TECHNOLOGICAL FORECASTING AND SOCIAL CHANGE 7, 319-396 (1975) 379

Technology, Productivity and Process Change

WILLIAM J. ABERNATHY and PHILLIP L. TOWNSEND

ABSTRACT

A descriptive model of process evolution is proposed as a new basis for clarifying the relationships
among technological innovation and productivity improvement within an organization. As a produc-
tion process develops over time it does so with a characteristic pattern: process flows become more
rational, tasks more specific, processes more capital intensive, product designs more standardized, etc.
As this development continues over time the overall nature of a process is significantly altered and
various stages of development can be noted that are similar in different industries and sectors. These
stages have important implications for research on technological innovation and productivity improve-
ment from a managerial and policy perspective. The paper explores the nature of this process
evolution, the sources that stimulate this evolution, and several implications for research on techno-
logical innovation and productivity improvement.

Introduction
The innovation and intensive utilization of technology have been the keystones of
industrial vitality, the high standard of living, increased leisure and the competitive
advantage in world trade which the U.S. has enjoyed since the 1930s. Yet the U.S. seems
to have recently lost ground to other nations in the use of technology to generate
economic strength. A balance of trade is no longer assured by technological superiority,
economical atomic power has not proceeded rapidly enough to meet energy needs, and in
several mature but critical industries such as steel, other nations are nearing parity even in
direct labor productivity.
In principle, these problems can be solved in a straightforward manner by investing
more heavily in science and technology to achieve higher rates of new product innovation
and greater infusion of technology to improve productivity. This approach is based on the
widely held view that productivity improvement and the development of new kinds of

PHILLIP L. TOWNSEND has a B.S. in Economics from M.I.T. and an MS. in Chemical Engineer-
ing from Purdue University, and is currently completing a doctoral dissertation at Harvard Business
School. His recently active areas of academic research and industrial consulting have been technology
forecasting and the managerial implications of scale economics and experience curves.
WILLIAM J. ABERNATHY, Associate Professor of Business Administration, Harvard University,
Graduate School of Business Administration, is the coordinator for the school’s new field of research
and doctoral studies on the Management of Technology, and he is involved in teaching and research in
this new field, in the Production and Operations Management area and in the school’s Program for
Health Systems Management. He formerly held full time faculty appointments at Stanford University’s
Graduate School of Business and prior to this at UCLA. He was with General Dynamics/Electronics
prior to this where he had management and electronic systems engineering responsibilities. He holds
DBA and MBA degrees from Harvard University (1967 and 1964) and a B.S. in electrical engineering
from the University of Tennessee.

0 American Elsevier Publishing Company, Inc., 1974

380 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

Diffusion
Scientific Invention - Development - Innovation- (Broad
Discovery - Application)

FIG. 1. Linear model of technological advance.

goods and services can be stimulated through more rigorous action in creating new
scientific or engineering knowledge and in making this knowledge broadly accessible.
The underlying assumption, as illustrated by Fig. 1, is that a new scientific discovery
will trigger a linear sequence of events that in turn will result in a broad economic impact
such as that achieved with technology based innovations like computers, television,
magnetic tape recording, etc.
In practice, however, the problem is not nearly this simple. The assumptions of the
linear model are at best only partially relevant in explaining the relationship that exists
between scientific advances and economic results.
In traditional and basic industries, like steel, textiles, shoe manufacturing, construc-
tion, etc., technological advances have had little perceivable impact. Even in the so-called
“science based” industries, where scientific advance has obviously led to major new
economic development, no clear pattern is evident. In some instances, such as the
transistor, synthetic fibers, and antibiotics, essential theoretical results indeed led almost
directly to consequences of enormous societal value. In other cases, however, advances of
major potential economic relevance lie fallow until other enabling circumstances fit into
place. For example, the economic impact of scientific advances in wireless communica-
tions, nuclear energy, fuel cell technology and computers, etc., have lagged the essential
discoveries by a decade or more and have critically depended upon other enabling
circumstances.
The relationship between research and development (R&D) activity and productivity
improvement is also clouded. The kind of mixed relationships between R&D input and
productivity improvement that are obtained in practice are illustrated by the data given in
Table 1.

TABLE 1
Productivity and R&D Input; 1959-1969 Average

R&D Input Levelb

Annual Productivity’ (R&D Scientists
Improvement per 1,000 Employees)

Electronic Computing Equipment 8.6% High (3 5 and more)

Petroleum Refining 6.6% Moderate (15 to 35)
Railroad Transportation 6.3% Low (15 or less)
Paper Mills 4.5% Low (15 or less)
Motor Vehicles 4.1% Moderate (15 to 35)
Aircraft Engines and Parts 2.2% High (35 and more)

‘Source for annual productivity statistics is Business Week, May 13, 1972, p. 122.
bEstimates based on National Science Foundation data for the number of R&D
scientists and engineers per 1,000 employees for the major industry category that
includes the identified industry subsegment. Research and Development in Industry,
NSF 71-18, April, 1971.
TECHNOLOGY. PRODUCTIVITY AND PROCESS CHANGE 381

