IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 3, NO.

2, JUNE 1991 149

The Role of Artificial Intelligence in Understanding

the Strategic Decision-Making Process
William E. Spangler

Abstract-Research in decision support systems (DSS) is con- Second, not only is strategy formulation a complex individ-
ducted from two perspectives. The first views the mission of ual task, but it also involves communication and negotiation
DSS research in terms of building specific tools for supporting with other individuals as part of an organizational decision-
decision making, while the other views a DSS as a mechanism for
modeling, and thus better understanding, the decision process. making process [29], [67], [92]. This group-oriented, heterar-
Expert systems for strategic planning have generally taken the chical process in turn has tended to complicate the construction
former approach. This paper argues that the latter approach of expert systems (ES), or other AI-based approaches, which
is required in order to shed light on the early, and perhaps have traditionally been more suited to modeling a single
most important stages of strategic planning, particularly the individual.
early interpretive stages involving strategic intelligence analysis
and issue diagnosis. Research in artificial intelligence-including Third, executives tend to resist using computer-based deci-
investigations into diagnosis and situation assessment, analog- sion support systems of any kind, let alone AI-based systems
ical reasoning, plan recognition, nonmonotonic reasoning, and [24], [36], [115]. Formulation of strategy is instead often
distributed intelligence, among others-can be used to build viewed as an art, a highly creative process which is not
models of strategic decision making that help researchers in conducive to the intrusion of computers.
better understanding this traditionally unstructured activity.
Fourth, long-range strategic planning does not invite the
Index Terms-Artificial intelligence, cognitive modeling, com- type of validation found in the traditional domains. There
petitive intelligence, decision support systems, expert systems, may not be a “right” or “wrong” answer. Instead, the ultimate
strategic decision making.
strategy may be a compromise from among several alterna-
tives, constrained by many environmental factors, and subject
I. INTRODUCTION to dispute from other interested parties in the organization.
LTHOUGH the use of artificial intelligence (AI) tech- Furthermore, it may take years before the consequences of
A nologies continues to expand in the corporate world, the an implemented strategy are known. Despite these issues,
growth of that same technology for the support of corporate or however, researchers and practitioners are slowly pushing
organizational strategic planning is somewhat slower. This is AI technology, and its applications, up the corporate ladder.
particularly apparent when comparing the strategy area to other This survey will focus on research issues involving strategic
domains, such as medicine [lo91 or engineering [84], or even planning and AI, including the nature of the formulation task,
to other traditional management disciplines, such as finance cognitive studies of strategic planners, and computer-based
[53] or accounting [30], [110]. There are several reasons for support for strategic planning. We will begin by reviewing the
this. research to date, and will argue that, like traditional decision
First, while the application of AI requires an adequate support systems (DSS) research, much of the potential for
understanding of a task and a problem-solver, the task of future research in this area lies in modeling the ill-structured,
corporate strategy formulation is still relatively unknown. early stages of the strategic decision making process. There-
As a result, strategic problems continue to be characterized fore, we will discuss a specific model-based approach to the
as complex, unstructured, fuzzy, ambiguous, and notoriously study of these early stages: strategic intelligence analysis and
difficult to formulate [l],[29], (671, [72], [102]. Mason and issue diagnosis. We will suggest relevant research in AI that
Mitroff echo this sentiment by describing these so-called can support this type of effort, and conclude by offering a
“wicked” problems in explicit metaphorical terms: research agenda intended to guide future efforts.

Wicked problems are not necessarily wicked in the

11. THEMISSIONOF DECISION
SUPPORT SYSTEMS RESEARCH
perverse sense of being evil. Rather, they are wicked
like the head of a hydra. They are an ensnarled web of There is a continuing discussion within the research com-
tentacles. The more you attempt to tame them, the more munity concerning the objectives of DSS research. One school
complicated they become. [67] of thought views DSS research in terms of the computer-based
support issues. Goul, for example, described the validation cri-
teria for his strategic planning ES (described in the following
Manuscript received October 1, 1989; revised October 1, 1990. section) as follows:
The author is with the Artificial Intelligence in Management Laboratory, Designers have typically validated systems by seeking to
Joseph M. Katz Graduate School of Business, University of Pittsburgh,
Pittsburgh, PA 15260. answer the question, ‘Does the system produce the same
IEEE Log Number 9144302. decision as the expert when given a particular problem?’
10414347/91/060G0149$01,00 0 1991 IEEE

- ~~
150 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 3, NO. 2, JUNE 1991

There has been little attempt to validate a system by Like AI-based applications in other fields, expert systems
attempting to answer the question, ‘Does the user of an in organizational strategic planning were spawned as rule-
ES make better decisions by using it?’ [45] based production systems, and have since evolved with the
Other researchers, however, regard the building of DSS as maturation of newer AI techniques developed in the computer
a technique for studying the process of decision making. If and cognitive sciences. This evolution, from rule- and frame-
the system does not directly improve the decision making of based approaches to the use of techniques such as analogical
its user, but it does, through computational modeling, allow reasoning, fuzzy sets, and distributed AI, is illustrated in the
some insight into how decisions are made, then the research discussion below.’
is validated based on this criterion. Long-term benefits will
accrue in the form of subsequent improvements in decision A. Rule-Based Approaches
aids and decision making. For example, Keen [55] suggests The feasibility of a rule-based approach to modeling cer-
that, with a greater understanding of the decision process, a tain aspects of strategic decision making has been discussed
DSS can become a more active partner in the decision-making previously [24], and has been demonstrated in a number
process, perhaps guiding the decision maker while still leaving of expert systems. Goul [44], [45], for example, developed
the “primacy of judgement” with the user. a rule-based ES to assist a decision maker in performing
The decision-making focus is also reflected in King’s def- situation assessment. The system contains 475 rules, and took
inition of strategic decision support systems (SDSS) [57]. approximately two years to complete. Like many rule-based
King recommends that SDSS target the earlier phases of the systems, Goul’s ES uses a deductive strategy. The system asks
decision-making process (that is, to borrow from Simon’s a series of questions from a particular “folder” chosen from
[lo51 terminology, SDSS should target the intelligence and the following: 1) diagnostic questions, 2) emerging industries,
design phases, rather than the choice phase, of the three-phase 3) capacity questions, 4) buyers and suppliers, 5) industry
process). Other researchers concur 1431, [76], [93], [114]. evolution, 6) competitive actions, 7) market signals, 8) entry:
Indeed, these earlier phases are perhaps the essential element new business, 9) fragmented industry, 10) mature industry,
of strategic decision making, since they encompass the for- 11) declining industry, and 12) global industry. Based on the
mulative processes of environmental assessment and strategic answers to the questions asked, the system responds with an
issue diagnosis. However, the earlier phases, and associated assessment of the company’s situation. In an example dealing
strategic problems, are also the most poorly understood. with emerging industries, an assessment states: “the greater the
An understanding of the cognitive aspects of strategic chance that technological change will make early investments
decision making (and of decision making in general) is crucial obsolete, the more risky is early entry into a market.” Then,
in the future development of DSS. Keen notes that as computer like other deduction-based schemes, Goul’s system allows the
technology becomes less of a bottleneck in the development of user to ask “Why?” The system responds: “(because) firms
DSS, the lack of understanding of decision making becomes entering later will have the advantage of the newest processes
more so [ S I . Mockler concurs: without paying the development costs for early technologies
While progress is being made in computer technology (see Porter, 233)” [45, p. 1381. It should be noted that the
and in software, not enough is yet known about the system’s knowledge was derived from the expertise of a
expert thought processes involved in strategic corporate strategic planning professor, who had extensive planning ex-
planning for much more sophisticated software devel- perience, as well as from the prescriptive literature of strategic
opment, even if sufficiently advanced technology were planning. Goul later incorporated the knowledge and (perhaps
available. conflicting) advice of multiple experts, citing the resulting
Only as the intellectual activities involved in strategic beneficial effect on situation assessment. Each conclusion or
corporate planning are defined more and more precisely recommendation, however, is offered independently.
will the groundwork be laid for developing more sophis- Lynch and Hertz [65], [50] have developed a rule-based
ticated decision support software for strategic planning. system, called PAD, which can be used to develop policies
While a great deal has been done to define the tasks and procedures in a manufacturing setting. Specifically, the
involved in planning, very little has been done to define system provides policy support for the procurement of parts
the various thought processes involved in doing these for personal computer manufacturing, assignment of plants for
tasks. [77]. assembly of the computers, and distribution of the finished
products. The system is an extension of traditional rule- or
frame-based ES, in that the reasoning is driven by broad policy
111. EXPERT SYSTEMS FOR STRATEGIC DECISION MAKING
statements (for example: “Buy at the lowest prices consistent
A large number of decision support systems for strategic
decision making (also referred to as executive information As an aside, we should note that the distinction between business-related
systems (EIS) or executive support systems (ESS)) have been The strategic planning and planning in other settings can be somewhat ambiguous.
RAND Corporation, for example, has built a number of AI-based systems
described in the literature (see, for example, [4], [20], [28], for the US. military, mainly involving situation analysis, and strategic and
[31], [39], [51], [52], [58], [94], [95], and [103]). Of late, a tactical planning for a variety of scenarios [9],[104]. Similarly, Anderson
growing number of researchers have begun to integrate results and Thorson [3] have developed rule-based models of foreign policy decision
making, in an attempt to model the decision-making cognitive processes of
from artificial intelligence and expert systems research within political leaders during international crises. Many of the issues raised in those
the strategic planning domain. studies are relevant to the notion of decision making models in business.
SPANGLER: ARTIFICIAL INTELLIGENCE IN UNDERSTANDING STRATEGIC DECISION-MAKING PROCESS 151

