working paper series · to cover an increasing range of applications. the evolution of sap’s core...
TRANSCRIPT
College of Business Administration
University of Rhode Island
2004/2005 No. 5
This working paper series is intended tofacilitate discussion and encourage the
exchange of ideas. Inclusion here does notpreclude publication elsewhere.
It is the original work of the author(s) andsubject to copyright regulations.
WORKING PAPER SERIESencouraging creative research
Office of the DeanCollege of Business AdministrationBallentine Hall7 Lippitt RoadKingston, RI 02881401-874-2337www.cba.uri.edu
William A. Orme
Mark Lehrer
Business Software as a General-Purpose Technology:Experience-Driven Design Innovation in ERP Systems
Business Software as a General-Purpose Technology:
Experience-Driven Design Innovation in ERP Systems
Prof. Mark Lehrer College of Business Administration
University of Rhode Island Kingston, RI 02881 USA
phone: (401) 874-7882
fax: (401) 874-4312
11 November 2004
University of Rhode Island Working Paper
Please do not Quote or Cite without Permission
Business Software as a General-Purpose Technology:
Experience-Driven Design Innovation in ERP Systems
ABSTRACT
General Purpose Technologies (Bresnahan and Trajtenberg, 1995; Helpman, 1998) have been hitherto conceptualized as upstream technical inventions whose downstream commercialization requires sector-specific complementary investments. This article points up the existence of a different class of general purpose technologies that – unlike the classic cases of dynamos, transistors, and computers – originate downstream as special purpose technologies in specific sectors and then evolve upstream into general purpose technologies through discovery of broader applications that lead to design breakthroughs in product embodiment of the technology. The evolving design of enterprise resource planning (ERP) software exemplifies this process of „technology generalization.“ Technology generalization is related to the process of technological convergence (Rosenberg, 1972; 1976), but lays the emphasis on design improvements enabled by the accumulation of experience and exposure to multiple environments rather than on changes in industry structure.
1
Business Software as a General-Purpose Technology: The Rise of ERP Systems
1. Introduction
The advent of enterprise resource planning (ERP) systems in the 1990s created a
major new category of standardized software dominated by a handful of large software
companies (SAP, PeopleSoft, Oracle, Baan). Notwithstanding the odd terminology, ERP
refers simply to an integrated suite of business application software supporting the firm’s
basic functions (manufacturing, finance, marketing, etc.). The hallmark of modern ERP
software is that, with the necessary customization, it can be implemented across a wide
range of firms and industries, displacing the traditional reliance on idiosyncratic firm-
specific business software. As such, ERP systems would appear to be a good example of
General Purpose Technologies, or GPTs (Bresnahan and Trajtenberg, 1995; Helpman,
1998). Frequently cited examples of GPTs include electric motors and steam engines
(Bresnahan and Trajtenberg, 1992; Rosenberg and Trajtenberg, 2001), dynamos and
computers (David, 1990), transistors and semiconductor chips (Helpman and Trajtenberg,
1998b). Lipsey, Bekar and Carlaw (1998) endeavor to draw up a first inventory of GPTs.
GPTs are technological innovations with the capacity to diffuse across a wide set of
industries. Related notions include macroinventions (Mokyr, 1990), strategic sectors
(Zysman and Tyson, 1983), technoeconomic paradigms (Freeman and Perez, 1988) and
2
other variations on Kondratiev’s waves-of-innovation model (Schumpeter, 1939; Freeman
and Louca, 2001). Common to all of these concepts is the ambition to provide a micro-
macro link between the generation of new technological knowledge at local levels and
economic growth at more global levels. A fundamental issue is one of multiple complex
complementarities entailed by new technologies with wide-scale applications. Most new
technologies, far from being plug-and-play products, require substantial complementary
investments in order to fit into the existing production system (Tushman and Anderson,
1986; Anderson and Tushman, 1990).1 The existence of innovation complementarities
across sectors in the diffusion of new technologies leads to appropriability problems and
hence also to innovation lags (Bresnahan and Trajtenberg, 1995), productivity lags (David,
1990) and cyclical patterns of expansion and investment (Freeman and Perez, 1988;
Helpman and Trajtenberg, 1998a).
Innovation complementarities also create a potential critical mass problem.
Adoption of a new technology may hinge on additional R&D investment, standardization,
and large-scale production of the core technology in question; yet these economically
necessary and efficient supply conditions may fail to materialize in the absence of demand
across a wide number of industrial sectors. In other words, GPTs present an economic
chicken-and-egg problem akin to that treated in the literature on dominant designs
(Abernathy and Utterback, 1978; Utterback, 1994).
The interest of the EPR case lies in the particular path of resolution of this chicken-
and-egg problem. In a nutshell, ERP systems did not begin as general purpose
technologies. Instead, they began as specific purpose technologies that evolved into
generic purpose technologies through repeated installation and redesign of the core product
3
to cover an increasing range of applications. The evolution of SAP’s core ERP products,
the R/2 and R/3 systems as recounted below, exemplifies the process of “technology
generalization”: the process of refining a technology product so that it accommodates the
needs of a range of industries rather a specific industrial application. This process is not
specific to ERP systems nor is it even uncommon; parallels will be pointed out in machine
tools and chemical engineering.
ERP systems belong to a class of GPTs that differ from classic technologies like
dynamos, transistors, and computers. For while these latter GPTs began upstream as
general-purpose technologies that had to be adapted to specific applications, ERP systems
are representative of technologies that first arise downstream as special-purpose
applications and evolve through learning and design improvements into GPTs. Their
economic implications therefore differ from those of the GPTs most commonly analyzed in
prior literature. In particular, the appropriability problem posed by innovation
complementarities is less pronounced than for innovations like dynamos, steam engines,
and semiconductors.
2. General Purpose Technologies: Definitions and Variations
An advantage of the GPT concept, originally formulated by Bresnahan and
Trajtenberg (1992), is definitional parsimony. As represented in Figure 1, a GPT is a
technology that can be applied across a variety of application sectors (AS). All variations
of the GPT concept assume the marginal cost of adapting the GPT to a new AS to be non-
trivial, although economic models of the cost function vary (Bresnahan and Trajtenberg,
4
1995; Bresnahan and Gambardella, 1998; Helpman and Trajtenberg, 1998a). The marginal
cost of sector adoption depends also upon the level of investment in the GPT occasioned by
adoption among other sectors (innovation complementarities). For example, the cost of
introducing microprocessor chips into telephones will be lower if microprocessors have
already been adopted by other sectors (computing, radio, etc.) whose demand has
stimulated prior R&D and improvements in the basic GPT. Two types of positive
externalities may lead to socially sub-optimal diffusion of GPTs due to the
nonappropriability of rents. First, “horizontal” externalities among different application
sectors result in a coordination problem in which latent demands for a common GPT lack
effective aggregation. Second, a “vertical” externality between a GPT and an application
sector may prevent GPT inventors from appropriating the full benefits the GPT, thereby
leading to underinvestment and undersupply of the GPT (Bresnahan and Trajtenberg,
1995).
-- Insert Figure 1 About Here --
Innovation complementarities affect the sequence of application sectors adopting
GPTs. GPTs diffuse first to sectors in which adoption costs or low and then to sectors for
which the cost of adoption is lowered by prior diffusion of the GPT. The history of
electricity exemplifies the process (Passer, 1953). After the invention of electromagnetic
dynamos in the 1860s, large-scale industrial applications of high-voltage electricity
emerged first in lighting (1870s), then transportation (1880s), and later capital equipment
(1890s). The complementarities were both horizontal and vertical. The proven feasibility
5
of high-voltage electricity for lighting (first arc-lighting in the 1870s, then incandescent
lighting in the 1880s) drove the supply and technology development of central electricity
plants (i.e. investment and R&D in the core GPT), which was crucial to the next major AS,
namely electric streetcars, which required vast amounts of power from a centralized source.