It is true that careful econometric analysis has shown a relationship between the level
of R&D inputs and productivity improvement in selected industries [7,9,15,18]. The
problem with these results, however, is that strong relationships only seem to be obtained
for industries that are already known to be achieving growth through technological
innovation, e.g., science-based industries like electronics and chemicals. The implication is
that technological inputs have the least impact where they are needed most, in mature or
stagnant industries. The difficulty is that limitations in data and the complexity of the
econometric methodologies involved severely limit the application of results to manage-
ment and policy decisions in either firms or the government. In short, the linear model
does not go far in explaining how technology can be managed to achieve economic or
social objectives. It is certainly valid in the sense that events do occur in the suggested
sequence. It is more misleading than useful as a tool in policy formulation and manage-
ment, however, since it accounts for only one sequence in the many causal factors that
shape outcomes. Unfortunately, other models that attempt to explain technological
outcomes on the basis of some different factor fare little better.
The purpose of the present paper is to propose a new model to clarify the factors that
facilitate or inhibit the successful application of new technological knowledge to improve
productivity, where productivity is defined to include the effectiveness of the good or
service that is produced as well as efficiency in the factors of production. The approach
differs from much prior work in two important respects: the characteristics of technology
users are seen as more important to the success of technological innovation than the
characteristics of technology creators (i.e., the problem is more on the side of utilization
and diffusion than in managing science and engineering), and that the factors which
critically enable innovation are best described as patterns of conditions rather than in
terms of single important variables.
The unit of analysis and the definition of the “technology user” in this paper is the
overall production process which is employed to create a product, whether the product is
goods or a service. More specifically, the unit of analysis is defined to include the physical
process (capital equipment, tasks, labor skills and process flow configuration), the
product, the characteristics of input materials and the characteristics of product demand
that are incident on the process. The term productive segment is used to describe these
included elements which can otherwise be conceived as the vertical span of a production
process that would typically be managed by the senior operating executive in an
organization. The essential idea underlying the model is that a process, or productive
segment, tends to evolve and change over time in a consistent and identifiable manner.
That is, as a given productive segment develops over time, it follows a predictable profile
that will be common among different industries.
The hypothesis is that the stage of development which a productive segment has
reached along this profile will determine its propensity to host particular types of
innovation. The most likely form of successful innovation, the typical barriers and
enabling conditions, and appropriate management skills all tend to depend upon the stage
of development which the recipient process has reached. A model which explains these
variations promises to reduce the number of variables and the complexity which the
decisionmaker must face in managing technological innovation. In addition, it will
provide a framework for transferring management expertise from one economic sector to
another.
The following section explores previous studies of successful technological innovation
as a starting point in tracing out the role of process and product innovation in stimulating
382 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

change within a productive segment. The results suggest that distinctions between new
product and new process innovation are largely irrelevant so far as overall productivity is
concerned. Product innovation, process innovation, and changes within a productive
segment appear to feed on one another, setting up the conditions for evolutionary
development. No single external force, such as market factors or technological factors, is
dominant in stimulating technological innovation. Sources of stimulation that arise within
a productive segment are more frequently the critical factor that sparks technological
innovation.
The second section of the paper considers the key forces within a productive segment
that stimulate or pace innovation and productivity improvement. Historical patterns of
development in several productive segments suggest that the efforts of engineers and
managers in improving production processes themselves may be a key factor in stimulat-
ing technological innovation.
The modes of process change and the stages of process development which result are
summarized as a descriptive model of process evolution, described in the conclusion of
this second section. The final section suggests some of the implications for management
and hypothesis for further study.

I. Characteristics of Innovation
The direct result from several studies of successful technological innovation [3, 161
indicate that most innovations are stimulated by market, as opposed to technical or
production considerations, and that they are most frequently applied to new products
rather than processes. Table 2 shows the results of one such study by Meyers and Marquis
[16]. This study reports data on 371 innovations drawn from two different segments of
the railroad and computer industries. As suggested, these data show that market factors
are the most frequent source of stimulation, involving 156 (42%) of the innovations
overall and that most innovations, 189 (51%), are applicable to products as opposed to
processes. Casual inspection of aggregate statistics in this way seems to imply that

TABLE 2

Successful Innovations’

Railroads Computer Industry

Components
Equipment Transportation and supply Computer
manufacturers companies manufacturers manufacturers Total

Number of Innovations 12s 19 17 90 371

Stimulation Source
Market 66 (53%) 15 (19%) 47 (61%) 28 (3 1%) 156
Production 2s (20%) 38 (48%) 17 (22%) 32 (36%) 112
Technical 28 (22%) 25 (32%) 10 (13%) 20 (22%) 83
Administrative 6 (5%) 1 (1%) 3 (4%) 10 (11%) 20
Application Impact
Product 80 (64%) 60 (78%) 49 (54%) 189
Component 32 (26%) 10 (13%) 28 (31%) 70
Process 13 (10%) 79 (100%) 7 (9%) 13 (14%) 112

aSource: Meyers and Marquis, Successful Industrial Innovations [ 161

TECHNOLOGY, PRODUCTIVITY AND PROCESS CHANGE 383

Rai1ro.d
Tran.port.tion

Railroad Suppliers I Sailroad - Transport.tion

* in additional Ii were stimulated by “admlnistrativc factors.”