with quality statements and the requirement that not more than Kochen and Min [59], for example, address the early
half the required quantity of any part be purchased from a stages of strategic decision making by providing a theoretical
single vendor.”), rather than by more specific, and narrow, example of a rule-based strategic intelligence system (SIS).
input. SIS is intended to incorporate frames as a mechanism for
The work of Mockler and Dologite [77]-[79] is gener- representing strategic, particularly environmental, knowledge.
ally similar to that described above, in that they have built SIS was conceived in recognition that strategic planning is
deductive, rule-based systems from commercial ES devel- driven by a decision-maker’s cognitive map of the business
opment tools (also known as “shells”) to support strategic environment, including his or her perceptions of relevant
decision making. The systems are intended to address the environmental entities (such as competitors, customers, and
entire spectrum of strategic planning, from environmental technologies), the relationships and associations among these
analysis to recommendation of strategy, ultimately suggesting entities, and expectations of environmental change over time.
generic strategies (such as “selective-investment” or “phased- By providing structural and procedural (generally hypothesis-
withdrawal”) based upon a number of environmental factors. driven) support for this type of strategic knowledge, the
The systems are generally derived from the knowledge of the authors argue that an SIS can assist in improving the cognitive
authors, as well as from the strategic planning methodolo- maps of decision makers. This improvement is achieved
gies prescribed by researchers such as Porter [88], [89] and through the revision of rules in the face of unexpected en-
Hamermesh (481. vironmental events.
1) Limitations of Rule-Based Approaches: The approaches The approach taken by Kochen and Min is essentially
taken above work well when strategic problem solving can be an extension of existing rule-based paradigms. The authors
represented heuristically, as a deductive process. Nevertheless, discuss using the meta-rule and frame-based capabilities of
the feasibility of using if-then rules as a comprehensive existing commercial development tools, such as EMYCIN-
model of complex decision situations has come into question. based shells. The proposed system architecture is modeled
Dhar and Pople [26], for example, describe an aborted effort around traditional DSS research, encompassing a data base
to model business resource planning in the form of rules. containing relevant external and internal domain information, a
Rules, they suggest, are likely to hide important contextual model base containing the qualitative and quantitative models
knowledge, particularly in complex domains. Too often in such described above, and a knowledge (rule) base of heuristics
circumstances, rules are conditioned by other factors, and an related to strategic analysis and selection.
attempt to include all possible qualifications and preconditions Berqvist and Lounamaa [IO] used an object-oriented ap-
in the rule set may not only make the system intractable, proach in building an expert system that can support a decision
but will also depart from the actual reasoning process of the maker in comparing and evaluating strategic investment oppor-
expert.* tunities (SIO) available to a company. SIO’s are potential new
Again, the broad nature of strategic decision making re- business units within an organization, and are evaluated based
quires computational approaches that also model and support on required and expected return on investment (ROI). Required
the early, ill-structured stages of strategic problem solving ROI is adjusted up or down depending on other contributions
[99]. Rule-based systems, particularly those based on the of the particular SIO, such as expected impact on market share.
MYCIN model [13], arguably focus on preformulated strategic The expert system was developed within an organizational
problems. Furthermore, as we discuss below, the knowledge framework, incorporating the various analytic techniques and
structures provided within (purely) rule-based development heuristics used by experienced managers, and simulating the
environments are generally not adequate to model the special process by which SIO’s are evaluated at progressively higher
nature of strategic knowledge. In light of this, AI-based levels within the corporation.
strategic DSS research has begun to draw on AI research in a Carlson and Ram [16] describe a prototype for a frame-
number of other compatible areas. based hypermedia system called SPRINT (Strategic Plan and
Resource Integration), which is designed to support a manager
B. Frame-Based and Object-Oriented Approaches: by providing a computational representation of the important
Hierarchy and Inheritance elements of the manager’s mental model for a strategic plan.
In order to address some of the limitations of rules, re- The individual model is integrated with the models of other
searchers have begun developing frame-based and object- managers in order to produce a global strategic plan intended
oriented mechanisms, often within rule-based systems, in order to address the company’s objectives.
to facilitate reasoning using hierarchical, interrelated strategic The SPRINT system incorporates a frame-based paradigm
information. Strategic decisions are generally derived through within an architecture of distributed knowledge bases, or
aggregated data, through general trends and perceptions based agents. This distributed AI approach is described in a later
on a lower-level analysis of more detailed data. Computer- section.
based knowledge represented in this form aids in modeling
C. Analogical Approaches
the schema or cognitive maps [6], [102], as well as the
associational reasoning [79], used by strategic planners. Rather than view strategic decision making primarily as a
deductive process, Sullivan and Yates [lo71 instead focus on
2This is also an aspect of the “qualification problem’ described by McCarthy the ability of seasoned managers to associate current circum-
~91. stances with analogous previous experience. Their research
~

