Composition-Based Approach to Organization Design:
Application to Reactive Problem-Solving in Manufacturing
Ravi Kalakota
Andrew B. Whinston
Center For Information
Systems Management
Department of Management Science and Information
Systems
University of Texas at Austin
Austin, Texas 78712-1175
E-Mail: abw@emx.cc.utexas.edu
insights and design techniques that enable the transition
1. Introduction
In a world of unrelenting change and competition,
designing flexible
and responsive organizations
effectiveness.
Thus, central
to the
management task is the design and implementation of
various workflows/processes linking interacting
tasks,
assessing their effectiveness over time and modifying them
to meet new demands. Terms like,
workflows, cross-functional
boundary-spanning
nature of "theoretical" statements (Schoonhoven,1981). For
instance, theoretical statements such as, organizations are
more successful when their structure
networked organizations and boundaryless cooperation are
frequently used by senior managementto describe the type
of flexible organizations they aspire to build. These terms
technologies (Woodward,1965), organization’s
and vision.
demands (Lawrence and Lorsch, 1969), technology and
"communication sla’uctures"
Whatis the reason for this discrepancy? It is the
lack of realization that the traditional organization design,
regardless of howdelayered, decentralized or re-engineered,
and responsiveness
results
in a changing
economy.Traditional organization design, according to the
four decades of theoretical
and empirical
should match nature of task
(Tushman,1978), are rich in meaningbut quite useless from
a practical
point of view because they provide no
implementationdirection or guidance.
In other words, the bridge linking the macro-focus
of top managementand organizational
been weakin implementationand delivering results.
required to produce sustainable
internal
states and processes should be "consistent" with external
However,while strong in crafting a vision, managementhas
cannot muster the speed, flexibility
"conform" to their
structure need to be properly "aligned" (Khandwalla,1974),
teams, re-engineering,
make a statement about business strategy
Onereason for the lack of clarity is the ambiguous
is an
important area for research given its implications for
organizational
from theory to practical implementation.
work in
organization and managementtheory (OMT), is understood
to be a "fit" betweenthe various contingencyfactors (e.g.
micro-focus of practioners
models does not exist.
theorists
and the
implementing these macro
Howdo we bridge this gap? What
analytical and computationaltools are required to implement
this bridge? Howcan AI help? Howcan we measure the
effectiveness of this bridging process? Whattype of metrics
do we need to measureeffectiveness? It is these challenging
questions that we have tried to address in our research on
flexible organization designs for effective reactive problemsolving in electronic printed circuit-board manufacturing.
environmental uncertainty, rate of technological change,
etc.) deemedcritical
to organizational functioning. While
rich in metaphors and symbols, these OMTtheories
are
poorly developed with respect to normative implications,
2. Building
Blocks of Computational Organization
Design
It is evident that organizationaldesign of the future
must incorporate or embedcomputingto a significant extent
118
to be effective. But, what is organization design? Weview
representational
organizationdesign as an architectural/structural abstraction,
done. Hence we see a gradual transition
encapsulating a society of distributed,
entities
agents
(human or computational)
loosely-coupled
who are working
entities
knownas documents, to get work
from the macro
to the more micro elements. Wewill examine each
one of these in later sub-sections.
The research challenge is: howto design effective
concurrently, competitively or cooperatively to resolve
problems and perform work related tasks to achieve certain
organizations composedof these building blocks? However,
organizational
before embarking on design, two assumptions have to be
objectives or goals. The agent metaphor
for modeling embeddedcomputing, as
stated clearly. First, there is no one best wayto organize; the
software and processors becomeincreasingly sophisticated
secondis that any wayof organizing is not equally effective
in their operation makingthe distinction betweenhumanand
under all conditions (Galbraith, 1973). Keepingthese two
computational agents difficult.
assumptions in mind, let us delve into the details of the
seems appropriate
These agents can have
different levels of autonomydependingon their role in the
building blocks for computational organization design.
process being studied. For instance, agents can be part of
automated processes such as intelligent
networks. On the
2.1 Networks, Processes or Workflows
Dynamicreconfigurable processes, workflows or
other hand, agents can be part of a highly-integrated humancomputing design in which there is an interdependent
networks has come to the fore as the way of improving an
division of labor betweenpeople and computing.
organizations’ ability to react to threats and challenges.
But, what are the building blocks of computational
Dynamic networks are designed to support
superior
organization design? This is a fundamental question. Upon
execution by a clear understanding of the urgency and
detailed analysis, we see three major clusters emerging(see
volatility
Figure 1): (1) Networks,processes or work_flows;(2) agents
decision makingbased purely on hierarchical and functional
(humanor computational); and (3) mformationelements.
authority. Yet there remains muchconfusion over just what
of their environment and the need to break with
dynamic networks are, how they are created and how they
Organization Design
(Networkof
interrelated
operate? Also, very little
is knownabout the computational
building blocks (agents and informaion elements) needed for
developing and sustaining dynamic networks? The focus of
EachProcess is composedI
of Agentarchitectures
our research
I
has been on understanding
how dynamic
networks are designed, implementedand reconfigured using
computing to enable work to be done more effectively
[ EachAgentrequires
/ ~ ~
IInformation
Elements
/ ~
~ [ to Complete Assigned Task
Technical Pr6blem Design
Manuals Reports Schematics
Figure 1: Elementsof OrganizationDesign
Howare these related?
in
real-time environments(e.g. production and manufacturing.)