** An additional j? were srim,,iated by “admini.trstivc fWtor8.”

FIG. 2. Technological innovation analysis: Railroads. (Taken from Tables 5 and 12, Meyers and
Marquis, Successful Industrial Innovations [ 161.

innovation is largely a new product concern with little implication for process change or
productivity improvement. The same data can, however, be viewed in a different perspec-
tive with considerably different implications.
Figures 2 and 3 present the same data in graphical perspective with the structure of
vertical integration within the two industries. The circles in the two figures represent the
incidence of innovation in each segment of the two industries. The dotted lines with
arrows leading into the circles indicate the source of stimulation, and the solid lines
emerging from the circles indicate the frequency and area of application impact. For
example, on the left side of Fig. 2, “market factors” represent the primary source of
stimulation for 53% of the innovations produced by the railroad supplies segment. Market
factors here, however, are the same as the process equipment needs of the next higher
process in the chain of vertical integration, the railroad transportation segment. The
dotted line connects sources and effect of stimulating factors to establish such equiv-
alency between segments.
Several characteristics of technological innovation in these two industries can be
derived from Table 2 and the schematic presentation of Figs. 2 and 3. First, the
predominant impact of innovation on productivity is “up stream.” Most innovations in
the lower levels of the vertical integration chain are product innovations. These are at the
same time process innovations for processes at higher levels of vertical integration and as
384 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

FIG. 3. Technological innovation analysis: Computers. (Taken from Tables 5 and 12, Meyers and
Marquis [ 161.

such have direct productivity implications. In fact, from a strict perspective of the process
at the highest level of vertical integration in both industries, all the innovations
considered here are process innovations and all have implications for process productivity.
Second, there is a definite shift in the immediate application impact of innovations,
from predominantly a product application impact for those originating early in the
integration chain to a higher frequency of process improvement application for those
originating at higher levels in the chain. This tendency probably results in part from the
characteristics of the particular industries considered. The processes themselves tend to
become more elaborated, highly developed, capital intensive and important as an element
of economic activity at higher levels of integration. Also both industries culminate in a
process that produces a service rather than a product. The implication, however, is that
the use of innovation shifts to serve the economic needs of the organizations that develop
them. The needs are in turn dependent upon the characteristics of the productive
segments itself.
Finally, when viewing one entire vertical process chain in aggregate, a different
perspective on sources of stimulation is obtained. For both computers and railroads the
innovations at the lower levels of vertical integration are predominantly stimulated by
market factors and are largely used in product applications. At this level, however, to be
market stimulated means to be stimulated by an opportunity to serve process needs of an
organization higher in the integration chain. Similarly, a product innovation at this level
must be accepted as a change in the process of an organization at higher levels. Therefore,
innovation by organizations in primary process stages is both predominantly stimulated
and paced by actual or anticipated opportunities for process change at higher process
TECHNOLOGY, PRODUCTIVITY AND PROCESS CHANGE 385

levels. Of course innovations may stimulate as well as be stimulated by opportunities for

process change at higher levels but only if organizations at higher levels are receptive to
innovative change. The net effect is an interlocked, closed-loop system of stimuli and
result that establish conditions for self-regulating process change.
If “within segment” sources of stimulation are netted out to reflect this intra segment
equivalency between market stimulation at lower levels and process change opportunities
at higher levels, a better picture of internally and externally generated sources of
stimulation can be obtained. For railroads as a whole 129, or 62% of all innovations, are
stimultaed by opportunities for process change, either within the originating segment or
in another process-related segment (25 + 38 t 66 = 129). Market factors external to the
process are the source of stimulation in 15 innovations (7% of all) and technical factors
account for 53 innovations (26% of all).
A somewhat similar picture is obtained for the computer industry. Of the 197
innovations, 96 (49% of all) are stimulated by process change opportunities, 28 (14%) are
stimulated by external market factors and 30 (15%) by new technical approaches.
The result is a surprisingly similar pattern in the two industries, even though they are
much different in character. In both, opportunities for process change within the chain of
vertically related processes provide the primary source of stimulation for innovation, with
twice the frequency of other sources. Technology is the most frequent source of external
stimulation, even though it is a poor second in terms of total frequency. External market
opportunities are third in significance. To summarize, the picture of innovation described
above has three salient features.

(1) Virtually all the technological innovations considered here (not 92 and 20 as
implied by narrowly defined data) are process innovations in terms of final industry
output and as such they represent increments to process technology and impact over-ah
productivity.’ While such a categorical statement cannot be made for other industries, it
will be more nearly true than is frequently acknowledged. Distinctions between product
and process innovations are moot in- respect to the issue of productivity impact. Most
industrial product innovations are destined to be an input to a process in the same or
another segment. As such they are also process innovations and must meet the test of
being accepted as a process change. Even though process innovation is sometimes
conceived as a slower, less spirited problem than new product introduction, the diffusion
of both are governed by the same sort of considerations.
(2) The primary stimulant for the innovations, at least in the present data, is an
opportunity or a mandate to introduce change in a productive process-either the process
the innovator uses to produce his own goods or that of his customers. Opportunities for
change at the output end of a process represent a source of stimulation to primary
segments, while innovations from primary segments may in turn stimulate change in later
stage processes, etc. A system of interactive forces is thus created that can cause process
change and innovation to feed on itself in a self-generating pattern. Approximately 60%
of the innovations in the present sample are linked in this way to process change as a
source of stimulation (63% in railroads and 49% in computers). Other sources of

’ The significance of this point is augmented by Arrow’s argument that most process improve-
ments are caused by “learning” in producing a product [ 21. If this argument is extended to the
present data, then 60% of the innovations are subject to this same source. This is in essence the
argument of the second section of this paper.
386 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

stimulation occur with only half this frequency. The important implication is that process
change must be treated as a principal variable in technological innovation.
(3) The force of new technology is a second order effect in stimulating innovation.
For the present data, only 61 of the 83 technically stimulated innovations were original,
so that of 371 only 17% are innovations that were stimulated by an original technical
solution in search of a problem (13% in computers and 19% in railroads).’ Of these,
many instances were very likely achieved with existent technology so that the real
contribution may often be assigned to creative problem solving rather than the stimula-
tion of advancing technology.