152 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 3, NO. 2, JUNE 1991

resulted in a computer system, called Reasoning by Analogy system described in the following section or the RAND
(RBA), which attempts to support this aspect of strategy studies cited previously, some systems have incorporated
formulation. It retrieves cases that most closely relate to a numerical regression techniques within expert systems in order
particular situation, and thus allows the decision maker to to improve decision support for business forecasting. Round
understand, and perhaps implement, the strategies that were [98], for instance, describes the Retail Planning System and
successful in these analogous situations. Advisor [62], which joins a demographic simulation model
RBA was conceived in response to a common observation, with knowledge of how to run the model, thus allowing retail-
namely that systematic and consistent analogical reasoning, ers to formulate optimal plans for their distribution outlets.
even for an experienced manager, is a difficult, perhaps even
intractable task. This becomes readily apparent when one F. Distributed Artijicial Intelligence
considers the hundreds of cases that might be relevant to a
particular problem, as well as the thousands of factors that There is a relatively new but growing body of research in
may describe them. One or two cases may not be adequate, AI dealing with distributed problem solving, where indepen-
particularly if the analogy used to retrieve them is less than dent intelligent agents, each having particular knowledge and
perfect [loll. abilities, interact in order to solve a problem [96]. Research in
Thus, Sullivan and Yates developed RBA, which is essen- distributed AI has been conducted in a number of domains,
tially a sophisticated database retrieval system. It contains a including signal processing [64], robot planning [21], and,
large repository of cases, with an extensive index of factors most relevant to this discussion, social and organizational
related to each case. The RBA system uses AI-based heuristic modeling [8], [16], [17], [32], [66], [68], [108].
search techniques to scan through the cases, compare factors, The relevance of Distributed Artificial Intelligence (DAI)
and retrieve cases which are most analogous to the current to strategic decision making is further supported by noting the
problem. As an example, the authors describe a problem that computational mechanics involved in interagent communica-
involves the use of information technology for competitive tion, which bear a fascinating similarity to the description of
advantage, and note that the system retrieved only the cases heterarchy, or collective strategic decision making, offered by
which were most relevant to that particular problem. While Dutton et al.:
doing so, the RBA system rejected analogies to all of the Revisions in judgement have been characterized as
“classic” cases involving the use of information technology recursive at the individual level. At the collective level,
for competitive advantage. RBA judged that the other factors the process is analogous to an array of recursive proce-
were sufficiently dissimilar to warrant the rejection. dures cyclically calling on each other, yet rarely having
a single highest level. [29].
D. Futty Set Theory Researchers are beginning to apply AI-based distributed
Hall [47] has included work in fuzzy set theory [116] within models to the support of strategic decision making. Syed
a rule-based expert system for strategic planning. This ‘‘fuzzy and Tse [lo81 describe the Integrated Consulting System
decision support system,” called Stratassist, is intended to (ICs), a distributed system developed at SRI. ICs integrates
improve strategic decision making by allowing the user to quantitative and qualitative knowledge of technology, firms,
provide input in fuzzy terms: market share is “more-or-less and consumers, as well as market adjustment processes, in
high,” competitive position is “somewhat strong,” and so on. order to support business decision makers in competitive
This frees the decision maker from having to quantify values analysis and strategic planning. ICs combines a heuristic,
that are inherently qualitative, or that are known only within frame-based approach with analytic/algorithmicprocedures, all
particular ranges. within a distributed blackboard architecture. The system mod-
Like other researchers, Hall has derived the rules for her els firms and consumers as purposeful objects having goals or
system using previously published normative theory from the aspirations and undertaking actions intended to achieve those
strategic planning literature. In this case, Porter’s [88] five goals. The objects are also capable of modifying their goals
competitive forces: and the generic strategies for dealing with in reaction to environmental events.
them, form the foundation for the system’s knowledge. ICs provides decision support through a knowledge-based
simulation of the competitive environment, which includes
E. Knowledge-Based Simulation the independent behavior of the objects described above.
The user can influence the simulation by interacting with
AI researchers are beginning to combine expert system the system to control the timing and nature of the problem-
technology with traditional simulation methods in an attempt directed search processes employed by the firms/objects. As
to build systems that are more robust in complex proba- a result, the system allows “what-if” analyses for various
bilistic situations [1121. The result, termed knowledge-based components of the competitive environment, such as possible
simulation [97], is intended to solve problems in a variety new markets and new product grades. ICs also permits the user
of functional areas, including some within strategic planning. to interrogate the model in order to determine the competitive
For example, in addition to systems that attempt to simulate balance in each market under investigation.
the behavior of environmental agents, such as the distributed Carlson and Ram [17] describe a Distributed Knowledge
Intra-industry rivalry, threat of new entries, threat of substitution, buyer
Base Management System (DKBMS), derived from social
power, and supplier power. epistemic theory and designed to support the SPRINT model
SPANGLER: ARTIFICIAL INTELLIGENCE IN UNDERSTANDING STRATEGIC DECISION-MAKING PROCESS 153

described above. SPRINT models the knowledge and inter- development of large, complex systems that must contend with
action of heterogeneous, distributed agents communicating changing environmental circumstances.
within an integrated social community. The model essentially
represents a business organization, where the knowledge re-
quired to solve a problem is often distributed among various Iv. FUTUREDIRECTIONS:
MODELING
THE EARLYSTAGES
managers, each of whom may contribute to the solution with- The research above can generally be described as investi-
out understanding every aspect of the problem. The complexity gations into decision support, technique-driven explorations
of the corporate environment and the decentralized nature within the framework of specific AI approaches. We sug-
of modern corporate decision making almost ensure that no gest, however, that considerable opportunity exists in taking
single individual can be knowledgeable about all aspects of an alternative, cognitive modeling approach to the study of
the strategic domain (with the possible exception of certain strategic decision making. That is, knowledge of strategic
entrepreneurships [73] ). As a result, strategic decision making decision making can be enhanced through observation and
is generally undertaken by “partial experts” [26], and requires computational simulation of the detailed cognitive processes of
the differing perspectives and insights of all relevant parties an expert or experts who are reasoning about the ill-structured
in order to make an informed organizational decision. early stages of planning, related primarily to environmental
In the DKBMS, there is no single entry point, no single assessment (i.e., information gathering, analysis, and interpre-
authority that coordinates problem solving throughout the tation). Toward this end, we suggest setting aside the specific
network and directs a response back to the user. Control is AI techniques, at least initially, and asking the following
instead delegated to local authorities, individual knowledge questions in a very precise way: what do experts do when they
base systems, each of which is responsible for its own user assess the business environment, and how do we formulate
interface and processing. our inquiry in order to find this out? Then, rather than
The kind of distributed architecture exhibited in these constraining the investigation, relevant AI research can be
studies has several potential advantages. The architecture selected in an eclectic manner to support an empirically-based,
is, for example, patterned after the global decision making computational model of expert behavior.
processes of organizations, thus making it more conducive to The cognitive perspective on strategic decision making,
the support of the distributed problem solving typically found exclusive of computational models, has been addressed by a
in many companies. The architecture is also modular, meaning number of researchers in strategic planning. Schwenk [ 1021,
that individual agents can be added as needed expertise is for example, discusses cognitive models in the early stages of
uncovered. This facilitates the knowledge acquisition process, strategic planning, particularly in terms of default reasoning
and simplifies the development and maintenance of large, (strategic assumptions) [67], [74], [75] and schematic struc-
complex systems. tures (cognitive maps), and suggests that these play a key role
in environmental assessment and interpretation. Assumptions
in effect are a part of the schemata, as is a subjective
G. Machine Learning understanding of causal relationships and predictive judgments
Elofson and Konsynski [32], also noting the distribution based on those perceived relationships.
of required organizational expertise, developed a prototype Dutton et al. [29], futhermore, suggest that well-developed
of a distributed system based on many of the same factors models tend to distinguish expertise in strategic decision
motivating Carlson and Ram. They state, for example, that making, and add that certain characteristics typify the strategic
experts in a company are rarely skilled in every aspect of a interpretation process. They note, for example, that the early
strategic problem, and suggest that a distributed computational stage known as strategic issue diagnosis (SID), as the name
model is most realistic in collecting and integrating diverse implies, is essentially an iterative, diagnostic process, in which
organizational expertise. preliminary hypotheses lead to the search for confirming
This research departs, however, by focusing on commu- or disconfirming data [19], [29], [72], [92]. It is also a
nication and knowledge sharing among agents, and by in- dynamic process, in which interpretations are subject to con-
corporating elements of machine learning in order to allow tinuing revision as new information invalidates or modifies
“apprentice” agents to learn to classify various patterns of old information. They also suggest that a combination of
strategic environmental data. The apprentices, described as inductive and deductive reasoning is commonly observed in
semi-autonomous agents, comprise what is termed the Knowl- strategic decision making. While the process of developing
edge Cache. Within the Knowledge Cache architecture, an new schematic structures is often inductive in nature, infer-
apprentice acts as an agent for an expert, asking the same ring conclusions from assumptions (in existing structures) is
questions, and with initial guidance, classifying responses to typically deductive.
the questions within a framework of attribute-value pairs. We suggest extending these cognitive studies by turning to
Through a machine learning algorithm [63], the apprentices an information processing theory of human cognition [81],
are subsequently able to independently classify similar patterns [82]. From this perspective, a computational model of human
that may arise, thus off-loading much of the effort from problem solving provides a rigorous empirical mechanism
the expert. Furthermore, machine learning, by unobtrusively for exploring, in detail, the cognitive processes suggested
incorporating the knowledge of experts, has the potential to by the researchers cited above. This type of approach has
improve the knowledge acquisition process, particularly in the been used to study problem solving in a number of other ill-
 154 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 3, NO. 2, JUNE 1991