The industrial use of electricity for applications like electrochemistry accelerated when
centralized power generation was achieved on a high scale in the 1890s using AC instead
of DC, another major advance in the core GPT (Passer, 1953).
In similar fashion, transistors diffused quickly to sectors like hearing aids and
computers but much more slowly in sectors like telecom and automobiles where they were
difficult to incorporate into a complex pre-existing production infrastructure (Braun and
MacDonald, 1978; Helpman and Trajtenberg, 1998b). The adoption sequence is depicted
in Figure 2. The dynamo (electromagnetic generator) and transistor are two examples of
GPTs that were invented in research laboratories as major technical breakthroughs whose
path of industrial adoption was difficult to foresee. While Werner Siemens was a co-
inventor of the dynamo and instantly intuited electricity’s potential to transform the
industrial world (Feldenkirchen, 1997a: 53-4), his company subsequently lagged in the
development of high-voltage applications after failing to anticipate the boom in electric
lighting and central generating plants (Feldenkirchen, 1997b). A similar irony prevailed in
transistors, invented by Bell Labs for use in telephony but not actually adopted in
telecommunications until much later than in other application sectors (Braun and
MacDonald, 1978). In sum, the dynamo and transistor are good examples of GPTs whose
invention preceded commercial innovation. Put differently, the dynamo and transistors fit
6
the linear R&D model with sequential stages of basic research, development for specific
applications, and industrial commercialization.
-- Insert Figure 2 About Here --
The friendly amendment to the innovation literature raised here is to point out the
existence of GPTs that originate through a different kind of process In particular, many
GPTs actually begin as special purpose technologies whose general applications only
become apparent over time. Here, too, complementary investments in the core technology
are necessary for diffusion across application sectors, but the basic technical challenge is
different. Instead of investing in development of a generic technological invention to fulfill
a specific commercial purpose, the basic task may be one of redesigning an already
commercialized technology so as to widen its range of deployment. As producers of a
novel technology learn about its broader applications, they may find ways to unbundle the
technology from its specific original application and redesign it as a GPT suitable to serve a
multitude of application sectors.
This process of “technology generalization” is schematized in Figure 3. In contrast
to the linear model of R&D, technology generalization underlines the incorporation of
information feedback from downstream market applications into the upstream innovation
(R&D) process (von Hippel, 1988; Gibbons et al., 1994). Figure 3 illustrates a process of
GPT emergence driven by the accumulation of high-level design knowledge as a new
technology finds application beyond its initial AS. In top-down GPTs like the dynamo and
transistor, the first commercial breakthrough involves tailoring the upstream technology to
7
a downstream application sector; the critical complementarity is a vertical one. In the class
of GPTs examined here, the initial complementarities are more horizontal; the accumulated
experience of applying a similar technology across multiple application sectors results in a
GPT based on a breakthrough in design. This is somewhat analogous to processes of
market feedback leading to a dominant design in “lead markets” (Beise, 2001).
-- Insert Figure 3 About Here --
One question left open by the foregoing discussion is whether GPTs are
components of the end product or of the capital goods that produce the end product. In
fact, GPTs can refer to both. Since ERP systems are capital goods, much of the following
discussion will be most germane to capital goods. The process of technology
generalization can clearly apply to consumer products as well, however; the personal
computer notably comes to mind as a product whose range of applications has increased
over time, with feedback from each generation of technology leading to design
improvements facilitating yet further extensions of product functionality. Of course, the
personal computer has also been a capital good since the inception of spreadsheets;
generally speaking, the macroeconomic effects of GPTs are most significant for
technologies with complementarities across sectors in both business and consumer product
sectors, for it is here that GPTs are most tangible as forces generating increasing returns
and nonconvexities across the entire economy (Helpman and Trajtenberg, 1998a: 82). The
ensuing focus on capital goods GPTs is thus duly noted here as a limitation of the present
analysis.
8
3. Technology Generalization in Capital Goods: ERP and Its Predecessors
ERP systems did not begin as general purpose capital goods. Instead, they evolved
into them. The history of machine tools in the 19th century reveals similar dynamics. The
American system of manufacturers, based on interchangeable parts, first took root in arms
manufacture but later diffused to other manufacturing sectors. The concrete transmission
mechanism of cross-sector diffusion was machine tool production (Rosenberg, 1972;
1976). For example, turret lathes, which allowed multiple cutting operations to be
performed on a workpiece, originated in firearms production but later spread to virtually all
metalworking industries.
Another key innovation was the milling machine, which allowed regular spiral
grooves to be imparted into metal pieces so that the pieces could be assembled with less
filing and fitting of parts. The origin of the milling machine is considerably different from
that of the transistor. At first the milling machine was simply a superior piece of
machinery for producing firearms. Its qualities as a GPT emerged gradually through a
process of iterative market experience and design improvement:
The firearms industry was also the source of the milling machine … The versatility of the milling machine was greatly increased during the Civil War when the Brown and Sharpe Company of Providence, Rhode Island, a firm which had been engaged in sewing machine production, attempted to develop a new method for making twist drills … The result was the universal milling machine … Within ten years after Brown and Sharpe produced the first universal milling machine in 1862, the firm had sold similar machines to manufacturers of hardware, tools, cutlery, locks, arms, sewing machines, textile machinery, printing machines, professional and scientific instruments, locomotives, to machine shops and foundries, and to machine tool manufacturers. In subsequent decades, with each new product innovation, universal milling machines were sold to a succession of firms producing cash registers,
9
calculating machines, typewriters, agricultural implements, bicycles, and automobiles (Rosenberg, 1972: 104-5). The progressively widened application realm of the milling machine exemplifies the
process of technology generalization, which consists above all in design improvement, a
point central to the analysis of ERP systems below. Rosenberg’s own term, “technological
convergence” (1976: 16), is obviously related but lays more emphasis on changes in
industry structure resulting from the growing similarity of basic production techniques
employed across manufacturing sectors. The central 19th-century industrial development
studied by Rosenberg is the emergence of machine tools as an independent sector; prior to
the 1850s, most manufacturers had to produce their own specialized machine tools.
“Technological convergence” refers to the economic process by which a general-purpose
machine tool industry emerged in place of specialized in-house machine tool production by
final manufacturers; following Stigler (1951), Rosenberg’s chief interest is the division of
labor and vertical disintegration. “Technology generalization,” in contrast, refers more to
intellectual breakthroughs underlying the design of products like machine tools; the
following analysis stresses the accumulation of technical and market knowledge resulting
in breakthrough or dominant designs (Utterback, 1994; Beise, 2001). In the case of milling
machines, this process is summarized in Figure 4.
-- Insert Figure 4 About Here –
The intellectual breakthrough required by technology generalization is above all
one of abstraction. Making a specific purpose technology tractable to a wider range of
industrial application requires a broader grasp of the technology’s basic function. In
10
milling machines, for example, a leap had to be made from “a machine that engraves
interchangeable stocks and barrels so that they can be screwed together” to “a machine that
creates grooves in any interchangeable parts that need to be screwed together.” In practice
such abstraction may result from the exposure to multiple industries and thus the
opportunity to perceive components and procedures that are common to a broad range of
industrial applications. First movers in the development of GPTs out of specific purpose
technologies are often those firms who were early in exploring multiple user industries
rather than remaining specialized in a given user group. This was fairly evident in the
history of Brown and Sharpe in the 19th century (Hounshell, 1984: Ch. 2) and SAP in the
20th century. Like milling machines, SAP’s business software did not begin as a general
purpose capital good. Instead, it evolved into one. What began as a product most useful to
chemical and machinery companies proved adaptable to ever wider categories of firms.