In total, the picture of technological advance which emerges is much different than
that inspired by the linear model of Fig. 1. In the principal mode, innovation does not of
itself proceed downstream from science to broad application and ultimate productivity
improvement. Science and technology is drawn into the picture, as needed, under the
stimulant of needs, and changes occur in the nexus of new product development and
process change. At the present time there is apparently a sufficient inventory of scientific
knowledge available to decouple technological innovation from a dependency on new
scientific advance. In recognizing that new science and technology is not the controlling
variable, the pressing need becomes one of identifying and delimiting the elements that
do pace process change so that the stock of available capabilities can be utilized.

II. Understanding Process Change as a Pacing Element

To better understand the sources that pace innovation and diffusion, it is instructive to
consider how the application of new technology has been related to process development
in specific examples. Evidence concerning the course of evolutionary development in a
number of diverse industries suggests that management’s effort in improving the produc-
tion process is itself a major pacing element that explains success over the long term.
Technology cannot be applied until the recipient production process has reached an
essential minimum level of definition and consistency. This is true whether the techno-
logical application is conceived as a process innovation from within the organization or as
a new product innovation by some external organization.
One particularly interesting example is the development of the electric light bulb
manufacturing process described in J. R. Bright’s book [6]. During a period from 1908 to
1955, the manufacturing process of electric light bulbs evolved from rather chaotic
conditions, manual operations and rudimentary process equipment, almost to complete
automation involving extensive process technology. Output per direct manhour increased
eighteenfold in this period. Although the application of much new potential technology
was important to the increased productivity, Bright found that these applications were
enabled by managerial changes in the process itself.
Bright describes a series of incremental changes which enabled the conversion of
potential technology to operational practice, First, during the early phases a series of
improvements were introduced that led the process from batch to continuous operation.
These were achieved by adopting systematic layout, standardized operations, and produc-
tion sequence changes. Second, the product underwent redesign a number of times to

* Appreciation is expressed to Associate Professor James Utterback of Indiana University, who

made a special computer tabulation of the Meyers and Marquis data to provide the statistics on
original versus adopted innovations.
TECHNOLOGY, PRODUCTIVITY AND PROCESS CHANGE 381

facilitate these process changes and permit mechanical manipulation. Third, individual
manual operations were replaced with direct mechanical analogies. Finally, automatic
work feeding, removal and material handling devices were used to integrate all machines
to form a highly automated process.
The important implication of this study is that management has a critical role in
causing the process to evolve and in readying it for technological innovation. Initially, the
process had to be rational before even simple technology could be effectively applied.
Then individual manual operations could be first mechanically assisted and then auto-
mated. Later in the period (1936 forward in Bright’s study) as process uncertainty or
ambiguity was reduced, the process equipment began to be designed as an integrated
whole-a systems development effort as it were.
Studies of other processes show that they have evolved with a strikingly similar
pattern. Early gains in automobile manufacturing resulted from a similar evolution in the
manufacturing process at Ford. The basic concepts here were not new. The use of
interchangeable parts and the assembly line were accepted practices by the time Ford
applied them to automobiles.3 They represented concepts of process organization rather
than scientific advances. They became the basis for numerous changes in management
practices and a stream of changes in many process operations, product design, material
inputs, etc., that made the automation of engine manufacture and automobile assembly
possible.
This same pattern of process development is evident in nonindustrial sectors as diverse
as medicine and health care delivery. As the physician’s technique and knowledge in new
fields like open-heart surgery are improved and the uncertainties and ambiguities in the
process begin to be resolved, then the prospects for further significant improvement
through innovation in capital embodied technology grow. Important technological inno-
vations here have followed practice, however, and attempts to promote progress in reverse
order have not been notable for their success. In several instances highly automated,
comprehensive systems, like automated multitest facilities and complete computer based
hospital information systems have been developed to promote substantial improvements
in the current practice of health care delivery. The impact and acceptance of these
technological innovations have been minimal, however, particularly when compared with
development cost. These patterns, like those in light bulb manufacture, demonstrate an
important role for new technology in supporting the development of important new
medical fields, but a limited and risky role in pacing improvement.
The coordinated pattern of product and process change, developed by example above,
is also evident in at least two independent fields of systematic research: the work done on
“operations technology” reported by Harvey [ 111 , Hickson et al. [ 121, and Woodward
[27] ; and studies of “the product life cycle” of international trade by Vernon [25,10],
Stobaugh [21] , and others. The research on “operations technology” based on studies of
several hundred firms in the U.S. and Great Britain shows that as a firm’s production
process becomes more highly integrated, and of a line flow nature, there are correspond-
ing broad changes in product and the firm’s organizational structure. For example,
Harvey’s [ 1 l] results show that with such changes the product beomces more stan-

