulaşılmıştır. Ürün bileşeninin tasarım geliştirme sürecindeki hızı ile tasarımcıya; üretimi, montajı, taşımacılığı ve maliyeti gibi konularda üreticiye kolaylık sağladığı görülmektedir. Akademik bir perspektiften bakıldığında, tasarımda ve üretimde modülerlik devamında kullanımda modülerliğe de olanak sağlayarak kullanıcıda tercih sebebini arttıracaktır. Bu nedenle çalışma sonuçları dikkate alındığında yapım sistemleri arasında teşvik edilmesi gereken bir yöntem olarak önerilmektedir.

MASS CUSTOMIZATION IN SERVICE OPERATIONS

Article

Sep 2010

Wang An

The concept of Mass customization (MC) is the ability to provide individually designed products and services to each customer through high process flexibility and integration. In recent years, MC has been identified as a competitive strategy; however, the investigations of MC in the service context are limited. In this dissertation, the published classification of MC approaches have been discussed and tested through cases which are from four different service industries. The limitations of these methods and the main characteristics which they have not covered have been highlighted. The analysis of the findings leads to the development of a service operations process framework, which comprises five fundamental service MC modes based on the framework of MC modes for manufacturing developed by MacCarthy et al. (2003). The framework and models have been extended through integrating modularity approach in order to improve the ability of full utilization of operations resource in the service industries. Some feasible approaches related to the application of modularity approaches have been suggested for the enterprises, who adopt mass customization strategy, in the service industry.

Dynamic capabilities of Module Suppliers

Conference Paper

Full-text available

Jul 2017

This paper seeks to determine the nature and scope of dynamic capabilities associated with the provision of modular supply in the automotive sector. Whilst the concept of dynamic capabilities is now widely understood, there has been very little research exploring such capabilities within the context of automotive cockpit provision. We contend that the development of dynamic capabilities within this context is contingent upon the abilities of module suppliers to identify and operationalize capabilities reflecting the unique attributes associated with this highly customised and bespoke module.

DİNAMİK YETENEKLERİN PAZAR VE MÜŞTERİ PERFORMANSINA ETKİSİ: HİZMET SEKTÖRÜNDE NİCEL BİR ARAŞTIRMA

Article

Full-text available

Jul 2019

Günümüz dünyasında firmalar rakiplerinden daha fazla müşteriye hizmet vermek ve rakiplerinden daha fazla pazar payı elde edebilmek amacıyla sürekli rekabet hali içerisindedirler. Firmaların rekabet sürecinde kalıcı avantajlar elde edebilmesi rakiplerine nazaran daha fazla yeteneğe sahip olmasını gerektirmektedir. Önemli olan firmaların sahip oldukları yeteneklerini pazarın dinamizmine göre yeniden yapılandırabilmesi ve pazarın koşullarına duyarlı hale getirebilmesidir. Bu süreçte firmaların ve yöneticilerin yeteneklerinin farkına varabilmeleri ve sürdürülebilir rekabet avantajı elde edebilmenin kaynağına ulaşabilmeleri gereklidir. Çalışmada; öncelikli olarak, organizasyonların sahip oldukları yeteneklerin dinamik mi yoksa operasyonel mi olarak değerlendirilmesi gerektiğinin ayrımı literatür desteğiyle ortaya konulmuştur. Bu bağlamda; dinamik yetenekler görüşünün firmaların pazar ve müşteri performanslarına nasıl katkıda bulunduklarını belirleyerek, hizmet sektöründe ağırlıklı rolü olan lojistik firmaları üzerinde yapılan nicel bir araştırma ile ortaya koymaktır. Araştırmanın bağımsız değişkenini dinamik yetenekler, bağımlı değişkenini müşteri ve pazar performansı oluşturmaktadır. Araştırma modelinde, değişkenler arasında bağlantı kurularak lojistik sektörü üzerindeki etkisi nicel bir çalışma ile ortaya konulmaya çalışılmıştır. Çalışmada, anket yöntemi kullanılmıştır. Tesadüfi örnekleme yöntemi ile Marmara bölgesinde İstanbul ilinde yer alan ve lojistik sektöründe yönetici olarak çalışan 122 yöneticiye anket uygulanarak veriler elde edilmiştir. Veriler, SPSS istatistik paket programı kullanılarak analiz edilmiştir. Yapılan analizler sonucunda, dinamik yeteneklerin müşteri ve pazar performansı üzerinde anlamlı bir etkisinin olduğunu göstermektedir.

Understanding Architecture Principles as Working Mechanisms: 8th Enterprise Engineering Working Conference, EEWC 2018, Luxembourg, Luxembourg, May 28 – June 1, 2018, Proceedings

Chapter

Full-text available

Jan 2019

Mark Paauwe

This paper is the introduction of a new way of formulating architecture principles as working mechanisms by enterprise architects. This alternative way of formulating is aimed at producing more effective architecture principles than enterprise architects are currently producing. This new way of formulating architecture principles aligns with practices in physics and building architecture. Architecture principles need to be made much more effective because in them lies the added value of working with enterprise architecture: using principles to guide the design and realization of enterprise-wide solutions compliant to strategy and stakeholders requirements. In this paper, principles are proposed to be formulated as working mechanisms rather than general rules (guidelines). This will make them communicate contextual truths and will have them cause more effect in guiding the design and engineering of systems. This paper presents intermediate results of an ongoing research on architecture principles.