An example of technology generalization through abstraction is the discipline of
chemical engineering, which Rosenberg (1998) classifies as a GPT and which bears
resemblance to ERP software in its pattern of emergence. Before the establishment of the
chemical engineering field, the process of making chemicals was highly firm-idiosyncratic,
as was the associated terminology. Though chemical engineering courses had been offered
since the 1880s as a mixture of mechanical engineering and chemistry, the field’s
boundaries remained murky until WW1 when the analysis of “unit operations” caught on as
the specific province of chemical engineers. Arthur D. Little first articulated the basic
principles of what he initially called unit actions in 1915 as follows: “Any chemical
process, on whatever scale conducted, may be resolved into a coordinated series of what
may be termed ‘unit actions,’ as pulverizing, mixing, heating, roasting, absorbing,
11
condensing, lixiviating, precipitating, crystallizing, filtering, dissolving, electrolyzing and
so on” (Little, 1933: 7). The accomplishment of Little and his colleagues at MIT was to
transcend specific chemical properties by articulating at an abstract level what sets of
operations were involved in the handling and making of chemicals. The character of
chemical engineering as a GPT was subsequently shown by the discipline’s fecund
applications not only to chemicals, but also to petroleum, plastics, synthetic fibers, and
synthetic rubber. The chemical engineering discipline furnished a technical vocabulary and
accumulating stock of scientific knowledge that could be utilized across multiple industries
(Rosenberg, 1998).
A.D. Little’s formulation of fundamental principles of chemical engineering
apparently derived from his prior experience as a consultant in various manufacturing
industries, especially paper and pulp. Rosenberg (1998: 178) surmises: “It seems
reasonable to conclude that Little’s consulting experience … sharpened his awareness of
the extent to which entirely different product lines were making use of similar operations.”2
The pioneering of ERP software by SAP reveals a quite analogous process: as SAP gained
acquaintance with the requirements of a widening circle of corporate customers, it also
gained a more abstract understanding of firms’ generic “unit operations” (to borrow Little’s
term) and gradually acquired the knowledge necessary to design a business software
package supporting the business functions of firms from virtually any industry setting.
4. Technology Generalization in ERP Systems: The Case of SAP
12
The origin of modern ERP systems is largely coextensive with the history of SAP
AG of Germany. When SAP introduced its R/3 software suite in 1992, there was no other
product on the market that could match either its functionality (a suite of integrated
business applications for the various functions of a firm: Finance, Production, Human
Resources, etc.) or its technical architecture (R/3 had a client-server architecture at the time
when client-server architectures were fairly new but making dramatically rapid inroads on
mainframe and minicomputer architectures). Other subsequent vendors of ERP software
packages (PeopleSoft, Baan, Oracle, JD Edwards) were essentially followers and imitators
of SAP; SAP thus enjoyed a first-mover advantage and has consistently enjoyed a 30-40%
global market share in ERP installations, three times the share of the next-largest ERP
vendor, Oracle (Norris et al., 1998: Ch. 8; Rashid, Hossain, and Patrick, 2002). Its initial
monopoly allowed SAP to charge very high prices (typically about $2000-$3000 per
company employee for the basic license) while accumulating over 17,000 customers by
2001 (Rashid, Hossain, and Patrick, 2002: 10). Growing to over $3 billion in annual sales
since 1998, SAP has consistently remained the fourth-largest software company in sales,
behind only Microsoft, Oracle, and Computer Associates.
Although R/3 was SAP’s blockbuster product on world markets, it was not the first
SAP product that could be considered a GPT. R/3’s predecessor R/2 was likewise a suite
of integrated business software suitable for use by almost any firm in any industry. R/2
came be widely installed throughout corporate Germany in the 1980s, but outside the
German-speaking world R/2’s application was limited mainly to SAP’s bread-and-butter
industries of chemicals and petroleum (Plattner, 2000). In the 1980s the alternative to an
ERP system was simply not to have one at all and to rely instead on a patchwork of
13
heterogeneous business software systems that were loosely coupled rather than tightly
integrated. This was the route taken by most companies outside Germany prior to the
1990s, and of course loose coupling offers certain advantages (Orton and Weick, 1990;
Brusoni, Prencipe, and Pavitt, 2001).
The sudden preference of companies worldwide for the tightly integrated software
that SAP offered derived from a conjunction of supply and demand shifts. The transition to
client-server networks entailed a massive overhaul in corporate hardware systems, thus
providing a window of opportunity for one-time large-scale overhauls of corporate IT
systems in which accumulated legacy business software was jettisoned and replaced with a
unified ERP package. The Y2K scare exacerbated this tendency toward wholesale
replacement. Since “re-engineering” was in vogue in the 1990s, many companies
overhauled their business processes and IT systems in synch. As a cohort of Cooper &
Lybrand consultants put it, “an SAP software installation and a business process re-
engineering effort go hand in hand” (Norris et al., 1998: 7). Finally, SAP’s track record
was firmly established by the 1990s; the feasibility of tightly integrated software systems
(in contrast to loosely coupled patchworks) was now empirically demonstrated and
believed in by the critical partners of SAP, the Big Six accounting firms and IT consulting
companies.
A stylized representation of sectoral diffusion of ERP systems, based on the
experience path of SAP, is depicted in Figure 5. Chemical and petroleum firms were lead
adopters of ERP systems, followed by machinery makers. This can be seen in the customer
history of SAP. Its first customer (1972) was the newly established German subsidiary of
UK chemical giant ICI. SAP’s first foreign subsidiary was in Switzerland where SAP
14
predominantly served the Swiss chemical, pharmaceutical and machinery companies. The
first UK customer was ICI, while DuPont, Mobil and Dow Chemical were among the first
R/2 buyers in the US. Indeed, SAP expressly set up its first US subsidiary in Philadelphia
“so as to be near to the chemical and petroleum companies” (Plattner, 2000: 40). R/3 was
originally developed for medium-sized firms but soon proved a favorite of multinational
companies looking to coordinate their cross-border operations on client-server systems. In
addition to the core base of chemicals, petroleum and machines, manufacturers of food,
electronics, and computers were also rapid adopters of SAP’s ERP software.3 In contrast,
SAP software was slower to diffuse to service industries and the public sector. Especially
for government services, PeopleSoft was perceived to offer a more suitable product,
presumably on the basis of its own experience trajectory.
-- Insert Figure 5 About Here –
In the case of IT products, it makes sense to relate technology generalization not
only to generalization across industries, but also across hardware platforms. SAP’s early
products were tailored to IBM mainframes owned mainly by large companies. R/2 ran on
both IBM and Siemens mainframes. R/3, developed primarily in UNIX, was hardware-
independent and was suitable for large as well as for medium-sized companies, thus
allowing SAP to enlarge its customer base substantially. The evolution of SAP’s business
software from a specific purpose to a general purpose technology therefore consisted of
two main strands, namely increasing functional generality across industries and easier
portability across hardware platforms, both of which served to diffuse ERP systems more
15
widely. In the 1990s, hardware specifications ceased to be an important constraint. For
example, the process of porting R/3 onto Windows NT required only five days of coding
(Plattner, 2000: 42).