3 For example, line operations were employed by both the 15th century arsenal at Venice, and
early meatpacking operations in Chicago. Eli Whitney used interchangeable parts in manufacturing
guns for John Adams. Ford made some contributions, apparently by combining and refining the two
concepts.
388 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

dardized and the product line less diverse. Hickson’s [12] work suggests that the use of
scientists and engineers (R&D input) also increases with increased integration of the
production process and scale increase.
Life cycle studies in international trade demonstrate a related pattern of change over
time. As a product becomes more standardized, then more efficient, less flexible produc-
tion technologies are chosen, and the basis of competition shifts increasingly from one
that emphasizes innovation in product performance features to one of minimizing cost.
These changes have important implications for the source and nature of technological
innovation.
Observations that may be drawn from the preceding examples and literature lead to
several hypotheses about the systematic properties of development in a productive
segment. Five basic concepts comprise the underpinnings for a descriptive model. First, as
costs are reduced and productivity is increased production processes evolve toward
greater predictability and more ordered or systematic relationships.4 A series of almost
imperceptible changes are involved even if some major process alterations are readily
apparent. Some productive segments remain essentially unchanged for long periods of
time, but rapid productivity improvement will be associated with a movement toward the
more systematic state.
Second, development requires consistent progress in four different factors: process
continuity and predictability, product improvement and standardization, process scale
and improved material inputs. Progress in each of these factors is apparent in light bulb
manufacturing, automobiles, the results from studies of operations technology, techno-
logical change in international life cycle studies, and less obviously in many service sector
examples. All contribute to advancing productivity. This not only implies an inter-
dependency between factors, but also as a third concept, that a minimum degree of
evenness in advance among productivity factors is needed. The product standardization
(any color Ford as long as it is black), the use of the assembly line to rationalize
production, division of labor, and increase in scale cannot be separated out as sufficient
conditions for Ford’s results. Change in all of these aspects went hand in hand.
Fourth, the innovation or utilization of product as well as process technology will be
heavily influenced by the level of development that has been attained within the
productive segment. This will also influence the origin and type of applicable innovation.
In early stages of a productive segment’s development, most technological applications
are to new products, if new technology is a significant factor at all. The innovation in
light bulbs was the light bulb itself. The important changes in the process at this stage are
to make it predictable and rational and technology is of little importance. As the
productive segment develops, technological inputs to process change are increasingly
enabled but the product design necessarily stabilizes. This in turn reduces the input of
new technology to the product. With further progression process innovations and applica-
tions become more technologically complex, costly, and ultimately become the vehicle
for linking the process into an integrated system. In the extreme the disrupture cost of
process change becomes the source of inhibition for product innovation. Many have
argued that this is the state which the automotive industry has now achieved.
Fifth and finally, the stream of changes within a productive segment builds cumula-
tively over time to signi’cantly alter its aggregate characteristics. Three definite stages can

4 While such progression is evident in many examples, it is systematically documented in Bright’s

[6] study and Hollander’s study of rayon manufacturer [ 131.
TECHNOLOGY. PRODUCTIVITY AND PROCESS CHANGE 389

be identified that are distinctly different from one another but common to productive
segments in the same phase of development, even though the segments are in different
industries. These stages are described here as unconnected, segmental and systemic. These
are named for the major structural properties of each stage, as summarized in Exhibit I.
The rows in this exhibit represent stages and entries under each column identify
characteristics of the four productivity factors. The differences among stages are pervasive
and beyond a purely technical structure have implications for the organization and the
innovative climate of the organization that manages the productive segment. The three
stages are discussed next and then the typical modes of change between stages are
identified.
Uncoordinated. Early, in the life of process and product, technical advances and
market expansion and redefinition result in frequent competitive improvements. The
rates of product and process changes are high and there is great product diversity among
competitors. The process itself typically is composed largely of unstandardized and
manual operations or operations that rely upon general purpose equipment. During this
state, the process is fluid, with loose and unsettled relationships between process ele-
ments. Such a system is “organic” and responds easily to environmental change, but
necessarily has “slack” and is “inefficient .”
Segmental. As an industry and its market matures, price competition becomes more
intense. Production systems, designed increasingly for efficiency, become mechanistic and
rigid. Tasks become more specialized and are subjected to more formal operating
controls. In terms of process, the production system tends to become elaborated and
tightly integrated through automation and process control. Some subprocesses may be
highly automated with process specific technology while others may still be essentially
manual or rely upon general purpose equipment. As a result, production processes in this
state will have a segmented quality. Such expensive development cannot occur until a
productive segment is mature enough to have sufficient product volume and at least a few
stable product designs.
Systemic. As a process becomes elaborated, increasingly integrated, and investment in
it becomes large, selective improvement of process elements becomes increasingly diffi-
cult. The process becomes so well integrated that changes become very costly since they
must consider the whole processing system. Process redesign comes more slowly but may
be spurred either by the development of a new technology or by a sudden or cumulative
shift in the requirements of the market environment. If changes are resisted as market and
technical forces continue to evolve, then the stage is set for revolutionary as opposed to
evolutionary change within a productive segment.

MODES OF PROCESS CHANGE

The righthand side of Exhibit 1 describes the type of process change activities that are
associated with movement from one stage to another: Process Rationalization, Systemic
Development and Process/Product Realignment. Change in the early stages is the easiest
to achieve.