2.2 Agents
Any process can be visualized as a composition of
Process or workflows
a set of interrelated
tasks performed by agents (e.g.
represent the coarse-grain structural design elements that
engineers, salespersons,
software programs, etc). These
encapsulate the more action-oriented elements -- agents.
tasks represent certain aspects of a larger problemthat the
Agents act on fine-grain information elements embeddedin
parent process describes.
An agent is commonlyused to
refer to an entity that is continuously and autonomously
119
operating in an environment in which other processes and
agents exist concurrently. Agents execute in parallel and
2.3 Information Elements
The paperless
contribute to the overall objective by solving sub-problems.
flow of documents through
They cooperate and share information and resources with
enterprises
each other. Suchcooperation, called coordinated interaction,
quarters as being workflows.This is a critical element of the
among the agents
organization design process that is frequently ignored in
arises
due to data
and task
large and small has been defined in some
interdependencies which impose constraints on howactions
theoretical work. Werapidly movingtowards having a large
are performed.
amountof corporate information and data in electronic form
Clearly an agent is an embedded
system, i.e. it is a
composed of different
data types, ranging from text to
component of a larger system. Kaelbling and Roseuschein
multimedia, e.g. audio, video, graphics etc. This electronic
(1990) refer to "computersystems that sense and act on their
information is not located in one location but spread around
environments"
and linked by wide area networkse.g. Internet.
traditional
as
~ or
~a~te,
d_at~.
A
This new form of workflow organization
example of an embedded agent is a robot
controller that is a componentof a robot system consisting
many questions.
of one or more mechanical
require to be most effective? If distributed throughout the
arms, servo-mechanisms
What information
raises
elements do agents
controlling
axis motion and sensors and actuators for
organization, what form of effective search and retrieval
interfacing
with the external environment. However, the
methods should be developed to provide the information?
notion
of
~ computing
in
the
design
of
Howdo we utilize
the vast amount of information to be
organizational process interactions is a relatively unexplored
design effective organizations? Let us examinesomeof the
area of research. This embeddednotion of a2enthood in
capabilities
addition to the fact that someagents face timing constraints
solving on the manufacturingshopfloor.
i.e. need to process events and data within a given time
(1) Cross Referencing: Strong cross-referencing facility is
frame, movesorganizational agents close to realm of real-
neededto link related information together. For instance, in
time systems where time-critical
the course of repairing a piece of equipment, a technician
deadlines must be met
(agent) mayneed to consult a large numberof documents,
otherwise catastrophic system failure mayoccur.
Several questions have to answered in designing
integrated agent architectures,
necessary for productive work in problem-
i.e. humanand computing
and/or manydifferent parts of each document.In designing
processes
or workflows, we must identify
elements working coherently to produce output. These
information and ways of linking
questions
together.
range from internal
decomposition
of existing
design
of agents,
worflows into isolated
related
location
information
(2) Conditional Branching: To identify a technical problem
componentsand agent composition, i.e. integrating available
with a piece of equipment, a technician performs a number
components to create
of tests, on the basis of which a diagnosis is made and a
a workflow. The goal of agent
composition is to shorten the risk and time of developing
repair procedure is prescribed. The workflow or process
new organizational
must outline each step of the testing and repair procedure,
forms by using incremental design
methods which leverage off existing models, architectures
and there must be a specific series of steps to be taken for
and processes. Hence, organization design can be restated
each possible problem. The workflow model can utilize
as a problem of composing well-coordinated interacting
hyperlinks to represent the possible branchings.
agent structures.
(3) Knowledgerepresentation: Impose a standard knowledge
120
representation methodfor organizing the various activities
frequent problem because of the vast numberof components
associated with different kinds of information. For instance,
(over 7000 components) used in over 120 PCBdesigns, with
provide for configuration/version control of problems(e.g.,
each card typically
review, date, priority, classification/category,
diagnosis,
components. The plant, in our case study, consumes 4-5
involves the assembly of 100 to 300
status).
These knowledge representation
schemes are
million unit components per day. The material shortage
essential
for information sharing amongagents and for
workflowis composite process involving several interacting
creating a knowledge-basefor organizational learning.
workflows
(4) References to External Objects: Manyorganizational
identification,
documents may contain multimedia objects,
and coordinating
motion videos demonstrating
inspection
including
dealing
with
tasks
such as problem
problem diagnosis, production rescheduling
the implementation of the reactive
schedule. Anyfactor that maycause an alteration - schedule
and repair
procedures, audible spoken instructions, full length meeting
change, inventory variance,
minutes, etc. The research from an AI perspective is to
inventory record error - must be compensated for through
create rich structures that can link together parts of the
reactive scheduling and is termed a reactive problem. These
documentsto create knewinformation entity specific to the
problemsarise because of failure to anticipate contingencies
process at hand.
and often motivate actions which are solution centered,
restrict
3. Application:
Reactive
Problem-Solving
in
unplanned usage,
innovation and limit the number of alternatives
considered due to intense time and performance pressures.