structured management domains, including financial diagnosis detailed assessment of U.S. government intelligence expertise,
and planning [111, inherent risk assessment in auditing [83], primarily in the context of the Cuban missile crisis, which
and manufacturing resource planning [25]. We illustrate this generally concurs with Meyer’s description of CI. The notion
essentially inductive approach to model building within strate- of generic expertise is further supported by noting that much
gic decision making by focusing in some detail on Competitive of the expertise in CI is derived from the ex-government
ZnteZIigence (CI) analysis, the preformulation stage of strategic intelligence analysts who move to the private sector. These
planning, related to SID, that has been addressed from various analysts are valued for the intelligence skills, exclusive of any
perspectives in some of the expert systems studies cited in the particular domain, that they bring to a company.
previous section. CI is the systematic study of a company’s in-
dustry and competitive environment [18], [56], [90], [106], and B. Diagnosis in Competitive Intelligence: An Example
represents a highly complex, analytic component of strategic Meyer [71, p. 191 gives an example of the differences in
decision making. It therefore provides a fertile environment for impact that may be expected across a spectrum of interested
qualitative, computer-based exploration of cognitive decision parties, based on a single piece of intelligence data. He asks us
models within this domain. to consider a scenario where two inventors have just developed
an automobile engine that allows a mid-sized car to drive 50
A. Evidence of Expertise in competitive Intelligence miles on a gallon of gasoline.
While many studies have investigated the organizational The inventors had originally worked at a major U.S. au-
or systematic process of CI, there is little, if any, research tomaker, but resigned to start their own company. This news
into the cognitive processes of CI analysts. Related research is obviously of great importance to a number of players in
has been conducted at the Environmental Protection Agency the industry, each of which will generate intelligence related
(EPA) of the U.S. government [35]. The EPA gathers sensitive specifically to its own interests and objectives. For example:
corporate data from a large number of companies. The agency
must take care not only to avoid directly disclosing sensitive U. S . Automaker
information to the companies’ potential competitors, but also The new engine looks to be a great success. Our own
to avoid disclosing secondary information that could be used people haven’t yet finished their analysis, but prelimi-
to derive the sensitive information. Analysts at the EPA thus nary results suggest that the engine can be modified for
analyze requests for data in a manner similar to military our cars. We know for certain that Ford, Chrysler, Nis-
intelligence analysts. Through a statistical, diagnostic process San, Honda, and Volvo also want licensing agreements;
of “information fusion,” which pieces together the information we don’t know what terms they want. [71, pp. 19, 201
to be disclosed as well as any publicly available information,
an EPA analyst will attempt to derive any potentially sensitive Japanese Automaker
information.
The inventors will not sell exclusive rights, as we
By observing expert analysts in this domain, EPA re-
predicted they would not. . . . No doubt they will also be
searchers have developed a computer system incorporating
negotiating with our competitors. . .. (Nevertheless) if
a statistical algorithm that is intended to calculate ranges of
we begin to produce cars equipped with this new engine
potentially sensitive quantitative data. Heuristic knowledge is
before our U.S. and European competitors, these coun-
included, providing diagnostic support to the decision maker in
tries are likely to retaliate by lowering import quotas on
determining whether the derived ranges are sufficiently “close
our exports. ... [71, p. 201
enough” to the protected information to warrant withholding
the requested information.
With respect to CI, Meyer [71], while not conducting Wall Street Brokerage House
an in-depth research study of the cognitive processes of CI There should be lots of action today when the mar-
‘ analysts, has described an expert’s perspective (his own) ket opens. Our guys expect that GM and maybe even
within the competitive domain. Meyer asserts that CI, like SID, Chrysler will nosedive, since most people figure they’ll
is an iterative, diagnostic process of data collection, analysis, find a way to come out last on this. As for the oil
and reporting. The expert analyst is capable of knowing companies, they’re already down so far that this won’t
what information is needed in response to top management make too much difference, at least in the short run. If
objectives, and where to find the information. This is facilitated this sets off a panic in the Mideast, we could see a fast
by rich mental models of the environment, particularly in terms run-up in the price of gold. [71, p. 211
of the goals, intentions, and capabilities of competitors, as
well as the external forces acting upon all of the competitive What is particularly interesting about this, and other
entities. These models enable the analysts to piece together examples of the subjective interpretation of intelligence that
diverse information in the context of their own experience, Meyer discusses, is the process of assimilation of the new
and within the constraints imposed by corporate objectives. datum into a mental framework that is assumed to preexist in
Meyer’s observations strongly suggest a generic “intel- each of the hearers, which determines what additional hypothe-
ligence” expertise that encompasses not only, the business ses will be adduced, what consequences the hearer expects
environment, but also the more established intelligence activ- to unfold, and what reactions the hearer deems appropriate
ities of world governments. Allison [2], for example, offers a and necessary. Among other consequences, presumably, will
SPANGLER: ARTIFICIAL INTELLIGENCE IN UNDERSTANDING STRATEGIC DECISION-MAKING PROCESS 155