(1) Process Rationalization. The transition from the Uncoordinated state is achieved
after standard methods of production are adopted, process flows are routinized and (in
general) part of the production process is rendered stable enough to justify greater
expenditure for more automated and special equipment. A predictable and rationalized
process will require that at least one product be produced in high volume with standard-
390 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND
TECHNOLOGY, PRODUCTIVITY AND PROCESS CHANGE 391

ized specifications, which is normally associated with a middle stage of the product life
cycle. The process will come to have islands of automation interspersed with manual
operations. Innovations will increasingly result from the development of new technology.
Suppliers develop a more continuous focus on the productive segment and on alternative
technologies which might improve the process. Suppliers increasingly develop internal
capability to innovate technology and increasingly initiate innovations in the processes of
customers.
(2) Systemic Technological Development. A combination of process rationalization,
stabilization in product design, and application of increasingly advanced technology must
render the entire process much more stable before systemic development may occur.
There is high emphasis on process engineering, including issues of inventory control,
process balance and optimal equipment selection. Eventually, technological advances
need to be treated as systems development problems. A specialized equipment supplier
industry may form and be sustained in larger industries to conduct the development of
process-specific equipment, since development costs are then amortized over more instal-
lations. This contrasts with the earlier stages where the process equipment suppliers will
provide or adapt general-purpose (nonprocess-specific) equipment. Relationships with
suppliers may also change. Unmanageable or difficult-to-automate process tasks will tend
to be separated out of the main process and often turned over to feeder contractors.
Considerable control is also gained over material suppliers so that inputs can be optimized
to the needs of the process.
(3) Process/Product Realignment. As a process evolves through its stages, productiv-
ity gains result from changes in the four factors-material inputs improvements, changed
process technology and labor skills, larger process scale, and a tailoring of product
characteristics. The development will continue as long as the productive segment faces a
similar environment. But what if a productive segment becomes stalled in any one stage?
If no movement occurs because progression in one of the four productivity factors is
barred or because management is ineffective, then productivity gains can be expected to
diminish or even become negative.
A dilemma is apparent. Continued evolution toward a systemic state offers the
benefits of high productivity but only at the cost of decreased flexibility and innovative
capability. A productive segment places itself in a very vulnerable position when it
achieves a systemic state and may then face limits to further development through
blockage of one or more of the productivity factors. It must face competition from:
innovative products that are produced by other flexible segments that are more capable
of substituting products; foreign imports; competing products from other industries with
high cross-elasticity of demand; or process changes by customers to eliminate the product
directly. Once such a state is reached and further development becomes blocked then
stagnation or death of the productive segment may be threatened. This is essentially the
problem that faced Western Union, around the turn of the century, when as the giant of
the communications industry it decided to shun the innovations in voice communications
and specialize itself in efficient telegraphic operations. It is also reflected in Ford’s
position following the demise of the Model T in 1926, and in that of the manual
typewriter producers in the 1940’s and 1950’s, etc.
There are choices to be made other than for the management of a segment to accept
economic demise. First, firms may elect to achieve lower costs by moving production to a
foreign country. In fact, cases of foreign direct investment which are motivated by
efficiency (most are not) may be viewed largely in just this way-as competitive reaction
392 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

to approaching stagnation of a productive segment. For example, establishment of foreign

shoe production operations by U.S. companies at least partially reflects the inability of
the shoe industry to move toward more productive stages of process development.
Alternatively, a decision may be made to backtrack along the traditional course of
evolutionary process development to a more flexible state with greater capability for
innovation but less process integration, efficiency and capital intensity. This essentially is
the course of action that major productive segments in the automotive industry followed
after the end of the Ford Model T era in 1926. The assembly, engine and body plants at
Ford reached their highest level of integration during the Model T era when the basic
vehicle was standardized and the price was brought down to $890 (in 1958 dollars) and
thru put time from iron ore receipt to vehicle delivery was reported to be as short as four
days. Thereafter vehicle performance was increased together with price and the major
process segments such as engine and assembly operations were more decoupled one from
the other.’
This choice is a major strategic long range decision that management must make either
implicitly or hopefully in an explicit manner that recognizes advantages and con-
sequences. In effect, it is a trade-off decision between the capability for innovation and
productivity improvement. Whether the firm chooses its response or is forced to react,
environmental change will eventually cause most productive segments to choose among:
reverting to an earlier stage of the process life cycle, or “escaping” by international
migration, or its own economic death.

III. Implications for Management and Research

The proposed model of process evolution should be viewed as a new theory about
process change and its implication for innovation. The five concepts that were introduced
in section II above represent the essential hypotheses of this model.
The oasis for extending the essentially descriptive model of process evolution (as in
Exhibit 1) to one that is measurable and replicable, is demonstrated by the explicit
methodological approach developed and applied by Harvey [I I] and Woodward [27].
Such a model promises a method that can be used to assess the characteristics of a given
productive segment in a way that can usefully clarify its capability‘for innovative activity
and productivity improvement. Research is currently underway to develop the process
model in this way and to test its implications for innovation and productivity improve-
ment .
The implications for managers and policymakers fall into three major categories:
managing technological innovation, improving the productivity within a productive seg-
ment itself, and the use of the model as a framework for formulating policy in respect to
both technology and productivity.
Several important issues in managing technological innovation are addressed by the
model: the natural locus of innovation (or the most potentially fruitful source); the most
appropriate type of innovation; and the array of barriers to innovation.

