THE COORDINATION OF GLOBAL MANUFACTURING
Table of Contents
Appears in "Globalization, Technology and Competition:
The Fusion of Computers and Telecommunications", Edited by Stephen
P. Bradley, Jerry A. Hausman and Richard L. Nolan, Harvard Business School
Press, Boston, MA., 1993.
Graduate School of Business Administration,
Harvard University, Boston, MA 02163
Production capacity is now sufficiently flexible in some industries, to
be viewed as a commodity. Technological change has raised the prospect
of global markets for a variety of types of flexible manufacturing capacity.
This paper outlines technological and commercial conditions under which
markets for flexible manufacturing capacity are likely to arise, describes
an industry in which a capacity market exists, and explores desiderata
for such markets.
Conditions for Global Commodity Capacity Markets
Important changes in manufacturing technology have occurred in many industries.
Machines for many processes are now very flexible, in that they may be
programmed to perform a wide variety of manufacturing tasks and are able
to accommodate diverse product characteristics while providing both high
quality and low cost. Although the products manufactured by such technology
are highly differentiated, the broad diffusion and standardization of the
technology has made productive capacity comparatively common and indistinguishable.
Indeed, for many highly differentiated products (metal-machined parts or
socks for example), the precise source of the product is becoming much
less relevant since the technology of production allows consistent quality
and cost performance regardless of source. This suggests a novel situation
--- one in which flexible capacity may be seen as a commodity ---
inasmuch as units of flexible capacity are comparatively lacking in distinguishing
qualities. The large number of product variations which such capacity may
effectively produce mean that the primary input for the process is information,
in the form of a well-codified, often computer readable product descriptions.
Advances in global telecommunication mean that such information may be
effortlessly transmitted between customer and source, making the precise
location of the manufacturer progressively less relevant.
Diffusion of the supporting manufacturing technology to an ever-growing
range of processes, along with advances in telecommunication have had far-reaching
consequences, from the shop-floor to the global structure of industries.
We are seeing fundamental change in such basic elements of production as
firm size, the nature and form of product and process specifications, transaction
processing, and labor.
Why should factories be large? Companies have traditionally collected the
machinery of production into large factories for two reasons:
Technological economies of scale have never been strong in industries that
produce highly differentiated products: machines in such industries are
gathered together primarily to share overhead costs. With the new technologies,
even this source of pressure to aggregate has been greatly relieved.
to gain economies of scale;
to share overhead costs, such as maintenance, quality assurance, materials
tracking and so forth.
Today's machines are highly reliable and require very little maintenance.
What little is needed may be performed by operators. Because machining
processes are under programmable computer control and are, in general,
well understood and predictable, machines produce good quality output time
after time, eliminating the need for separate quality departments to weed
out defects. Finally, technological solutions make overhead due to information
and material flow very small, regardless of the size of the operation.
Minimum efficient scale for a modern manufacturing operation in many industries
is a manufacturing cell of about six machines and fewer than a dozen people.
Such a cell functions as a factory within a factory, effecting an entire
production process under computer control, often including materials handling
Product and Process Specifications
With computer-based manufacturing technologies, product and process specifications
exist as computational procedures developed on specialized computer-aided
design (CAD) systems. These procedures are transportable, via standardized
telecommunication links, to the machine controllers that govern the manufacturing
process. Moreover, to the extent that the people who write them are able
to anticipate and solve every possible contingency, these procedures guarantee
precise reproducibility, that is, every part made by any machine running
a particular procedure will be exactly the same. These two characteristics
of modern product and process descriptions---transportability and precise
reproducibility---reduce the need in many industries to collocate engineering
and design with manufacturing, except for pilot production. The standardization
and predictability of the link between design and manufacturing weakens
traditional dogma that insists that design engineers be on hand during
volume manufacturing, to make design trade-offs as manufacturing problems
occur. Technologies in many industries have solved this problem at its
root---by anticipating and eliminating such problems systematically rather
than on an ad hoc basis, allowing engineers to be absent from volume production
and rendering the location of volume manufacturing less relevant. Indeed,
manufacturing units need not necessarily even have their own engineering
functions; in the context that is evolving, engineering and design can
be effectively supplied by physically and functionally distinct organizations.
A certain economy of scale to which firms do have access in the new manufacturing
context is through sales and distribution. With the right communication
links, one marketing department can today serve the world.