be further changes in the hearer’s mental interpretative frame- and Pople, would be useful in discovering and modeling
work-setting the stage for assimilation of further intelligence the cognitive structures and processes involved. Similarly,
nuggets. protocol analysis of the discourse between an analyst and a
specialist would facilitate an understanding of their problem
C. General Approach to Formal Model Construction solving collaboration, particularly in terms of how the analyst
We suggest that formalization of the early decision pro- communicates requirements to the specialist and then incorpo-
cesses can begin with an investigation of the behavior of rates the response back into his or her existing mental model
real intelligence analysts engaged in the sort of reasoning of the business and competitive environments.
hinted at by Meyer-where the raw data become the trigger
for the creation of an enhanced interpretation of the meaning
of that item, where the analyst’s knowledge of the industry, D. AI Research Supporting Model Development
the players, the competitive environment, his own company’s Several continuing streams of research in AI provide sup-
strategic plans, and the impact of all of this on those plans-all port for a cognitive modeling approach. We briefly describe
are brought to bear. Cognitive modeling research would seek selected areas of AI research that are particularly appropriate
the primitive constituents out of which such interpretative for modeling strategic decision making behavior, particularly
models are crafted, as well as the mechanisms by which those CI analysis. Based on empirical observation of specific expert
constituents are synthesized into meaningful, organized, often behavior, any or all of this research might be used to facilitate
action-oriented summaries. computational modeling.
The construction of a formal model can follow the general 1) Diagnosis and Situation Assessment: Because the role
inductive approach of researchers (cited earlier) who have of diagnosis is arguably fundamental to the cognitive model
modeled ill-structured management tasks using methodolog- of a CI analyst, construction of a computational model would
ical techniques borrowed from cognitive psychology [ 111, benefit from related studies in diagnosis and diagnostic model-
[25], [83]. Development of a cognitive model, for example, ing. Gadd and Pople, for example, developed a computational
begins with a thorough analysis of the task under study model, based in part on an object-oriented representation,
[81], including informal interviews and careful review of which was judged to be compatible with their empirically-
written documentation, combined with detailed observation derived theory of diagnosis. Objects and daemons provided
and concurrent verbal protocol analysis [111], [33] of expert an appropriate mechanism for modeling the expectations,
problem solving. Empirical observation, using a mixture of assumptions, and suspicions utilized in diagnostic problem
real and simulated problem cases, forms the foundation for solving, for instantiating knowledge structures associated with
a conceptual representation of the expert’s cognitive struc- observed data, and for guiding the search for confirming or
tures and processes, leading eventually to construction of a disconfirming evidence.
computer-based model of that representation. The Gadd and Pople model is an extension of prior investi-
In order to account for specific issues of CI analysis not gations into diagnosis within a number of areas, first in the
related to the above studies, we can draw upon analogous domain of internal medicine (with Internist and Caduceus)
methodological approaches used in other domains. One issue [85], [86], and more recently in three additional domains
in particular is the need for a CI analyst to confer with (Nuclear Power Plants, NASA operations, and Critical Care
a variety of individuals, including managers with particular Medicine) [113], [87]. The applicability of this particular
information needs, and specialists in areas outside of the diagnostic model to these varying domains strongly suggests
analysts’s expertise [40]. This type of collaborative problem the existence of invariants in the general task of diagnosis and
solving is common in the medical domain, for example, where situation assessment, and argues for a similar type of approach
Gadd and Pople have used protocol analysis to study and in other diagnostic tasks, such as CI analysis.
model person-to-person interaction in medical teaching rounds 2) Plan Recognition: Plan recognition refers to the act of
[41]. In their research, a physician-teacher was modeled as “explaining a set of observed or described actions.. . by
utilizing two diagnostic models: one of the student (to evaluate constructing a plan which contains them.” [54]. Because plans
the level of the student’s knowledge), and one of the patient typically contain goals, intentions, beliefs, and assumptions,
(to diagnose the disease). The physician‘s evaluation of the techniques of plan recognition can aid an observing agent in
student was based on a reconciliation of the two models, understanding the potential actions of another agent, and the
essentially using a dual diagnosis process. feasible rationale for those actions.
The result of their study, a structural representation and The intentions and goals of another agent may be inferred
process model constructed within a general framework of diag- in one of two ways. First, an agent’s plans may be expressed
nostic problem solving, is appropriate to similar investigations (implicitly or explicitly) as a part of interactive conversation or
of collaborative diagnosis in strategic decision making. A CI discourse, such that the listening (or observing) agent expects
analyst, for example, responds to management inquiries by the other agent intentionally to communicate its plans [46].
tailoring his or her response within the dual context of the The second form of plan recognition arises in the absence
executive’s objectives and intentions, and the external business of discourse, and requires an agent to infer the intentions
environment. Because those intentions may be discovered and goals of another agent without interacting with that agent
(or diagnosed) through communications between the analyst (often labeled keyhole recognition) [ 1001. AI researchers have
and the manager, protocol analysis, as employed by Gadd investigated the problems of keyhole recognition from various
156 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 3, NO. 2, JUNE 1991

perspectives, including multiagent planning [121 and user making, particularly the evaluative approach to hypotheticals
interface design [38], [70]. and adversarial reasoning taken by Ashley [5].
This mode of plan recognition, unlike the analysis of
discourse, more clearly reflects the dilemma of inferring the
strategies of competitors in the CI domain. Because commu-
nication with and control over competitors is generally. limited v. SUMMARY AND CONCLUSIONS: SETTING THE
(at best), the CI analyst is required to infer the strategies and in- RESEARCHAGENDA
tentions of competitors based on somewhat limited knowledge. The CI task is complex, and as a result, a research agenda for
Although researchers in AI have generally conceded that this the task is necessarily ambitious. We describe below general
is much more difficult than inference through direct discourse, objectives and research questions intended to guide future
we might nonetheless draw on research into representational efforts in this area,
schemes and algorithms developed to infer plans based on
partial information [12], [54], [loo].
A. Objectives
3) Real-Time Problem Solving: Real-time knowledge-based
systems have come under increased investigation in the last 1) To understand more fully the cognitive processes and
decade as AI researchers have sought to understand the issues structures of CI analysts, and with that the CI task itself.
of intelligent support involving tasks that are time-dependent Specifically:
and reason within a changing environment [61]. The areas of What is the analyst’s cognitive representation of his
investigation have included process control [go], [113] and or her business, particularly in terms of the goals
medicine [34], among others, but can potentially be expanded and objectives of upper-level management? How does
to include the dynamic, time-dependent CI environment. Al- this representation reconcile with the analyst’s mental
though the time frames may vary (some real-time systems model of the strategic environment?
must react to situations in a matter of milliseconds), many How does the analyst access and utilize supporting
of the issues relative to CI are similar. From the criteria of resources (of all kinds, including human resources)?
real-time problem solving proposed by Laffey et al. [61], a How are these resources integrated into the analyst’s
cognitive model of CI analysis would require 1) gathering existing mental model?
and synthesizing data across time, 2) reasoning about past, How does the analyst decide which data sources to
present, and future events, 3) confronting nonmonotonicity by include in the analysis?
How does the analyst judge the relative importance of
updating interpretations and assumptions in the face of new
the various data?
information, 4) handling missing or uncertain data, and 5)
How does the analyst judge the validity of the data?
focusing attention on events of special significance. On-going
How does the analyst handle incomplete, inconsistent,
research in temporal database management [23] would be
and unreliable data?
particularly relevant to the time-dependent nature of strategic
How does the analyst synthesize the problem de-
data.
scription with current and past data to formulate an
4) Truth Maintenance and Nonmonotonicity: The dynamic
assessment?
nature of CI and the general business environment require Do the cognitive processes vary among analysts?
the capability of storing data and assumptions that may later 2) To build a computational model of the cognitive pro-
be retracted as new information is received. As data and/or cesses of a CI analyst. Specifically:
assumptions are changed, the model must be able to propa- What knowledge representation and procedural ap-
gate those changes to all other dependent knowledge in or- proaches are most appropriate and should be included?
der to maintain the consistency of the knowledge base and How should analytical techniques from CI [91] be
resolve any inconsistencies that might arise [37]. Therefore, a incorporated? Can the output from them serve as input
cognitive model of decision making in volatile strategic en- to higher-level assessments?
vironments would likely benefit from research in dependency What new approaches should be developed in light of
networks and truth maintenance systems (TMS) [22], [27]. the unique nature of strategic assessment and decision
5) Analogical and Case-BasedReasoning: As described ear- making?
lier, interpretations of strategic data are often derived through 3) Ultimately to determine, given the foundation of a com-
analogous references to past experience. The work of Sullivan putational model and the insights into strategic decision
and Yates, described earlier, demonstrates the feasibility of us- making that it provides, the nature of computer support
ing an analogical approach in strategic DSS. Much opportunity required for CI. Specifically:
exists in extending this approach by incorporating theoretical Which aspects of expert behavior, gleaned from the
work in analogical [14], [15], [42] and case-based reasoning cognitive model, can be improved or enhanced through
[5], [7], [49], [60] within a model that can integrate current computer-based support? What types of problems do
situational data with stored experience, updating and revising these experts have, and what type of generalized
its knowledge base, and thus learning, as new information is support is required?
processed. Many of the theoretical subissues raised in other What parts of the task canhhould reasonably be del-
studies are directly related to aspects of strategic decision egated to a computer system?
SPANGLER: ARTIFICIAL INTELLIGENCEIN UNDERSTANDING STRATEGICDECISION-MAKING PROCESS 157