The following analysis provides an historical reconstruction of the design decisions
that transformed SAP’s business software into a GPT. Interestingly, the history of ERP
development at SAP reveals an-inverted U-curve in the level of technology generalization
over time, peaking in 1992 with the introduction of R/3. Thereafter, SAP actually began
introducing more industry-specific version of its software, thereby reversing the trend of
producing an ever more universal single standardized software package. The reasons for
this and its implication for our understanding of GPTs will be explored.
5. Evolution of Business Software Design at SAP
Methodology. The following reconstruction is based on a combination of
interviews with German IT experts and a variety of published sources. The author
interviewed the directors of two Fraunhofer Institutes specializing in information
technology and attended six industry association meetings and fairs in the German IT
sector to better understand the sectoral specificities of the German IT sector.4 SAP
permitted the author to attend an SAP-sponsored customer event in Berlin and conduct
brief interviews with a half dozen managers of SAP and SAP partner firms.5 SAP also
directed the author to two monographs that document the company’s history in
considerable detail on the basis of extensive interviews with top SAP managers (Meissner,
16
1997; Plattner, 2000). In the interest of verifiability, references to published sources will
be the preferred means of documenting the basic facts and conclusions derived below.
Background Information on Business Software. Until the 1990s, business processes
were considered too firm-idiosyncratic to permit standardization of business application
software other than for highly specialized tasks. Although standard software packages for
business did become common after 1969 (the date of IBM’s court-mandated unbundling of
hardware and software), these programs were function-specific (accounting, payroll, etc.).
All major ERP vendors started out selling function-specific software: SAP specialized
originally in accounting and purchasing, PeopleSoft in human resources, and Baan in
manufacturing (Meissner, 1997: 86-7; Rashid, Hossain, and Patrick, 2002). Packaged
software was co-specialized to specific hardware, with varying costs of portability across
hardware platforms.6
Reusable Software Code: SAP’s Original Idea. SAP was founded in Walldorf,
Germany by five ex-IBM programmers in 1972. At the time, computers were so expensive
that SAP had to conduct all of its product development on customer premises; SAP did not
own its own computer until 1980. Like many IT professionals in the 1970s, SAP’s first
manager-programmers had to store and transport their computer programs in boxes of
punch cards. If only for technical reasons, SAP’s early software programs had to be fairly
customized, given the need to optimize the use of costly CPU time and the fact that
reproducing software code was still relatively expensive; in other words, scale economies
were lower than in the PC era.
SAP’s basic strategy was to economize on coding costs by reusing algorithms as
programming work migrated from one customer to another. This strategy had to be
17
partially concealed, however, for the market at the time demanded that “you have to offer
users something specifically tailored to them” (firm founder Dietmar Hopp, cited in
Meissner, 1997: 48). In other words, SAP’s initial customers always believed they were
buying a specific purpose technology customized to their needs, which in part they were.
The economies of reusing code were so great that SAP often realized net profits of 50%
during the 1970s (Plattner, 2000: 22). Given the benefits of recycling software code, SAP
used its accumulated experience to enlarge the standardized core code with ever more
adjustable settings and parameters. By expanding the range of such settings and
parameters, SAP could reduce the amount of code that needed to be specially written for
each individual’s premises.
By way of illustration, SAP programmers developed parameters encompassing
language, time zone, tax laws and currency variables to serve the needs of multinational
companies. It was a locational advantage of SAP’s European base that the company had to
support multiple languages and tax laws from early on. One of the most significant product
enhancements resulted from the decision to support Japanese for R/2 in the 1980s, which
required developing a double-byte version of the entire software, the costliest engineering
modification SAP had ever engaged in until then. SAP took on the project because “the
German multinational chemical companies asked us to go to Japan, then DuPont asked us
to as well” (Plattner, 2000: 48). The result of this engineering change was that SAP’s
software could also support other languages requiring double-byte characters, such as
Chinese and Korean. SAP’s support of multiple currencies resulted in a competitive
advantage, since multiple currencies were “a complexity which US software has trouble
coping with” (Plattner, 2000: 49).
18
From Reusable Code to General Purpose Technology: R/2. By incorporating ever
more parameters to fit an ever greater variety of application environments, SAP’s business
software acquired an increasing level of functionality and application generality over time.
In 1979, SAP embarked on development of its first truly generic software suite, R/2. Its
outstanding characteristic was real-time entry and analysis of data on a time-sharing system
that served and integrated all of the firm’s different functions. Computer-generated reports
could be produced at any time, rather than, as was customary, at the end of the month when
printouts from different company IT systems were assembled and reconciled by human
analysts (Ceruzzi, 2003: Ch. 3).
Completed in 1981, R/2 was designed for mainframe systems using timeshare
terminals. Originally coded for IBM computers, R/2 was ported in the early 1980s onto a
common platform so that it could be installed on the mainframes of Siemens as well.
Within ten years, R/2 quietly conquered the German market for large-business software on
mainframes. The IT magazine Computerwoche noted in 1990: ‘Almost unnoticed by the
general public, SAP AG of Walldorf has acquired a quasi-monopoly in commercial
software for IBM-370 computers in the Federal Republic with the modular standard
software package R/2’ (Meissner, 1997: 53). At its height, R/2 had 2200 installations.
Its rapid diffusion notwithstanding, R/2 was costly to install. This was because
R/2’s general-purpose flexibility entailed a high level of product complexity. Each
installation necessitated years of parameter-setting to conform to specific customer
requirements. The installers of an R/2 or R/3 package typically fill out tens of thousands of
different tables over a process of 1-2 years; if ERP installation also coincides with a
process of business re-engineering, such an installation can take 2-4 years (Cooke and
19
Peterson, 1998; Norris et al., 1998). Indeed, R/2 and R/3 are so exceptional in their degree
of complexity that a normal installation utilizes only about 20% of the package’s total
functionality (Norris et al., 1998: 23).
Another implementation cost imposed by R/2 ensued from the requirement of
imposing homogenous terminology (common definitions, data sets, interfaces, etc.) across
sub-units within the company. Only so could R/2’s cross-functional capability be
operationalized. Given the high coordination, implementation, and complexity costs that
R/2 installation entailed, R/2 was only feasible in large companies whose business
processes were highly routine and relatively stable over time. Firms in Germany’s leading
industries generally fit this description. Germany has come to specialize in “medium-tech”
industries based on mature core technologies while lagging in faster-moving high-tech
industries (Casper, Lehrer, and Soskice, 1999; Siebert and Stolpe, 2002).7 Germany’s
home-country strength in chemicals, energy and machinery provided a good fit with ERP
software that was suited to highly codifiable and stable firm practices.