Figure 1 Traditional Firm to Market
The traditional firm to market scenario is depicted in Figure
1, the electronic marketplace in Figure
Figure 2 Electronic Marketplace
In the latter, we envision direct links between individual facilities
of a firm and the buyers of their products. Contracts in the electronic
marketplace would likely be many and small, and the market system would
learn through repeated transactions, enabling it, over time, to arrange
almost universally ideal contracts. The speed of the system would be such
that contracts would be renegotiated dynamically in the case of non-fulfillment,
utilizing surge capacity among the manufacturers. Such a system minimizes
transaction costs by effectively automating transactions. The telecommunications
basis for such a system exists. All that is required is sufficient speed
and memory, and reliability. If the memory is structured well, the market
functions could be performed without human participation.
The existence of standard flexible technologies has decreased the need
for firm-specific training. A number of industries may now draw from a
pool of labor whose skill level is lower and more homogeneous. This, in
turn, means lower costs for centralized training and personnel functions.
For example, with standardized CNC machine tool technology, a person trained
in CNC milling may very quickly learn CNC turning.
Collectively, these factors serve to substantially de-emphasize economies
of scale and reduce absolute cost outlay at the plant level. Manufacturing
concerns can now establish small, independent cells that operate effectively
and economically with only a modest capital investment. Moreover, these
small units of flexible capacity can be physically and organizationally
separate from design, marketing, and engineering. Small minimum efficient
scale, low capital requirement, and separability of volume manufacturing
operations, by effectively lowering entry barriers, ensures the prospect
of ample players in a market for flexible capacity.
Advantages of a capacity market.
There are a number of distinct advantages to organizing an industry's manufacturing
network so that providers of flexible capacity compete, among them the
The pooling effect of the market better insulates capacity buyers from
fluctuations in demand. Since products are diversified, it will
often be the case that when one capacity purchaser's demand is high (for
his/her particular product) another's may be low. The pooling effect enables
capacity purchasers to take advantage of the disparate temporal requirement
for flexible capacity and avoid the cost of capacity constrained operation
and low asset utilization that often face firms to which capacity is dedicated.
Poor capacity performance can be combated more quickly by switching supplier
than through the slower process of improving internal operations.
Managerial costs of coordinating and balancing capacity are avoided since
market mechanisms perform many of these functions. For example, the difficult
task of assigning capacity to the highest priority task can be performed
by the price mechanism--- when industry capacity is tight, it goes to the
highest bidder. This avoids the cost of bureaucratic internal groups having
to inefficiently juggle jobs in order of apparent priority, without information
that would enable them to reflect priorities accurately.
Capacity can be bought in the short term, providing the buyer faced with
a high degree of uncertainty access to capacity without committing to its
The factors listed above will be most beneficial when demand is highly
uncertain, the managerial costs of coordinating proprietary capacity are
high, and transactions are small and great in number. Provided the technological
conditions described above prevail, we would expect to see markets for
flexible capacity develop in industries that face such commercial conditions.
An Example: The Disintegration of the Textile Industry
Dating from the 14th century, the textile industry in Prato, Italy has
for centuries been the economic backbone of the Florence and Pistoia regions.
Once, armies of artisans carded and dyed, spun and weaved. But with the
technological changes that precipitated the Industrial Revolution, which
brought increased economies of scale to the industry, firms grew in size
and vertically integrated so as to be able to schedule and balance capacity
in these various process steps. Because starving assets of input materials
incurred substantial penalties, given the heavy required investment in
production technology, great advantage accrued to coordinating the process
steps directly to avoid such circumstances. Production equipment for the
various process steps was collocated to facilitate coordination and the
quick resolution of inter-process problems, and central marketing and design
departments were maintained on site to work with production and match customers
to production capacity and capabilities.
Rediscovery of Small Economic Scale
Most Prato mills were integrated in this way in the early 1970s, with fiber
production, dyeing, spinning, and weaving performed within the same company.
But many of these companies had progressively become unprofitable. Lower
market prices, global competition, and rising internal costs had gnawed
margins to the bone. Meanwhile, new dyeing and finishing techniques were
becoming available and the market was demanding an ever-broader range of
products from these new methods.