What type of computer-based support would facilitate -, “Modeling organizations as a social network of distributed
the communication and/or retrieval of information knowledge-based systems,” in Proc. Twenty-third Annu. Hawaii Int.
Con& Syst. Sei., 1990, pp. 271-280.
within this context? D. I. Cleland, W. R. King, “Competitive business intelligence systems,”
What should be the nature of the human-computer Business Horizons, vol. 18, no. 6, pp. 19-28, 1975.
interface? R. M. Cyert and J. G. March, A Behavioral Theory of the Firm. Engle-
wood Cliffs, NJ: Prentice-Hall, 1963.
How intrusive or assertive should the system be? F. David, “Computer-assisted strategic planning in small business,”
What should be the form of the inputs and outputs? J . Syst. Management, vol. 36, pp. 24-34, 1985.
How explicit and transparent should the system be R. Davis, and R. Smith, “Negotiation as a metaphor for distributed
problem solving,” Artif Intell., vol. 20, pp. 63-109, 1983.
in terms of its processing? What does the user need J. de Kleer, “An assumption-based TMS,” Artif Intell., vol. 28,
to know at each step of the analysis? How should pp. 127-162, 1986.
the system explain its behavior? T. Dean and D. McDermott, “Temporal data base management,” Art$
Intell., vol. 32, pp. 1-55, 1987.
Can a general support system for CI be developed, or D. Demetrius, “Expert systems and board level decisions,” Artificial
must it be tailored for a specific environment? Intelligence in Economics and Management, L. F. Pau, Ed. New York:
Can there be elements of each? Elsevier Science, 1986.
V. K. Dhar, “PLANET: An intelligent decision support system for
AI research provides models of human cognition that can the formulation and investigation of formal planning models,” Ph.D.
assist strategic planning researchers in better understanding the dissertation, Univ. of Pittsburgh, 1984.
V. Dhar, and H. E. Pople, “Rule-based versus structure-based models for
process of strategic decision making, and as the final objective explaining and generating expert behavior,” Commun. ACM, vol. 30,
above notes, might lead to improved notions of computer- pp. 542-555, 1987.
based support for strategic decision makers. Toward this end, J . Doyle, “A Truth maintenance system,” Artif Intell., vol. 12,
pp. 231-272, 1979.
AI modeling techniques offer a computational laboratory in B. K. Dutta, and W. R. King, “A competitive scenario modeling system,”
which proposed theories of strategic decision making can be Management Sei., vol. 26, no. 3, pp. 261-273, Mar. 1980.
developed and tested in a rigorous fashion. J. E. Dutton, L. Fahey, and V. K. Narayanan, “Toward understanding
strategic issue diagnosis,” Strategic Management., vol. 4, pp. 307-323,
1983.
A. Edwards, Expert Systems in Accounting. Englewood Cliffs, NJ:
ACKNOWLEDGMENT Prentice-Hall, 1990.
0. El Sawy, “Personal information systems for strategic scanning in
The author wishes to thank Dr. H. Pople for his valuable turbulent environments: Can the CEO go on-line?” MIS Quarterly,
insights and suggestions, as well as two anonymous reviewers pp. 53-60, Mar. 1985.
for their constructive comments. G. S . Elofson and B. R. Konsynski, “Supporting knowledge sharing in
environmental scanning,” in Proc. Twenty-third Annu. Hawaii Int. Conf
Syst. Sci., 1990, pp. 281-288.
K. A. Ericsson and H. A. Simon, Protocol Analysis: Verbal Reports as
REFERENCES Data. Cambridge, MA: MIT Press, 1984.
L. M. Fagan, “VM: Representing time-dependent relations in a medical
[I] R. Ackoff, Redesigning the future. New York: Wiley, 1974. setting,” Ph.D. dissertation, Stanford Univ., 1980.
[2] G. T. Allison, Essence of Decision: Explaining the Cuban Missile Crisis. J. Feinstein, “An expert system to prevent the disclosure of sensitive
Boston, MA: Little, Brown, 1971. information at the United States Environmental Protection Agency,”
[3] P. Anderson and S. Thorson, “Artificial intelligence based simulations in Artificial Intelligence in Economics and Management. L. F. Pau, Ed.
of foreign policy decision making,” Behavioral Sci., vol. 27, no. 2,
New York: Elsevier Science, 1986.
pp. 176-193, 1982.
H. Fersko-Weiss, “Personal computing at the top,” Business Week,
[4] H. I. Ansoff, “Competitive strategy analysis on the personal computer,”
J . Business Strategy, vol. 2, no. 3, pp. 28-36, 1986. vol. 9, pp. 68-71, 1985.
J. Fiksel and F. Hayes-Roth, Knowledge systems for planning support,”
[ 5 ] K.D. Ashley, “Modeling legal argument: Reasoning with cases and
hypotheticals,” Ph.D. dissertation, Univ. of Massachusetts, 1987. IEEE Expert, pp. 16-23, Fall 1989.
G. Fischer, A. Lemke, and T. Schwab, “Knowledge-based help sys-
[6] R. Axelrod, The Structure of Decision: Cognitive Maps of Political
Elites. Princeton. NJ: Princeton Univ. Press. 1976. tems,’’ in Proc. CHI’85 Conf Human Factors in Comput. Syst., ACM
W. M. Bain, “Case-based reasoning: A computer model of subjective SIGCHI, 1985, pp. 161-167.
assessment,” Ph.D. dissertation, Yale Univ., 1986. P. Fredericks and N. Venkatraman, “The rise of strategy support
S. Banerjee, “Reproduction of social structures: A n artificial intelligence systems,” Sloan Management Rev., pp. 47-54, Spring 1988.
model,” J. Conflict Resolution, vol. 30, no. 2, pp. 221-252, 1986. L. M. Fuld, Monitoring the Competition. New York: Wiley, 1988.
B. Bennett, C. Jones, A. Bullock, and P. Davis, “Main theater warfare C. S. Gadd and H. E. Pople, “Evidence from internal medicine teaching
modeling in the RAND strategy assessment system,” Tech. Rep., The rounds of the multiple roles of diagnosis in the transmission and testing
RAND Corp., Sept. 1988. of medical expertise,” in Diagnostic Monitoring of Skill and Knowledge
J. Bergqvist and P. Lounamaa, “Knowledge-based evaluation of strategic Acquisition, N. Frederikson, Ed. Hillsdale, NJ: Erlbaum, 1990.
investments,” in Economics and Artificial Intelligence. J. L. Roos, Ed. D. Gentner, “Structure mapping: A theoretical framework for analogy,”
New York: Permagon, 1987. Cognitive Sci., vol. 7, pp. 155- 170, 1983.
M. J. Bouwman, “Financial diagnosis: A cognitive model of the pro- M. Goslar, G. Green, and T. Hughes, “Decision support systems:
cesses involved,” Ph.D. dissertation, Carnegie-Mellon Univ., 1978. An empirical assessment for decision making,” Decision Sci., vol. 17,
B. Bruce, and D. Newman, “Interacting plans,” Cognitive Sei., vol. 2, pp. 79-91, 1986.
no. 3, pp. 195-233, 1978. [44] M. Goul, “The inclusion of expertise in a decision support system for
B. Buchanan and E. Shortliffe, Rule-based Expert Systems: The MYCIN strategic planning: An empirical study,” Ph.D. dissertation, Oregon State
Experiments of the Stanford Heuristic Programming Project. Reading, Univ., 1985.
MA: Addison-Wesley, 1984. [45] -, “On building expert systems for strategic planners: A knowledge
J. Carbonell, “A computational model of analogical problem solving,” engineer’s experience,” Inform. Management, vol. 12, pp. 131- 141,
in Proc. Seventh Int. Joint Con$ Artif Intell., IJCAI, Vancouver, B.C., 1987.
Canada, 1981, pp. 147-152. [46] B. J. Grosz and C. L. Sidner, “Attention, intentions, and the structure of
-, “Derivational analogy: A theory of reconstructive problem solv- discourse,” Computat. Linguist., vol. 12, no. 3, pp. 175-204, 1986.
ing and expertise acquisition,” in Machine Learning, R. S. Michalski, [47] N. Hall, “Diagnosing problems with the user interface for a strate-
J. Carbonell, and T. Mitchell, Eds. Palo Alto, CA: Tioga, 1986. gic planning fuzzy DSS,” IEEE Trans. Syst., Man, Cybern., vol. 18,
D. Carlson and S. Ram, “Hyperintelligence: The next frontier,” Com- pp. 638-646, 1988.
mun. ACM, vol. 33, pp. 311-321, Mar. 1990. [48] R. G. Hamermesh, Making Strategy Work. New York: Wiley, 1986.
 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 3, NO. 2, JUNE 1991