An outstanding feature of R/2, and inseparable from its character as a GPT, was the
level of abstraction that informed the product’s design. Most business software of the time
was designed to mirror existing practices in all of their functional idiosyncrasies. That is,
accounting software mirrored existing accounting practices, purchasing software was built
on existing purchasing practices, and so forth. One of Germany’s leading MIS professors,
August Wilhelm Scheer, recalled how SAP’s software transcended such tendencies through
unprecedented abstraction:
What especially impressed me was that the [SAP] data model made clear the parallel between purchasing and sales. Because more or less the same concepts
20
appear in both areas. With sales it’s called “customer,” with purchasing “supplier,” but in both you have bills, payment warnings, sales conditions, so that the whole structural similarity became clear. And I suddenly saw: this is a new approach to management thinking! This was the integration idea, that book value information like debit accounting belonged together with operative processes like sales. This involved different thinking from what we had in management school, where we were told something about financial accounting but where later in discussing operating issues like logistics the book values were no longer discussed. Both of these effects, integration and the formalization of representations, i.e. business modeling, got me interested in SAP (cited in Plattner, 2000: 82). The level of abstraction in ERP software increased in another aspect as well. As its
functional flexibility increased, SAP’s business software essentially evolved from a set of
programs into kind of operating system that required programming in order to become
functionally operative. The general-purpose nature of SAP’s software meant that R/2 and
R/3 did nothing when merely taken out of their box, but rather constituted a new IT layer
on top of which the firm-specific functionality had to be constructed as a separate layer of
parameters and supplemental software code.8 The process of abstraction is endemic to IT
development where multiple layers of software have accumulated to facilitate program
adjustments as well as to provide interfaces among different information and
communication systems. Indeed, IT developers actually use the term “abstraction layer” to
refer to a software program that provides a generic platform so that a given IT component
or layer can interface more easily with other IT components or layers.9
6. Complexity Limits to Technology Generalization, Part One: Scale – The Development
of R/3
21
For all intents and purposes, R/2 constituted a GPT. R/2 was a general purpose
software package that could and did diffuse to a variety of application sectors. ERP
technology continued to evolve, however. The further evolution of ERP systems sheds an
interesting light on bottlenecks entailed by the process of technology generalization. The
two bottlenecks of interest here are scale and implementation cost. The interesting aspect
of ERP systems, as experienced by leading developer SAP, was that the scale bottleneck
could be more or less overcome through design innovation, whereas the complexity
bottleneck could not. Design innovations were able to make ERP systems capable of
running on IT systems of any size, but they could not, in the final analysis, reduce the
complexity emanating from the diversity of task environments that ERP systems had to
operate in. SAP ultimately went back to developing more specific-purpose ERP software;
the result was an inverted U-shaped curve in the level of technological generality of SAP’s
ERP software over time.
The potential downside of technology generalization illustrated by ERP systems is
technical complexity. Since SAP achieved the broadening of application scope by adding
parameters to its ERP software, the ultimate result was a plethora of parameters and
settings that users had to make choices about. R/2 was predicated on the assumption that
only companies able to afford large mainframe computers could take on the effort and
expense of implementing ERP software. Following advances in hardware and new IT
architectures, SAP developed its next ERP product, R/3, to tap the market of medium-sized
firms; R/3 made ERP technology potentially accessible to companies of all but the lowest
size, as it could run on very large as well as very small IT systems. However, there was no
development from R/3 to any kind of R/4 that could overcome the implementation cost
22
bottleneck that had always dogged ERP systems. Instead of developing a new GPT called
R/4, SAP ultimately retreated from developing a single general-purpose ERP product and
began offering industry-specific versions of its ERP software. SAP overcame the scale
bottleneck by redesigning its GPT; it was not able to overcome the implementation cost
bottleneck in this way.
From Large to Scaleable GPT: R/3. SAP’s follow-up product, R/3, formed the
basis of its ERP market leadership beyond German borders. R/3 replaced the IBM-inspired
mainframe architecture of R/2 with UNIX-centered client-server architecture. Thanks to
this, R/3 was scalable, permitting the company’s ERP system to be augmented at will by
adding servers and clients. Through technical advances in PCs and microprocessors, the
power of client/server systems could equal that of mainframe systems. While SAP
managers did not foresee this turn of events, SAP in the early 1990s nonetheless found
itself with a monopoly on modular ERP systems that could run on a client/server
architecture. Serendipity played a major role in both the development and diffusion of R/3.
Development of R/3 began in 1988, when R/2 was slowly reaching saturation
within the German large-company market. In developing ERP software for medium-sized
companies, SAP originally conceived R/3 as a complement, not a successor to R/2. Even
when R/3 was introduced in 1992, SAP was unaware of the potential market of large
multinational companies for R/3. Expressed in GPT terms, SAP did not know, in
embarking on development of R/3, that it was about to generalize its ERP software from
large-firm applications to cover medium-sized firm applications as well. SAP did not
originally believe that a single ERP package could suit both large-firm and medium-sized
firm markets.
23
SAP’s ability to develop a GPT for firms of all sizes was demonstrably indebted to
the accumulation of new experience, albeit not so much in new application sectors as in
new IT programming environments. R/3 was not originally developed with UNIX,
Windows, PCs, workstations, or any of the other IT icons of the 1990s in mind.
Nonetheless, development work on R/3 did bring SAP into the realm of UNIX, relational
databases, and the programming language C, all part of the later hegemony of PC-based
client/server networks. SAP embarked on R/3 to follow the announced technology
trajectory of IBM, SAP’s primary hardware complementor. IBM had outlined its Systems
Application Architecture (SAA) as a method for creating overarching compatibility across
IBM equipment. Since IBM’s own operating system for SAA was not yet released, SAP
performed its development on a mixture of IBM and UNIX machines. R/3 ended up as a
UNIX-based system when the SAA program of IBM fell behind schedule. As coding of
R/3 neared completion, SAP found itself unable to make R/3 operational on IBM hardware.
It was a mixture of accumulated IT experience and unexpected crises that induced
SAP to generalize its ERP technology beyond IBM hardware specifications in 1991. SAP
was scheduled to present the debut of R/3 at the 1991 CeBIT fair in Hanover, yet only six
weeks before the fair SAP determined that it could not do so on IBM equipment. Rather
than cancel its presentation, SAP elected to engage in massive technical improvisation to
save face. Having just received new UNIX workstations, SAP managers scrambled to port
R/3 onto one of them. The unanticipated outcome was that R/3, a gigantic software
package designed to run on IBM 370 mainframes, actually fit entirely onto a single
workstation:
The only lie we told [at the Hanover fair] was this. R/3 itself was on the machine. It was a lie, because we said the R/3 system is in Walldorf on the mainframe, this
24
here is just the UNIX terminal. The truth was that the whole R/3 system ran locally in this tiny machine … We had to hide this sensation; nobody at the time could have comprehended it. (Plattner, 2000: 98-9) In preparing for the CeBIT fair, SAP deepened its experience with UNIX-based
equipment and awareness of what technological progress in workstations and PCs was
achieving. SAP’s management admitted that, due to its IBM orientation, “we almost
missed the PC because we thought it was a toy” (Plattner, 2000: 46). Shortly after the
Hanover fair some SAP customers began expressing interest in UNIX-based systems,
which previously were used mainly in academic institutions. A maneuver to save face at
the CeBIT fair thus resulted in a new design trajectory for R/3. SAP ended up developing
R/3 specifically for UNIX rather than for IBM equipment. By 1992 SAP was able to roll
out R/3 on UNIX machines independently of any manufacturer. By 1993, it was clear that
R/3 would scale higher than R/2 with improvements in hardware; in that year, a version of
R/3 for Windows NT was developed.
Though SAP initially sold R/3 as planned to medium-sized firms, Chevron’s
surprise purchase of R/3 opened SAP’s eyes to the potential of client/server ERP systems
for large firms. Thereafter R/3 diffused rapidly among sought by global companies looking
to integrate their cross-border operations on increasingly standardized server and PC
hardware. In the years 1990-1999, SAP’s turnover increased 20-fold from DM 500 million
to over DM 10 billion (5 billion Euro).
One final design detail is worth mentioning. In converting R/3 to a UNIX-based
client/server system, SAP pioneered the three-tier client-server architecture that has now
become common (Rashid, Hossain, and Patrick, 2002). In the early 1990s, client/server
architectures were mainly two-tier: either “fat-client” or “fat-server.” SAP’s three-tier
25
architecture places the database on a centralized server (“database level”) and the user
interface on the clients (“presentation level”). SAP’s design innovation, emanating from
trial and error in development of R/3, was to place the applications on an intermediate level
of decentralized servers (“application level”). The advantage of this design, later emulated
throughout the IT industry, was that it avoided informational traffic jams in accessing
applications.