The Theory of Rational Option Pricing

Article

Full-text available

Mar 1973
Bell J Econ Manag Sci

Robert C. Merton

Author begins by deducing a set of restrictions on option pricing formulas from the assumption that investors prefer more to less. These restrictions are necessary conditions for a formula to be consistent with a rational pricing theory. Attention is given to the problems created when dividends are paid on the underlying common stock and when the terms of the option contract can be changed explicitly by a change in exercise price or implicitly by a shift in the investment or capital structure policy of the firm. Since the deduced restrictions are not sufficient to uniquely determine an option pricing formula, additional assumptions are introduced to examine and extend the seminal Black-Scholes theory of option pricing. Explicit formulas for pricing both call and put options as well as for warrants and the new ″down-and-out″ option are derived. Other results.

IBM's 360 and Early 370 Systems

Book

Jan 2003

IBM's 360 and Early 370 Systems describes the creation of this remarkable system and the developments it spawned, including its successor, System/370. No new product offering has had greater impact on the computer industry than the IBM System/360. IBM's 360 and Early 370 Systems describes the creation of this remarkable system and the developments it spawned, including its successor, System/370. The authors tell how System/360's widely-copied architecture came into being and how IBM failed in an effort to replace it ten years later with a bold development effort called FS, the Future System. Along the way they detail the development of many computer innovations still in use, among them semiconductor memories, the cache, floppy disks, and Winchester disk files. They conclude by looking at issues involved in managing research and development and striving for product leadership. While numerous anecdotal and fragmentary accounts of System/360 and System/370 development exist, this is the first comprehensive account, a result of research into IBM records, published reports, and interviews with over a hundred participants. Covering the period from about 1960 to 1975, it highlights such important topics as the gamble on hybrid circuits, conception and achievement of a unified product line, memory and storage developments, software support, unique problems at the high end of the line, monolithic integrated circuit developments, and the trend toward terminal-oriented systems. System/360 was developed during the transition from discrete transistors to integrated circuits at the crucial time when the major source of IBM's revenue was changed from punched-card equipment to electronic computer systems. As the authors point out, the key to the system's success was compatibility among its many models. So important was this to customers that System/370 and its successors have remained compatible with System/360. Many companies in fact chose to develop and market their own 360-370 compatible systems. System/360 also spawned an entire industry dedicated to making plug-compatible products for attachment to it. The authors, all affiliated with IBM Research, are coauthors of IBM's Early Computers, a critically acclaimed technical history covering the period before 1960.

Science, Specific Knowledge and Total Quality Management

Article

Jan 1994

The Mythical Man-Month

Book

Jan 1975

Fred Brooks

Sunk Costs and Market Structure: Price Competition, Advertising, and the Evolution of Concentration

Article

May 1992

Sciences of The Artificial

Book

Jan 1996

Herbert A Simon

Brownian Motion and Stochastic Flow Systems.

Article

Jan 1987

Japan's Software Factories: A Challenge to U.S. Management.

Article

Feb 1992

Though Japan has successfully competed with US computer makers in manufacturing and marketing hardware, it has been less successful in developing software that competes with US software companies. This book analyses how a number of Japanese firms, in an effort to catch up, have created what are called 'software factories' in which large numbers of people are engaged in developing software in co-operative ways.

The Modular Structure of Complex Systems

Article

Mar 1985

This paper discusses the organization of software that is inherently complex because of very many arbitrary details that must be precisely right for the software to be correct. We show how the software design technique known as information hiding, or abstraction, can be supplemented by a hierarchically structured document, which we call a module guide. The guide is intended to allow both designers and maintainers to identify easily the parts of the software that they must understand, without reading irrelevant details about other parts of the software. The paper includes an extract from a software module guide to illustrate our proposals.

Structured Design

Article

Jan 1974
IBM SYST J

The HIPO Hierarchy chart is being used as an aid during general systems design. The considerations and techniques presented here are useful for evaluating alternatives for thos portions of the system that will be programmed on a computer. The charting technique used here depicts more details about the interfaces than the HIPO Hierarchy chart. This facilitates consideration during general program design of each individual connection and its associated passed parameters. The resulting design can be documented with the HIPO charts. (if the designer decides to have more than one function in any module, the structure chart should show them in the same block. However, the HIPO Hierarchy chart would still show all the functions in separate blocks.) The output of the general program design is the input for the detailed module design. The HIPO input-process-output chart is useful for describing and designing each module.

Modularity-In-Design: An Analysis Based On the Theory Of Real Options

Abstract and Figures

Recommended publications

Towards a Framework for Agent Coordination and Reorganization, AgentCoRe

Marketing in an Information-Intensive Environment: Strategic Implications of Knowledge as an Asset

A Lessons Learned Assessment Approach: The Role of Technology through User Requirements

Organization and Uncertainty