(1) The locus of innovation shifts with the stage of development.6 During the
unconnected stage in the development of a process, innovative insight comes from those
individuals or organizations that are intimately familiar with the recipient process, rather

5 The implications of this period at Ford are more fully developed in [ 1 ] which is forthcoming in
the Harvard Business Review under a different title.
’ A number of studies that have examined changes in the pattern of innovation over time indicate
TECHNOLOGY, PRODUCTIVITY AND PROCESS CHANGE 393

than those intimately familiar with new technologies. The critical input is not state-of-
the-art technology but new insights about the need. Later, in the systemic stage, needs are
well defined, “system like”, and easily articulated. These needs7 lend themselves to
complex technological solutions, and the innovator will frequently be one that brings new
technological insights to the problem. This may be a formal engineering or R&D group,
an equipment company, or some other external source. In undertaking managerial action
to stimulate innovation, it is important to appreciate these distinctions so that the most
likely sources of innovation can be identified, nurtured, and supported.
(2) The type of innovation that is likely to succeed, whether systems like and
technologically complex or simple, and whether applied to product or process, also
depends upon the stage of development. During the unconnected state most technological
applications are to the products that the productive segment will produce. Few are to
process improvement, and those that do occur tend to be simple in application and to
address single needs. Complex technological systems of process equipment do not “take”
well when the recipient process is ill defined, uncertain and unstructured. Systems
technology has not been very successfully applied to solve ill defined process needs. The
converse is true in the systemic stage. In a symmetric fashion, isolated radical innovations,
of even major significance, seldom gain ready acceptance when the recipient productive
segment is in the systemic stage. Like many of the theoretical concepts which seem to
have application to pollution problems in automobile engines, most have effects that are
too disruptive even though they promise high merit as a product improvement. The
seemingly isolated innovation must in reality be incorporated as change throughout the
systemic productive segment. A realistic assessment of the type of innovations that will
be successful, and how it should be introduced, depends upon an understanding of the
productive process that will receive it.
(3) The total array of barriers to an innovation, like the appropriate type of innova-
tion, changes composition with the stage of development. In the unconnected stage,
resistance centers around perceptions of irrelevance. Will the innovation work and meet a
need? In the systemic state resistance stems from the disruptive nature of innovation. Will
it displace vested interest and disrupt current practice? In both forms, resistance radiates
into many different dimensions, labor practices and union positions, legislation, product
acceptance, etc., but the model helps to clarify the changing nature of these barriers.

The implications of the model for improving productivity follow rather directly from
the five concepts that were defined earlier. Two are noted below to suggest its applica-
tion.
First, the major factors in process change (product and product characteristics shown
as columns in Exhibit 1) need to advance with a minimum degree of evenness. To the
extent that one or more factors are lagging, the risk increases and the return from
improvement in other factors decreases. The model directs attention to a balanced
consideration of all the productivity factors when change in any one is considered,
thereby helping to uncover inconsistencies.8

a dramatic shift in locus. Included among these are Knight’s study of computers [ 141 and Clarke’s
study of rocket engine innovations [ 81.
7 This point is the essential conclusion that is implied by numerous studies which show that
“need” is the principal stimulant of innovation. See Hindsight [ 201 or Williams [26].
a The major study which provides the most direct evidence of this pacing is Project Hindsight
[20]. Most breakthroughs resulted from the decision to build the system to which they contributed.
 394 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

TABLE 3

Industry Productivity Patterns

High Productivity Industries Some Comparative Low Productivity Industries

Urrconnected stage

Livestock and livestock products’ I‘orestry and fishery product?

Real estate and rental’ Medical and educational services’
Mining (ferrous ore)b Mining (copper)b

Air transportb New construction”

Radio and television receiver eq.’ Aircraft and parts’
Malt liquorsb Iron foundries’
Agricultural products’
Hosieryh l,‘ootwear’
Coal mining

Systemic sta,qe

Electronic componentsa
Office & computing machinerya
Petroleum refining Glass and glass productsb
Electric and pas utilitiesb Steelb
Communicationsa
Aluminum rolling and drawing0 Primary copperb

Product/process realignment

Railroad transportation’
Miscellaneous textiles
and floor coverings’

‘Industry definitions correspond to the input-output model of the economy. Produc-

tivity improvement rates are projections for the 1965-1980 period 3s reported by the
U.S. Dept. of Labor, Patterns of‘ Economic Growth [ 23 1. Industries with high rates of
productivity improvement (left side) are those with annual improvements above 3.6%.
Thslow category (right side) corresponds to rates below 2.5%.
Industry definitions correspond to SIC standards. Productivity improvement is the
annual average during the 1960-1971 period. The high category represents a rate of
improvement of 5%% or better (left side). The low category is below 2.5%‘. The source is
1241.

Second, a factor has an order ofprogression. A change which requires a reversal in the
natural order of progression in any particular factor will have implications for the
segment. For example, changes in the structure of manual tasks in a process may be
sought to enrich job content and job satisfaction to workers. If such changes lead to less
specialization, more discretion, and a general loss in the rationalization of major pro-
cesses, then these changes may represent a general reversal in the direction of process
evolution. As such they may thwart long range productivity improvement. The model
provides a basis for planning enrichment programs so that they need not conflict with
plans to improve productivity in the long run.
TECHNOLOGY, PRODUCTIVITY AND PROCESS CHANGE 395