7. Complexity Limits to Technology Generalization, Part Two: Implementation Cost – The
Development of Industry Solutions
The increasing application generality of ERP software came at the expense of
greater product complexity. The fact that most SAP installations use only about 20% of the
package’s total functionality are an indication of this complexity. Anything but a plug-and-
play product, R/3 encompasses about 25,000 different tables that can be filled out
(“parameterization”) to permit customization of the product to a given customer’s premises.
Such complexity entails high implementation costs, including supplemental software
coding to perform customer-specific tasks not covered within the R/3 package. One large-
scale survey estimated SAP installations in the US to cost an average of $20 million, with
larger firms frequently spending over $100 million (Cooke and Peterson, 1998). Of this,
only 14% on average was for the software license; 70% of the total cost went toward
implementation and 16% toward hardware. This mirrors the ERP rule-of-thumb that the
cost of implementing ERP software generally runs above five times the purchase price of
the basic software license (Schwarz, 2000: 27). In fact, the most widespread criticism of
26
SAP (and other ERP systems) concerns precisely their complexity, which almost always
requires SAP or (more often) an external accounting, consulting or IT company to play a
major hand in ERP implementation. An article specifically charging that R/3’s “high
system complexity results in inflexibility” was widely cited in the German business press,
much to SAP’s chagrin (Meissner, 1997: 139). The evolution of ERP software illustrates
the possible trade-off between complexity and generality of technology.
Not surprisingly, then, a major strategic effort of SAP since the mid 1990s centered
on reducing the complexity and cost of implementing R/3. An example is Accelerated SAP
(ASAP), a methodology for implementing SAP in 6-9 months (rather than 2-4 years) for
customers of moderate size and business complexity. More telling, however, are SAP’s
growing set of “industry solutions,” namely supplementary modules designed to suit the
requirements of users in specific industries. Such industry solutions, developed in
collaboration with key users and consulting partners, help users economize on the amount
of parameterization and supplementary coding required by ERP implementation.
One can regard such industry-specific modules as a step backward from prior efforts by
SAP to upgrade its ERP software into an ever more universal, one-size-fits-all technology.
Instead, the application universality of SAP’s ERP software peaked with the introduction of
R/3, while many subsequent developments essentially involve customization of R/3 to
different industries. Such industry solutions have been developed for the following sectors:
• Aerospace and defense • Automotive • Banking • Chemicals • Consumer products • Engineering and production • Health care
27
• High tech and electronics • Insurance • Oil and gas • Pharmaceuticals • Public sector • Retail • Telecommunications • Utilities
SAP’s shift to industry-specific versions of its ERP software amounts to a reversion to the
classic case of a GPT that needs to be tailored to industry-specific applications. The
industry solutions are overlays placed on top of the generic R/3 software; they are
essentially semi-customizations designed to lessen the burden of installation in specific
industries. SAP’s industry solutions constitute industry-specific investments in the basic
GPT, thereby conforming to the original conception of GPTs as technologies whose use in
specific application sectors requires supplementary sector-specific investment (Bresnahan
and Trajtenberg, 1995).
The process of technology generalization can thus be limited by complexity
constraints. The level of technological generality of ERP systems at SAP reveals
something of an inverted U-curve, with generality across industries and hardware platforms
rising and peaking with the introduction of R/3 in 1992, but decreasing thereafter with the
continuing introduction of new industry-specific modules from the mid 1990s on. Within
the company, the willingness to back away from an all-encompassing single product
involved something akin to a strategic inflection point (Burgelman and Grove, 1996) in
which variations were accepted precisely as variants and not as additional variables
needing to be programmed as new parameters in the basic package:
28
We always had a deep fear of them [variants] … We always had this fear and in historical perspective it came into our heads as we transformed R/1 into R/1.5 on IMS … So it is now an important process that is taking place and must take place with us, that we recognize variants simply as such and not compulsively make a simple variant into a complex one that is close to generic in nature (Plattner, 2000: 133-4). Another way to put the matter is that SAP encountered initially increasing and later
diminishing returns to technology generalization. Analogous patterns can probably also be
found in many histories of product standardization, though technology generalization is not
always the same kind of process: whereas standardization endeavors to homogenize a
product across application areas, technology generalization can also aim at enhancing the
product’s capability to perform heterogeneous tasks across a variety of environments.
8. Conclusion: Technology generalization and dominant designs
ERP systems constitute a class of GPTs for which the economic bind posed by
innovation complementarities is less significant than for dynamos and transistors whose
invention in the laboratory preceded industrial application. One can almost say of ERP
software that innovation preceded invention: it was in the act of reusing and enlarging
software code that a general purpose technology emerged in stages out of a specific
purpose one. Thus, ERP belongs to a class of general purpose technologies that palliate the
externality problem by originating as an economically feasible technology in a specific
sector and diffuse with decreasing marginal costs of adoption in other sectors, leaving in its
wake an accumulation of experience that can be used to broaden the design of the
technology to fit an increasing range of applications. We called this process technology
generalization.
29
The technology generalization process bears some resemblance to the
crystallization of a dominant design. A dominant design constitutes a key step in the
maturation of a new technology from one that is used in many changing variants to one
whose basic design is both stable and suitable for wide-scale application (Abernathy and
Utterback, 1978; Anderson and Tushman, 1990). Dominant designs facilitate serial
production. Prior to a dominant design, heterogeneity in supply and demand engenders
high production costs and prices, whereas the emergence of such a design permits
economies of scale and ultimately significant price reductions. The process of technology
generalization likewise facilitates serial production. In standard software products, of
course, scale economies are particularly pronounced.
In many cases, the dominant design can be seen as the logical and economic
endpoint of the technology generalization process. The dominant design is not to be
confused with a hit product like SAP’s R/3. Rather, the dominant design applies to general
product characteristics across vendors. In the case of ERP systems, such a design clearly
materialized. Rashid et al. (2002: 7-8) list the major characteristics of ERP systems as
follows:
• Modular design comprising many distinct business modules such as financial, manufacturing, accounting, distribution, etc.
• Use centralized common database management system (DBMS)
• The modules are integrated and provide seamless data flow among the modules, increasing operational transparency through standard interfaces…
The general practice is to have three-tier architecture:
• Presentation Layer: graphical user interface (GUI) or browser of data entry or accessing system functions;
• Application Layer: business rules, functions, logic, and programs acting on data received/transferred from/to the database servers;
30
• Database Layer: management of the organization’s operational or transactional data including metadata; mostly employs industry standard RDBMS with structured query language (SQL) provisions. (Rashid, Hossain, and Patrick, 2002: 7-8)
The quoted description of typical state-of-the-art ERP systems amounts to a list of the
major design decisions made by SAP and subsequently imitated by SAP’s followers. SAP
pioneered the integration of functional modules on top of a centralized database using
precisely the three-tier architecture of presentation, application, and database layers.
Nonetheless, these core design characteristics do not really capture the specific effects of
technology generalization in the evolution of ERP systems. The cross-sectoral suitability
of modern ERP systems like SAP’s R/3 depends principally on the programmed flexibility
within the various modules themselves (finance, manufacturing, etc.) rather than on the
software package’s overall product architecture. From the foregoing case study on SAP it
can be reasonably conjectured that this programmable flexibility resulted more from the
accumulated experience of writing business software than from general IT design decisions
governing the overall product architecture. In fact, the principles of customizable standard
business application software that informed SAP’s product development were entirely
foreign to the established principles of software engineering at the time (Plattner, 2000:
190).