K. J. Hammond, Case-based Planning: Hewing Planning as a Memory [81] A. Newell and H.A. Simon, Human Problem Solving. Englewood
Task. Boston, MA: Academic, 1989. Cliffs, NJ: Prentice-Hall, 1972.
D. Hertz, “Expert systems for the analysis and synthesis of strate- [82] -, “Computer science as empirical inquiry: Symbols and search,”
gic policy,” in Economics and Artificial Intelligence, J. L. Roos, Ed. Commun. ACM, pp. 113-126, Mar. 1976.
Elmsford, NY: Permagon, 1987. [83] J. M. Peters, “A knowledge based model of inherent audit risk assess-
C. Holloway and J. Pearce, “Computer-assisted strategic planning,” ment,” Ph.D. dissertation, Univ. of Pittsburgh, 1989.
Long Range Planning, vol. 15, pp. 56-63, 1982. [84] D. Pham, Expert Systems in Engineering. New York: Springer-Verlag,
G. Houdeshel and H.J. Watson, “The management information and 1988.
decision support (MIDS) system at Lockheed-Georgia,” in The Man- [85] H. E. Pople, “Heuristic methods for imposing structure on ill-structured
agement of Information Systems, P. Gray, W. King, E. McLean, and problems: The structuring of medical diagnosis,” in Artificial Intelli-
H. Watson, Eds. New York: Dryden, 1989. gence in Medicine, P. Szolovits, Ed. Boulder CO: Westview, 1982.
B. Hurnpert and P. Holley, “Expert systems in finance planning,” Expert 1861 -, “Evolution of an expert system: From Internist to Caduceus,” in
Syst., vol. 5, no. 2, pp. 88-101, 1988. Artificial Intelligence in Medicine. I. De Lotto, and M. Stefanelli, Eds.
H. A. Kautz and J. F. Allen, “Generalizedplan recognition,” in Proc. 5th Amsterdam, The Netherlands: Elsevier Science, 1985.
Nut. Con$ Artif Intell., AAAI, 1986, pp. 32-37. 1871 -, An Artificial Intelligence System for Process Monitoring, Situation
P. Keen, “Decision support systems: The next decade,”Decision Support Assessment and Response Planning. Tech. Rep., Final Rep. of SBIR
Syst., vol. 3, pp. 253-265, 1987. Phase 1 Project, NASA, 1989.
B. E. Keiser, “Practical competitor intelligence,” Planning Rev., vol. 15, 1881 M. Porter, Competitive Strategy. New York: The Free Press, 1980.
no. 5, 1987. 1891 -, CompetitiveAdvantage. New York The Free Press, 1985.
W. R. King, “Achieving the potential of decision support systems,” J. 1901 J. E. Prescott and D. C. Smith, “A project-based approach to competitive
Business Strategy, vol. 3, pp. 84-91, 1983. analysis,” Strategic Management J., vol. 8, pp. 411-423, 1987.
H. Klein and W. Newman, “How to Use SPIRE: A systematic procedure 1911 J. E. Prescott and 3. H. Grant, “A manager’s guide to evaluating com-
for identifying relevant environmentsfor strategic planning,” J. Business petitive analysis techniques,” Interfaces vol. 18, no. 3, pp. 10-22, 1988.
Strategy, pp. 32-45, Summer 1980. 1921 J. Quinn, “Strategic change: Logical incrementalism,” Sloan Manage-
M. Kochen and M. Min, “Intelligence for strategic planning,” in ment Rev., vol. 20, pp. 7-21, 1978.
Proc. Seventh Int. Workshop Expert Syst. Appl., Avignon, France, 1987, [93] B. Reimann, “Decision support systems: Strategic management tools for
pp. 41-57. the eighties,” Business Horizons, vol. 28, pp. 71-77, 1985.
J. L. Kolodner, Retrieval and Organizational Strategies in Conceptual [94] J. Rockart and D. Delong, Executive Support Systems. Homewood, F L
Memory: A Computer Model. Hillsdale, NJ: Erlbaum, 1984. Dow-Jones-Irwin, 1988.
T. J. Laffey, P.A. Cox, J. L. Schmidt, S. M. Kao, and J. Y. Read, “Real- [95] J.I. Rodriguez, “The design and evaluation of a strategic issue com-
petitive information system,” Ph.D. dissertation, Univ. of Pittsburgh,
.
time knowledge-based systems,” AI Mag., vol. 9, no. 1, pp. 27-45,
nnn
IYUU.
1977.
J. LaFrance, an expert system for commercial use,,, Tech, [96] J. Rosenschein, “Rational interaction: Cooperation among intelligent
agents,” Ph.D. dissertation, Stanford Univ., 1985.
Rep., MPSI Corp., Tulsa, OK, 1989. Cited in [98].
[971 J. Rothenberg, S. Narain, R. Steeb, C. Hefley, and N. Shapiro,
M. Lebowitz, “Generalization from natural text,” cognitive
“Knowledge-based simulation: An interim report,” Tech. Rep., The
Sci., vol. 7, pp. 1-40, 1983.
V. Lesser, J. Pavlin, and E. Durfee, “Approximate processing in real- RAND Corp., July 1989.
[98] A. Round, simulation,,7in The ofArtificial
time problem solving,” AI Mag., vol. 9, pp. 49-61, 1988.
Intelligence, Vol. w,A. Barr, P. Cohen, and E. Feigenbaum, E&.
J. Lynch and D. Hertz, ‘‘The anatomy of AI,,, in proc. TIMs,oRsA 1985 Reading, M A Addison-Wesley, 1989.
Annu. Meet., (unpublished), 1985. Cited in [SO].
1991 R. Sabherwal and V. Grover, “Computer support for strategic decision-
T. ‘‘Modeling coordination in organizations and markets,”
making processes: Review and analysis,” Decision Sci., vol. 20, no. 1,
Management Sci., vol. 33, no. 10, pp. 1317-1332, 1987.
R’ O’ pp. 54-76, 1989.
and I. I‘ Mitroff, chalrenging Assump- [100] C, F, Schmidt, N, S. Sridharan, and J, L,Goodson, “The plan recognition