Finally, as a framework for policy evaluation, the model provides a new basis for
strategy choice. It points out the course of action that can be pursued over both the short
and long term as a strategy for continuing productivity improvement. At the same time it
describes the consequences of following such a strategy to its ultimate conclusion. These
consequences include the loss of adaptability to external changes in needs and a reduction
in innovativeness. If these latter capabilities are important, then the best choice may be
an intentional decision to slow or reverse evolutionary progress or to remain in that
particular stage which offers the best trade-off between conflicting objectives. There is
reason to believe that in practice progress may frequently be halted for long periods and
then continued.
Some support for the hypothesized relationships between productivity improvement
and production process structure is available from industry statistics. Although the
elements included within a productive segment do not correspond well with the broad
industry categories that are used in national economic statistics, there is some similarity
among many segments in one industry. Table 3 lists several industries that are distin-
guished by their rates of productivity growth as defined by two different sources of
industry statistics. The list includes some with very high and some with very low
improvement rates. Those with high rates of productivity improvement are listed on the
lefthand side of Table 3. An effort has been made in this table to roughly and somewhat
arbitrarily sort these into the three stages on the basis of the definition that would best
describe the major productive segments included within the industry. A similar procedure
was followed on the righthand side with industries that have low rates of productivity
growth. Although industry data are slippery, the resulting pattern suggests that industries
with high rates of productivity improvement are those whose major included productive
segments are progressing most rapidly from stage to stage. The industries with low rates
of productivity improvement, on the other hand, seem to include segments that are best
characterized as stalled in one stage of development, often for many years. The low
productivity industries do not necessarily have the lowest rates of innovation. They do,
however, apparently host a low rate of process innovation.
An interpretation of the data in Table 3 is speculative, but it does suggest some
important policy considerations. High rates of productivity are achieved in periods of
rapid process change in an industry. The decision to pursue a strategy of productivity
improvement need not be followed to its ultimate extreme, the systemic state. Each
movement that is undertaken should be made in full recognition of the implicit trade-off
that is involved-flexibility and innovative capability versus productivity improvement.

References

1. W.J. Abernathy and K. Wayne, Limits of the Learning Curve: Harvard Business Review (Sept.-Oct.
1974), pp. 109-119.
2. K. H. Arrow, The Economic Implications of Learning by Doing, Rev. Econ. Studies 29(3), No. 80
(June 1962), pp. 155-173.
3. N. R. Baker et al., The Effects of Perceived Needs and Means on the Generation of Ideas for
Industrial R&D Projects,lEEE Pans. Eng. Management 14(4) (Dee. 1967), pp. 156-163.
4. R. M. Barnes,Motion and Time Study, John Wiley, New York, 1968.
5. M. Bldug, A Survey of the Theory of Process Innovations, Econometrica (Feb. 1963), pp. 13-32.
6. J. R. Bright, Automation and Management, Harvard University, Graduate School of Business
Administration, 1958.
7. M. Brown and A. H. Conrad, The Influence of Research and Education on CES Production
396 WILLIAM J. ABERNATHY AND PHILLIP L. TOWNSEND

Relations, The Theory and Empirical Analysis of Production, M. Brown (Ed.), NBER Columbia
University Press, 1967, pp. 341-394.
8. R. Clarke, Innovation in Liquid Propelled Rocket Engines, Ph.D. dissertation, Stanford University,
Graduate School of Business, 1968.
9. H. C. Grabowski, The Determinants and Effects of Industrial Research and Development, Econo-
metrica Research Program Memo 82, Princeton University, 1966.
10. W. Gruber, D. Menta, and R. Vernon, The R&D Factor in International Investment of United
States Industries,J. Pol. Economy, 75(l) (Feb. 1967).
11. E. Harvey, Technology and the Structure of Organizations,Amer. Social. Rev. 33(2) (April 1968).
12. D. J. Hickson et al., Operations Technology and Organization Structure: An Empirical Reap-
praisal, Admin. Sci. Quart. 14(3) (Sept. 1969), pp. 378-397.
13. S. Hollander, Sources of Increased Efficiency: A Study of Du Pont Rayon Plants, Cambridge,
Mass., MIT Press, 1965.
14. Kenneth E. Knight, A Study of Technological Innovation-The Evolution of Digital Computers,
unpublished Ph.D. dissertation, Carnegie Institute of Technology, 1963.
15. Edwin Mansfield, Contribution of R&D to Economic Growth in the United States, Science 157
(Feb. 1972), pp. 477485.
16. S. Meyers and D. Marquis. Successful Industrial Innovations, National Science Foundation, 1969;
Report No. 69-17.
17. R. E. Miller, Innovation, Organization and Environment, University of Quebec, 1971.
18. Jora R. Minasian, The Economics of Research and Development, The Rate and Direction of
Inventive Activity, NBER, Princeton University Press, 1962, pp. 93-141.
19. R. P. Olsen, “A History of Automation in Manufacturing and Its Implication for the Service
Industires,” a paper prepared for the Committee on Automation Opportunities in the Service
Industries (unpublished), Harvard University, Graduate School of Business, 1972.
20. C. W. Sherwin et al., First Interim Report on Project Hindsight, U.S. Department of Defense,
Washington, D.C., June 30, 1966.
21. R. B. Stobaugh, How Investment Abroad Creates Jobs at Home, Harvard Bus. Rev. (Sept.-Oct.
1972), pp. 118-126.
22. Technology Retrospect and Critical Events in Science (TRACES), Illinois Institute of Technology
(NSF Contract NSF C535), Dec. 15, 1968.
23. U.S. Bureau of Labor Statistics, Patterns of U.S. Economic Growth, BLS Bulletin 1672, 1970.
24. U.S. Bureau of Labor Statistics, Indexes of Output per Man Hour: Selected Industries, BLS
Bulletin 1758, 1972.
25. R. Vernon, International Investment and International Trade in the Product Cycle, Quart. J. Icon.
(May 1966).
26. B. R. Williams, Technology Investment and Growth, Chapman & Hall, London, 1967.
27. Joan Woodward, Management and Technology, Tavistock Institute, London, 1958.

Received April 5, 1974, revised June 28, 19 74