This raises the final question of whether technology generalization as defined here
is useful primarily for analysis of “programmable” products like software. Certainly one
recurrent strand in the history of software engineering progress consists in techniques to
leverage the easy reuse of code through subroutines, object-oriented programming, and so
forth. Many of the most useful computer programs are not those that perform the final user
31
functions, but are themselves tools to write other programs, the obvious example being
programming languages like FORTRAN, C, and Java. This fact has resulted in an
accumulation of programming layers in IT systems, illustrating the famous von Neumann
principle of the nonseparability of programs and data. Without a doubt, large-scale ERP
software packages fit into this trend. Requiring extensive parameterization and
customization, they are in some ways more like programming tools than final user
programs. They constitute additional layers in corporate IT systems in the sense that the
specific business functions have to be programmed on top of a generic ERP software
package like R/3.
However, one reason for thinking that technology generalization is not a software-
specific phenomenon is precisely that programming and programmability are not specific to
software. Programs are at the core of all complex systems, social, managerial, and
biological (Beniger, 1986). It is not just that antecedents of software go back to at least the
Jacquard loom with its manually programmable weaving patterns. A growing number of
scholars have concluded that the so-called information revolution forms a continuum with
the industrial revolution (Yates, 1989; Chandler and Cortada, 2000; Freeman and Louca,
2001). Technological advances in production since the 18th century have repeatedly
created the need for new means to coordinate and control larger-scale systems (Chandler,
1977; Beniger, 1986). The history of management is largely the history of coordination
and control technologies, and it is probably in the context of such technologies that ERP
systems are to be placed.
In whatever sectors new control and coordination technologies have emerged, the
potential to utilize them in other sectors (i.e. the potential for technology generalization)
32
has always been latent. For example, the technology of interchangeable parts began in
military hardware and remained largely confined to this sector for fifty years before
diffusing widely to other sectors (Hounshell, 1984). Similarly, the history of modern
managerial hierarchies using information technology to coordinate complex production
systems began with railroad companies using the telegraph and new principles of
organization that later diffused to other sectors (Chandler, 1977). Perhaps the most
spectacular example of technology generalization is the assembly line, generally thought to
have originated in meatpacking before making its spectacular breakthrough in Ford’s auto
works. The diffusion of the assembly line through many industries in the 20th century
subsequently demonstrated how this GPT had hardware components (physical
contraptions) as well as what one could broadly call “software” (here: managerial
practices). Assembly lines and ERP systems alike require that a basic tool be customized
to the firm’s particular context and implemented with a deep set of experience-based
managerial skills. Like ERP systems, the assembly line is no plug-and-play product.
33
REFERENCES Abernathy, W.J. and J.M. Utterback 1978 "Patterns of industrial innovation." Technology Review, 80(7): 40-47. Anderson, P. and M.L. Tushman 1990 "Technological discontinuities and dominant designs: A cyclical model of technological change." Administrative Science Quarterly, 35: 604-633. Beise, M. 2001 Lead markets: Country-specific success factors of the global diffusion of innovations. Heidelberg: Physica-Verlag. Beniger, J.R. 1986 The control revolution: Technological and economic origins of the information society. Cambridge, MA: Harvard University Press. Braun, E. and S. MacDonald 1978 Revolution in miniature: The history and impact of semiconductor electronics. Cambridge: Cambridge University Press. Bresnahan, T. and A. Gambardella 1998 "The division of inventive labor and the extent of the market." In E. Helpman (ed.), General purpose technologies and economic growth: 253-281. Cambridge, MA: MIT Press. Bresnahan, T.F. and M. Trajtenberg 1992 "General purpose technologies: "Engines of growth?"." Working Paper, NBER. Bresnahan, T.F. and M. Trajtenberg 1995 "General purpose technologies: "Engines of growth"?" Journal of Econometics, 65: 83-108. Brusoni, S., A. Prencipe, and K. Pavitt 2001 "Knowledge specialization, organizational coupling, and the boundaries of the firm: Why do firms know more than they make?" Administrative Science Quarterly, 46: 597-621. Burgelman, R.A. and A.S. Grove 1996 "Strategic dissonance." California Management Review, 38(2): 8-26. Casper, S., M. Lehrer, and D. Soskice
34
1999 "Can high-technology industries prosper in Germany? Institutional frameworks and the evolution of the German software and biotechnology industries." Industry and Innovation, 6(1): 5-24. Ceruzzi, P.E. 2003 A history of modern computing. Cambridge, MA: MIT Press. Chandler, A.D. 1977 The visible hand: The managerial revolution in American business. Cambridge, MA: Harvard University Press. Chandler, A.D. and J.W. Cortada (ed.) 2000 A nation transformed by information: How information has shaped the United States from colonial times to the present. Oxford: Oxford University Press. Cooke, D.P. and W.J. Peterson 1998 SAP implementation: Strategies and results. New York: The Conference Board. Curran, T. and G. Keller 1998 SAP R/3 business blueprint. Upper Saddle River: Prentice Hall PTR. David, P.A. 1990 "The dynamo and the computer: An historical perspective on the modern productivity paradox." American Economic Review, Papers and Procedings, 80(2): 355-361. Feldenkirchen, W. 1997a Siemens: Von der Werkstatt zum Weltunternehmen. Munich: Piper. Feldenkirchen, W. 1997b Werner von Siemens: Inventor and international entrepreneur. Columbus: Ohio State University Press. Freeman, C. and F. Louca 2001 As time goes by: From the industrial revolution to the information revolution. Oxford: Oxford University Press. Freeman, C. and C. Perez 1988 "Structural crises of adjustment, business cycles, and investment behavior." In G. Dosi et al. (ed.), Technical Change and Economic Theory: 38-66. London: Pinter. Gibbons, M. et al. 1994 The new production of knowledge: The dynamics of science and research in contemporary societies. London: Sage. Helpman, E. (ed.) 1998 General purpose technologies and economic growth. Cambridge, MA: MIT Press.
35
Helpman, E. and M. Trajtenberg 1998a "A time to sow and a time to reap: Growth based on general purpose technologies." In E. Helpman (ed.), General purpose technologies and economic growth: 55-83. Cambridge, MA: MIT Press. Helpman, E. and M. Trajtenberg 1998b "Diffusion of general purpose technologies." In E. Helpman (ed.), General purpose technologies and economic growth: 85-119. Cambridge, MA: MIT Press. Hounshell, D.A. 1984 From the American system to mass production 1800-1932. Baltimore: Johns Hopkins University Press. Lehrer, M. 2005 "Two Types of Organizational Modularity: SAP, ERP Product Architecture and the German Tipping Point in the Make/Buy Decision for IT Services." In D. Grimshaw and M. Miozzo (eds.), Knowledge intensive services and changing organizational forms: Cheltenham: Edwad Elgar. Lipsey, R.G., C. Bekar, and K. Carlaw 1998 "What requires explanation?" In E. Helpman (ed.), General purpose technologies and economic growth: 15-54. Cambridge, MA: MIT Press. Little, A.D. 1933 "Chemical engineering research: Lifeblood of American industry." In S.D. Kirkpatrick (ed.), Twenty-Five Years of Chemical Engineering Progress. New York: American Institute of Chemical Engineers. Meissner, G. 1997 SAP - Die heimliche Software-Macht. Hamburg: Hoffmann und Campe. Mokyr, J. 1990 The lever of riches: Technology, creativity and economic progress. Oxford: Oxford University Press. Norris, G. et al. 1998 SAP: An executive's comprehensive guide. New York: John Wiley & Sons. Orton, J.D. and K.E. Weick 1990 "Loosely coupled systems: A reconceptualization." Academy of Management Review, 15: 203-223. Passer, H.C. 1953 The electrical manufacturers 1875-1900. Cambridge, MA: Harvard University Press.