tions. New York Wiley, 1981.
problem,” Artif: Intell., vol. 11, pp. 45-83, 1978.
M’ and p‘ “Beyond garbage An AI Of [loll C. R. Schwenk, “Cognitive simplification processes in strategic decision
organizational choice,” Administrative Sci. Quarterly, vol. 34, no. 1, making,” Strategic Management J., vol. 5 , pp. 111-128, 1984.
pp. 38-67, 1989.
J. McCarthy, “Epistemological problems of artificial intelligence,” in 11021 -, ‘The cognitive perspective on strategic decision making,” J .
Management Studies, vol. 25, no. 1, pp. 41-55, 1988.
Readings in Knowledge Representation, R. Brachman and H. LevesqUe, [lo31 E. Segev and P. Gray, “Integrating an expert system and DSS for
Eds. San Mateo, C A Morgan-Kaufman, 1985, pp. 23-30. strategic decision support,” Inform. Resources Management, vol. 2,
D. McCue and V. Lesser, “Focusing and constraint management in pp, 1-12, 1989,
intelligent interface design,” Tech. Rep. COINS 83-36, Univ. of Massa- [io41 D. Shlapak, w, Schwabe, M, brell, and y, B ~ ~ - ‘<The H ~RAND ~ ~ ~ ,
chusetts, 1983. strategy assessmentsystem’s Green agent model of third-country behav-
H. E. Meyer, Real World Intelligence: Organized Information for Exec- ior in superpower crises and conflict,” Tech. R ~ ~The , , RAND carp,,
utives. New York Weidenfeld and Nicolson, 1987. Sept. 1986.
H. Mineberg, D. and A. Theoret, ‘The structure Of [lo51 H. A. Simon, The New Science of Management Decision. New York
unstructured decision processes,”Administrative Sci, Quarterly, vol. 21, Harper and Brothers, 1960.
pp. 246-275, 1976. [lo61 D. C. Smith and J. E. Prescott, “Demystifying competitive analysis,”
H. Mmtzberg and J. Waters, “Tracking strategy in the entrepreneurial planning R ~ ~vol. . , 15, no. 5 , pp. 8-13, 1987.
firm,” Acad Management., vol. 25, pp. 465-499, 1982. [lo71 C. Sullivan and C. Yates, “Reasoning by analogy-A tool for business
1.1. Mitroff and J.R. Emshoff, “On strategic assumption-making: A planning,” Sloan Management Rev., pp. 55-60, Spring 1988.
dialectical approach to policy and planning.” Acad Management Rev. [lo81 J. Syed and E. Tse, “An integrated consulting system for competitive
vol. 4, no. 4, pp. 1-12, 1979. analysis of the polyolefins industry,” in Proc. Seventh I t . Workshop
I. I. Mitroff, J. R. Emshoff, and R.H. Kilmann, “Assumption analysis: A Expert syst. Avignon, France, 1987, pp. 689-709.
methodology for strategic problem solving,” Management Sci., vol. 25, [lo91 P. Szolovits, Artificial Intelligence in Medicine. Boulder, CO: West-
no. 6, pp. 583-593, 1979. view, 1982.
R. J. Mockler, “Computer information systems and strategic corporate [110] M. Vasarhelyi, Artificial Intelligence in Accounting andduditing. New
planning,’’ Business Horizons, vol. 30, pp. 32-37, 1987. York: Markus Wiener, 1989.
-, Knowledge-basedSystemsfor Strategic CorporatePlanning. Ox- [ l l l ] D. A. Waterman and A. Newell, “Protocol analysis as a task for artificial
ford, OH: The Planning Forum, 1987. intelligence,” Artif Intell., vol. 2, pp. 285-318, 1971.
R. J. Mockler and D. G. Dologite, “Developing knowledge-based sys- [112] L. Widman and K. Loparo, “Artificial intelligence, simulation and
tems for strategic corporate planning,” Long Range Planning, vol. 21, modeling: A critical survey,” in Artificial Intelligence, Simulation and
no. 1, DD. 97-102, 1988. Modeling, L. Widman, K. Louaro, and N. Nielsen, Eds. New York
[79] R. J. Mkkler, Knowledge-based Systems for Strategic Planning. En- Wiley, 1989.
glewood Cliffs, NJ: Prentice-Hall, 1989. [113] D. D. Woods, H. E. Pople, and E. M. Roth, “The cognitive environment
[80] W. R. Nelson, “REACTOR An expert system for diagnosis and treat- simulation as a tool for modeling human performance and reliability,
ment of nuclear reactor accidents,” in Proc. Second Nut. Con$ Artif Tech. Rep. NUREG-CR-5213, U.S. Nuclear Regulatory Commission,
Intell., AAAI,1982, pp. 296-301. 1988.
SPANGLER: ARTIFICIAL INTELLIGENCE IN UNDERSTANDING STRATEGIC DECISION-MAKING PROCESS 159

S. Yo0 and L. Digman, “Decision support system: A new tool for William E. Spangler is currently a Ph.D. candidate
strategic management,” Long Range Planning, vol. 20, pp. 114- 124, in the Joseph M. Katz Graduate School of Business
1987. at the University of Pittsburgh, Pittsburgh, PA. He
L. Young, “A systems architecture for supporting senior management’s received the M.B.A. degree from the University of
messy tasks,” Inform. Management, vol. 13, pp. 85-94, 1987. Hawaii, and has seven years of experience in the
L. Zadeh, “Making computers think like people,” IEEE Spectrum, computer industry.
pp. 26-32, Aug. 1984. He has worked at Westinghouse Credit Corpora-
tion and Burroughs (now Unisys) Corporation, and
has consulted for a number of other companies,
including several Fortune 500 firms. His research
interests focus on the application of artificial intel-
ligence to strategic decision making.