Wheneffectively
Manufacturing
scrap,
done, reactive problem-solving prevents
and design human-computing
waste due to low process yield, accumulating unnecessary
integrated organizations, we concentrated on the printed
inventory and scrap. The magnitude of the problem presents
circuit
an important and exciting domainfor the application of our
To understand
board (PCB) manufacturing environment where
global competition, shortening product lifecycles
and
composition-based technique of designing flexible
production philosophies, such as JIT, have intensified the
responsive organizational
need for new organizational designs to support improved
problemseffectively.
and
forms for solving reactive
In our research, we have used the organizational
reactive problem-solving. The reactive problem-solving
phenomenon
exists in all service (e.g customer service) and
agent metaphoras the underlying conceptual foundation for
manufacturing organizations. In manufacturing, it is the
analyzing, modeling and building intelligent
mechanism for monitoring and correcting
agents who would serve as building
production
software
blocks for the
processes to accommodate unpredictable changes in the
computationally-based organization design theory. Agents
environment.
realize
An example of a reactive
problem in
greater
abstraction
power than traditional
manufacturingoperations is "materials shortage". Materials
objects/modules by stronger encapsulation
shortage refers to the problemof unanticipated shortage of
they encapsulate not only procedures and data, but also
parts during production.
Someof the commoncauses of
types, concurrency, knowledgeand ontological details of
materials shortage included excess usage of a part, delayed
the operating domain. The factors necessary to qualify as
delivery
intelligent
by the supplier,
high rejection
rate during
mechanisms:
agents are autonomous agency, learning and
inspection, and discrepancy betweenactual inventory on the
conditional intelligence.
production line and inventory records in the database.
the user can delegate a task to the agent whichthen executes
Materials shortage, in our case study, was found to be a
in the backgroundenabling the user to work on other tasks;
121
Autonomousagency implies that
learning means that the agent improves over time based on
it almost impossible to support an experimental and
feedback; and conditional intelligence meansthat the agent
simulation programwith available computational tools and
can operate in complex, changing domains that require
techniques, and biases research towards abstract theorizing.
autonomous decision
The emphasis on developing computational
making and problem-solving
tools
for
capabilities. For these internetworkedagent structures to be
supporting the design process is both timely and a necessary
successful and operate to our expectations, we must develop
next step. In our work, we have examinedreactive problem-
a theoretical basis for their design. Wemust understand the
solving in manufacturing to gather requirments for tool
characteristics and interface specifications of the workflows development. These requirements are being utilized
to be linked and the nature of organizational
agents
embeddedin these workflows,and the resultant specification
in
creating a new genre of computational tools for effective
organization design.
of the compositebusiness process.
Muchof this work shares a commonbasis with the
intelfigence, object-
[1] Schoonhoven, C.B., 1981."Problems with Contingency
programming and protocol design theory. Our
Theory: Testing Assumptions Hidden within the Language
developmentsin distributed artificial
oriented
References
current research focus is on developing agent semantics
of Contingency’Theory’." ASQ,26.
dealing with interface,
[2] Khandwalla, P., 1974. "Mass output orientation
composition,
specification,
of
verification issues and on the methodologyof organizational
operations technology and organization structure." ASQ,14.
agentification i.e. transforming an existing organizational
[3] Galbraith, J., 1973. Designing ComplexOrganizations.
process into one embeddedwith software agents.
Reading MA:Addison-Wesley.
[4] Lawrence, P. and J. Lorsch, 1967. Differentiation and
4. Summary
integration in complexorganizations." ASQ,12.
Increasing competition and rapid changes in the business
[5] Tushman, M., 1978. "Technical Communicationin R&D
environmenthave put an intense sWainon the organizations’
laboratories:
ability to react to threats and challenges. In responseto these
Academyof ManagementJournal, 22.
pressures,
[6] Woodward,
J., 1965. Industrial Organization: Theoryand
organizations
undergoing significant
are either
contemplating or
transformations
with many new
The impact of project work characteristics."
Practice. London:OxfordUniversity Press.
forms - such as virtual and networkorganizations, strategic
[7] Kaelbling, L.P and S.J. Rosenschein. 1990. Action and
alliances or partnerships, and practices - such as business
Planning as EmbeddedAgents, Robotics and Autonomous
process re-engineering, process improvement,and seamless
Systems, Vol. 6, nos. 1&2, pp. 35-48.
integration
emerging.
of computing with business
processes-
However, our comprehension
of the
organizational transformation process, involving the design
of new organizational
technology,
forms with embeddedinformation
is limited
due to the difficulty
of
conceptualization, operationalization and measurementdue
to its multi-dimensionalnature.
The complexity
involved
in designing
organizations with embeddedinformation technology makes
122