36
Plattner, H. 2000 Dem Wandel voraus. Bonn: Galileo Press. Rashid, M.A., L. Hossain, and J.D. Patrick 2002 "The evolution of ERP systems: A historical perspective." In L. Hossain, J.D. Patrick, and M.A. Rashid (eds.), Enterprise resource planning: Global opportunities and challenges: 1-16. Hershey: Idea Group Publishing. Rosenberg, N. 1972 Technology and American economic growth. New York: Harper. Rosenberg, N. 1976 Perspectives on technology. Cambridge: Cambridge University Press. Rosenberg, N. 1998 "Chemical engineering as a general purpose technology." In E. Helpman (ed.), General purpose technologies and economic growth: 167-192. Cambridge, MA: MIT Press. Rosenberg, N. and M. Trajtenberg 2001 "A general purpose technology at work: The Corliss steam engine in the late 19th century US." Working Paper, CEPR. Schumpeter, J.A. 1939 Business cycles: A theoretical, historical and statistical analysis of the capitalist process. New York: McGraw-Hill. Schwarz, M. 2000 ERP-Standardsoftware und organisatorischer Wandel. Wiesbaden: Deutscher Universitäts-Verlag. Siebert, H. and M. Stolpe 2002 "Germany." In B. Steil, D.G. Victor, and R.R. Nelson (eds.), Technological innovation and economic performance: 112-147. Princeton: Princeton University Press. Stigler, G.J. 1951 "The division of labor is limited by the extent of the market." Journal of Political Economy, 54(3): 185-193. Trescott, M.M. 1982 "Unit operations in the chemical industry: An American innovation in modern chemical engineering." In W.F. Furter (ed.), A century of chemical engineering: 1-18. New York: Plenum Press. Tushman, M.L. and P. Anderson 1986 "Technological discontinuities and organizational environments." Administrative Science Quarterly, 31: 439-465.
37
Utterback, J.M. 1994 Mastering the dynamics of innovation. Boston: Harvard Business School Press. von Hippel, E. 1988 The Sources of Innovation. New York: Oxford University Press. Yates, J. 1989 Control through communication: The rise of system in American management. Baltimore: Johns Hopkins University Press. Zysman, J. and L. Tyson 1983 American industry in international competition: Government policies and corporate strategies. Ithaca: Cornell U. Press.
38
Figure 1: Generic Framework of GPTs
GPT
AS1 AS2 AS3 AS4
GPT = General Purpose Technology
AS = Application Sector
Source: Bresnahan & Trajtenberg, 1992
Figure 2: Top-Down Diffusion Pattern
Transistor
Hearing Aids
Computers Radios Telecom
Basis: Bresnahan & Trajtenberg, 1992; Helpman & Trajtenberg, 1998
General Purpose Technology:
Autos
Early Adopting Application Sectors:
Later Adopting Application Sectors:
39
Figure 3: Bottom-Up Diffusion Pattern
GPT
AS1
Application Sectors for Newly Developed GPT
SPT
AS2 AS3 AS4 AS5
GPT = General Purpose
Technology
SPT = Special Purpose
Technology
Application Sectors for Technological Generalization of
SPT into a GPT
design in
novation
Figure 4: Development of Milling Machines
Universal Milling
Machine
Firearms
Application Sectors for Universal Milling
Machines
Basis: Rosenberg, 1972
Lincoln Miller
Sewing Machines
Textiles Trains Etc.
General Purpose
Technology
Dominant Design For Firearms
Production
Application Sectors for Technological Generalization of
Milling Machines
design innovation
40
Figure 5: ERP Development at SAP
R/2 & R/3
Chemicals/ Petroleum
Application Sectors for Newly Developed ERP
System Software
Basis: Plattner, 2001
R/1
Machinery MNCs High Tech
R/2 and R/3 = General Purpose
Technology
R/1 = Special Purpose
Technology (But With Reusable
Components)
Application Sectors for Technological Generalization of
SAP’s Business Software
Etc.
design
innova
tion
41
42
ENDNOTES
1 For example, Rosenberg and Trajtenberg (2001) show how George Corliss devised a variation of the19th-
century steam engine that was suitable for more delicate uses in textile manufacturing.
2 This is not to deny alternative interpretations of how the concept of unit operations arose. For example,
Trescott (1982) connects Little’s concept of unit operations to contemporaneous ideas of scientific
management and mass production.
3 A good idea of the sectors of early diffusion can be gleaned from an SAP primer which claims that as of
June 1996, the list of Fortune 500 American companies using SAP software included: “7 of the top 10
pharmaceutical companies; 7 of the top 10 computer companies; 7 of the top 10 petroleum companies; 6 of
the top 10 electronics companies; 8 of the top 10 chemical companies; 8 of the top 10 food companies”
(Curran and Keller, 1998: xxi)
4 The institutes were the Fraunhofer Institute for Computer Graphics (Darmstadt) and the Fraunhofer Institute
for Software and Systems Engineering (Berlin). 5 The gracious intervention of Barry Nalebuff in this matter is gratefully acknowledged.
6 While versions of high-level languages like FORTRAN might not vary greatly in semantics across different
mainframe manufacturers, the configuration of data storage and input/output devices could vary substantially,
making complete rewrites of the software necessary.
7 In other words, SAP benefited from a domestic industrial environment in which firm processes were more
standardized and, in all likelihood, more explicitly codified than in other countries. See Lehrer (2005).
8 SAP actually supplies its own high-level programming language, ABAP, to code tasks that are beyond the
scope of the software package’s generic capabilities; the percentage of specially programmed code frequently
comprises up to 10% of the total ERP installation code.
9 Abstraction layers are abstract in the sense that they convert the product-specific parameters of IT
components into more generic terms.
Our responsibility is to provide strong academic programs that instill excellence,confidence and strong leadership skills in our graduates. Our aim is to (1)promote critical and independent thinking, (2) foster personal responsibility and(3) develop students whose performance and commitment mark them as leaderscontributing to the business community and society. The College will serve as acenter for business scholarship, creative research and outreach activities to thecitizens and institutions of the State of Rhode Island as well as the regional,national and international communities.
Mission
The creation of this working paper serieshas been funded by an endowmentestablished by William A. Orme, URICollege of Business Administration,Class of 1949 and former head of theGeneral Electric Foundation. This workingpaper series is intended to permit facultymembers to obtain feedback on researchactivities before the research is submitted toacademic and professional journals andprofessional associations for presentations.
An award is presented annually for the mostoutstanding paper submitted.
Founded in 1892, the University of Rhode Island is one of eight land, urban, and sea grantuniversities in the United States. The 1,200-acre rural campus is lessthan ten miles from Narragansett Bay and highlights its traditions ofnatural resource, marine and urban related research. There are over14,000 undergraduate and graduate students enrolled in seven degree-granting colleges representing 48 states and the District of Columbia.More than 500 international students represent 59 different countries.Eighteen percent of the freshman class graduated in the top ten percentof their high school classes. The teaching and research faculty numbersover 600 and the University offers 101 undergraduate programs and 86advanced degree programs. URI students have received Rhodes,
Fulbright, Truman, Goldwater, and Udall scholarships. There are over 80,000 active alumnae.
The University of Rhode Island started to offer undergraduate businessadministration courses in 1923. In 1962, the MBA program was introduced and the PhDprogram began in the mid 1980s. The College of Business Administration is accredited byThe AACSB International - The Association to Advance Collegiate Schools of Business in1969. The College of Business enrolls over 1400 undergraduate students and more than 300graduate students.
Ballentine HallQuadrangle
Univ. of Rhode IslandKingston, Rhode Island