Some mill owners recognized that their integrated mills were an
encumbrance in the new regime. The processes had become so well understood,
and hence specifiable, that the various steps were largely independent
of one another, yet the flexibility of the individual processes remained
constrained by the particular output of the upstream, and requirements
of the downstream steps. Capacity, such as weaving or spinning, had become
cheaper and economical in much smaller units, eroding economies of scale.
The increasing overhead burden and need to effectively coordinate production
of the broader product ranges and take advantage of new-found flexibility
in each of the process steps led firms to look very closely at the manufacturing
structure that had been the industry paradigm for a century. In the face
of extinction, firms began to change.
Many mills followed the example of the Menichetti family, which
broke its mill into eight separate companies, one a realty company that
leased space and services to the rest. As much as 50% of the stock in these
companies, financed through profits, was transferred to employees. To ensure
competitiveness, Menichetti insisted that each company find 50% of existing
business outside the original business. At the same time, he established
a New York-based marketing company to create new designs and match product
with the best producer. This company was to provide no more than 30% of
the business of any company in the Menichetti fold.
Within three years, all units of the disintegrated Menichetti
mill were running at 90% utilization, product variety had increased ten-fold,
average in-process inventory been reduced from 4 months to 15 days, and
attrition had reduced the labor force by a third while production had risen
by 25% (largely due to the satellite firms investing in new technology).
By 1980, all but one of the Prato mills had undergone similar disintegration,
turning a sluggish, threatened industry into a thriving community of innovative,
flexible companies, each a world class competitor. This process continued
throughout the value chain (see Figure
3 Firms in the Prato Network by 1975
The predicament of textiles, which found itself on the brink of an important,
global change, is by no means unique. Disintegration may have progressed
further in textiles than in other industries, but the conditions outlined
at the beginning of this chapter apply to a growing enclave of industries.
Very similar circumstances prevail in small-batch metal machining, for
example, with the flexibility of the machining cell and its small economic
scale.1 But the lack of
standardization of part programs, some remaining machine-tool specialization,
and the indifference of machine-tool producers to small manufacturers have
so far prevented the production of part-programmed machined products from
following the example of textiles.2
Network Coordination and the Modern-day Impannatori3
The key to the success of the Prato system lays in the role of the modern-day
impannatori. A throwback to medieval times, these agents provide
central brokerage for the firms in the network, of which there are now
between 15 and 20 thousand employing some 70,000 people. Today, several
hundred such brokers draw from a hierarchical network of these thousands
of suppliers. Brokers' thorough knowledge of the capacity, capability,
and loading of each of the producers loosely collected in their folds enables
them to source production for customers, find customers for spare production
capacity, and intermediate in the negotiation process.
Effective management of this complex information set, coupled with trustworthiness
and honesty, are the hallmarks of the successful impannatori. Indeed, it
is this trustworthiness that avoids the problem that result from contracts
that are difficult to specify---the trust ensures that what is needed is
provided without the need for legal specification. Many specifications
are thus based on a tacit understanding of industry standards. The complexity
of the capacity assignment problem is rendered manageable by the autonomy
of the various actors in the system, who are able to concentrate on being
effective in their specialties while contributing to the performance of
the network as a whole. Assignments of capacity are made through the market
mechanism and the impannatori, enabling the system to avail itself of the
most appropriate vendor and thus the full flexibility of the market for
each element in the value chain.
Globalization and Communication
Among the technological changes dramatically expanding and changing the
textile industry is codification. Today, cloth required by the world market
may be uniquely specified with a code of 50 digits. Computer-aided design
systems permit rapid local prototyping of a fabric before sourcing to a
volume producer. Most importantly, computers facilitate an electronic marketplace
in which the complexity of the production hierarchy can be "managed." Global
telecommunications and continued disintegration suggest the possibility
of trading options on both products and production capacity. For example,
a fashion manufacturer uncertain of the season's demand for a currently
imprecisely-specified product may ensure against a lack of supply by buying
an option on the use of flexible capacity. Today, at least towards the
end of the value chain, it is capacity, not product, that is a commodity.
Units of flexible capacity are now relatively indistinguishable from one
another and have the capability of producing myriad products. The products
themselves, characterized by variety and customization to a particular
fashion, and not at all commodity-like.
Manufacturing and Negotiation
Integration of advanced telecommunications and information systems holds
promise for automating more fully the negotiation process in the Prato
textile industry by speeding information flows and allowing requirements
to be matched more quickly to supply. The extreme complexity of information
flows, given the plurality of operators acting at different hierarchical
levels (the top level comprising hundreds of Impannatori splitting control
to thousands of suppliers and manufacturers), makes control of an automated
negotiation system highly strategic. Such a system would be capable of
integrating single elements of the network, support real-time monitoring
of the entire negotiation process and its related services, and provide
the necessary control to achieve optimization. It would enable the artisans
and subcontractors, the largest group in the Prato system, to "see" the
market, to discern market trends and review other suppliers' capacities
in order to be able to react quickly to market demands. Access to such
information would greatly stabilize the activities of the many small firms.
Conditions for the effective marshalling of resources in manufacturing
systems that are coordinated using negotiation methods are currently of
interest at many levels. As the flexibility of manufacturing elements increases
and effective units become smaller and more independent, it becomes increasingly
advantageous to permit entities to negotiate with one another.
Upton describes a negotiation system that functions within a plant.4,5
In this system, the partly completed product (such as a raw casting) is
provided with a miniature manufacturing computer physically attached to
it. This computer uses artificial intelligence techniques to negotiate
the manufacture of the product (step by step) with the various processing
entities in the system, such as transport vehicles and machine tools. Machines
bid for the right to provide processing for the product and the product
selects the best bid at each step.6
Bidding machines take into account their prevailing workloads, commitments
and capabilities. After successfully visiting all the necessary stations,
the semi-finished product (which might be a component for a large earth-mover,
say) relinquishes its computer so that subsequent products may avail themselves
of its experience, such as knowledge about unreliable performers and optimistic
bidders. Thus, the system slowly builds expertise about itself in the product
stream. This system is able to adapt easily to the removal and addition
of machines, since removed machines stop bidding and new machines are simply
told to start. The technique solves many of the problems of centralized
computer-control in dynamic manufacturing systems.
At the intra-company level, Jaikumar has considered the optimal
behavior of users and providers of processing capability within the firm.7
He describes a negotiation system in which a firms' sales agents are the
buyers of capacity and the production resource managers the providers.
Jaikumar shows that such a decentralized system can both be optimal for
the firm and provide an efficient incentive system. At the level of global
coordination, we are most interested in exploring how a global, decentralized
negotiation system might best be constructed for manufacturing.
Issues in System Design and Objectives
A number of factors must be carefully considered in attempting to establish
a structure for a Prato-like market, among them: reliability; management
information; brokerage; and commodity-like transactions.
Capacity must be reliable and there must be effective mechanisms for ensuring
that unreliable suppliers discount appropriately. Information on the performance
of previous contracts will enable the market to take into account both
the quality and reliability of suppliers. The credit-worthiness of buyers
must similarly be assured. How might such information be promulgated and
what recourse provided to suppliers and buyers for correcting inaccuracies?
For example, suppliers might be required to specify the proportion of the
last hundred contracts on which they were late, or in which there was a
dispute concerning quality.
There will inevitably be transactions in which one party is aggrieved,
and assigns too much importance to one egregious event. For example, a
firm may suffer badly because of one instance of failure and feel the need
to take some punitive measure. Of course, the consequences of a party's
actions are not relevant in determining its on-going performance, so the
market should ensure that these kinds of occurrences are accommodated.
What information does a supplier need in order to compete effectively in
an automated market? Temporary differences in cost of capacity due to scheduling
constraints are inevitable. Manufacturers currently making pink T-shirts,
given sufficient demand, would very much like to continue doing so to avoid
changeover costs; where should they seek buyers of such temporarily cheap
What internal information about changeover costs does a firm need?
Decisions about what price to bid on a job rely on timely and accurate
internal information. The advantage will go to players that are able to
reliably predict their own performance for the purpose of determining their
own bids as well as to ensure satisfactory acquittal of the contract.
Under what conditions is it advantageous to use brokers of capacity? Clearly
if communications can be organized effectively in a distributed fashion,
the need for a centralized hub is reduced. With the ability to transmit
and receive information throughout a global network, the role of a central
broker as a channel for information is less clear. Individual firms could
begin to access the network to determine customer requirements and bid
directly on jobs as they arise. Users of capacity could themselves post
requirements on the network (for products as well as for capacity options
Methods for limiting damage arising from non-performance of contracts are
essential. Some capacity providers will be able to provide insurance by
maintaining spare capacity. A futures and options market in capacity would
provide a hedge against increases in price, for example. Whereas product
variety has previously limited such deals to commodity products, given
the flexible capacity to produce commodity-traded products we will expect
to see commodity-like transactions.
Moreover, we need to explore the various methods by which different
forms of capacity might be converted into financial instruments. Such forms
would include futures markets and spot markets as well as options. The
insurance of these instruments is also of interest, as is the entry and
modus operandi of third parties into bilateral transactions, which might
have a considerable effect.
The Growing Arena
The application of programmable computer control is likely to continue
to broaden and increase in sophistication. As the physics of more and more
processes becomes well-understood, it is becoming possible to automate
them and allow a computer to control precisely the variants of the items
they produce. This is true even of processes that have traditionally been
craft-based and required tremendous skill.
An example of such a process is sheet metal spinning which a flat
plate of metal is rotated on a lathe-like machine and forced over a metal
die using a mandrel. The metal deforms plastically. This operation has
traditionally required very high skill on the part of the operator. Many
different shapes can be produced (the process is most often used for the
production of shades for industrial lighting) but it is easy to push too
hard and tear the metal, or too softly and leave the metal too thick. This
operation may now be carried out automatically under computer numerical
control. Operators complain that the quality of the product is "not what
we can do," but it will not be long until control has been refined to such
an extent that a computer program will be able to produce dies, and spin
customized products to order.
The foregoing is an example of a process in the early stages of
programmability; other processes are much further advanced and have been
reliably programmable for a number of years. The chief constraint in such
processes is no longer the physical manufacturing process, but the information
required to tell the machine what to produce. For example, in the manufacture
of electronic circuit boards, all manufacturing instructions may be completely
specified by a set of computer programs, from drilling holes in the boards
to the exact placement of surface-mount and through-hole components before
soldering. What is more, these operations may be effected reliably and
consistently by programmable machines running standardized programs. Despite
the tremendous variation in electronic devices, circuit board manufacturing
capacity is becoming a commodity. Many firms now produce boards for products
ranging from modems to fashionable electronic toys in small facilities
with only one or two programmable machines. They are often in competition
with a large number of similar sub-contractors using identical machinery.
Spring-making has traditionally involved the precise cutting of
cams and gears to control an automatic spring making machine. This was
a task requiring high skill and years of expertise. Some manufacturers
were thus very much better than others, and required substantial skilled
machine shops to produce the appropriate cams. Today, springs can be produced
under programmable computer-control by small programmable machines that
run a standardized program. Each machine is capable of interpreting the
program and putting the appropriate kinks and hooks into any spring being
produced. Such machines have dramatically changed the industry many small
spring-making shops (often run by ex-employees of the larger manufacturers)
now bid readily on spring-making jobs for small electro-mechanical devices.
There are many other examples of industries in which the type
of coordination described in this chapter is becoming practicable. Although
the global computerization of such a market for flexible capacity
has yet to be seen in practice, we believe that such markets will soon
exist. As global telecommunications and information technology enable such
manufacturers to compete efficiently, with standardized technology and
minimal barriers to entry, sources of advantage for individual firms are
hard to identify. This leaves small firms faced with the prospect participating
in such a global market for flexible capacity with a very important question:
"What is it now important to do well?"
Jaikumar, R. Japanese Flexible Manufacturing Systems: Impact on the United
States. Japan and the World Economy. Volume 1 (2). 1987
Jaikumar, R. Statement before the Subcommittee on Innovation, Technology
and Productivity of the Senate Small Business Committee of the US Senate.
December 2, 1987.
The closest modern translation of this word is "rag-trader". Impannatori
existed during the Renaissance as coordinators of artisans.
Upton, D. M. The Operation of Large Computer-Controlled Manufacturing Systems.
Purdue University Ph.D. Dissertation, 1988
Upton, D. M. A Flexible Structure for Computer-Controlled Manufacturing
Systems, Manufacturing Review, vol. 5 , no. 1, pp. 58-74, 1992.
The question of why machines should "want" to bid for the right to work
results only from the anthropomorphic analogy, and not because this causes
any inherent functional problem.
Jaikumar, R. Resource Allocation in Automated Flexible Manufacturing Systems.
Harvard Business School Working Paper 88-026, 1988.