INTRODUCTION TO DECISION SUPPORT SYSTEMS

1. INTRODUCTION

In very general terms, a decision support system (DSS) is a system that supports technological and managerial decision making by assisting in the organization of knowledge about ill-structured, semi- structured, or unstructured issues. For our purposes, a structured issue is one that has a framework with elements and relations between them that are known and understood within a context brought by the experiential familiarity of a human assessing the issue or situation. The three primary com- ponents of a decision support system are generally described as:

• A database management system (DBMS)

• A model-based management system (MBMS)

• A dialog generation and management system (DGMS)

The emphasis in the use of a DSS is upon providing support to decision makers to increase the effectiveness of the decision making effort. This need should be a major factor in the definition of requirements for and subsequent development of a DSS. A DSS is generally used to support humans in the formal steps of problem solving, or systems engineering (Sage 1992, 1995; Sage and Rouse 1999a; Sage and Armstrong 2000), that involve:

• Formulation of alternatives

• Analysis of their impacts

• Interpretation and selection of appropriate options for implementation

Efficiency in terms of time required to evolve the decision, while important, is usually secondary to effectiveness. DSSs are intended more for use in strategic and tactical situations than in operational situations. In operational situations, which are often well structured, an expert system or an accounting system may often be gainfully employed to assist novices or support the myriads of data elements present in these situations. Those very proficient in (experientially familiar with) operational tasks generally do not require support, except perhaps for automation of some routine and repetitive chores. In many application areas the use of a decision support system is potentially promising, including management and planning, command and control, system design, health care, industrial management, and generally any area in which management has to cope with decision situations that have an unfamiliar structure.

Taxonomies for Decision Support

Numerous disciplinary areas have contributed to the development of decision support systems. One area is computer science, which provides the hardware and software tools necessary to implement decision support system design constructs. In particular, computer science provides the database design and programming support tools that are needed in a decision support system. Management science and operations research provides the theoretical framework in decision analysis that is nec- essary to design useful and relevant normative approaches to choice making, especially those con- cerned with systems analysis and model base management. Organizational behavior and behavioral and cognitive science provide rich sources of information concerning how humans and organizations process information and make judgments in a descriptive fashion. Background information from these areas is needed for the design of effective systems for dialog generation and management systems. Systems engineering is concerned with the process and technology management issues associated with the definition, development, and deployment of large systems of hardware and software, in- cluding systems for decision support.

Many attempts have been made to classify different types of decisions. Of particular interest here is the decision type taxonomy of Anthony, which describes four types of decisions (Anthony 1965; Anthony et al. 1992):

1. Strategic planning decisions are decisions related to choosing highest level policies and ob- jectives, and associated resource allocations.

2. Management control decisions are decisions made for the purpose of ensuring effectiveness in the acquisition and use of resources.

3. Operational control decisions are decisions made for the purpose of ensuring effectiveness in the performance of operations.

4. Operational performance decisions are the day-to-day decisions made while performing op- erations.

Figure 1 illustrates how these decisions are related and how they normatively influence organi- zational learning. Generally, low-consequence decisions are made more frequently than high- consequence decisions. Also, strategic decisions are associated with higher consequences and are likely to involve more significant risk and therefore must be made on the basis of considerably less perfect information than are most operational control decisions.

A decision support system should support a number of abilities. It should support the decision maker in the formulation or framing or assessment of the decision situation in the sense of recognizing needs, identifying appropriate objectives by which to measure successful resolution of an issue, and generating alternative courses of action that will resolve the needs and satisfy objectives. It should also provide support in enhancing the decision maker’s abilities to assess the possible impacts of these alternatives on the needs to be fulfilled and the objectives to be satisfied. This analysis capability must be associated with provision of capability to enhance the ability of the decision maker to provide an interpretation of these impacts in terms of objectives. This interpretation capability will lead to evaluation of the alternatives and selection of a preferred alternative option. These three steps of formulation, analysis, and interpretation are fundamental for formal analysis of difficult issues. They are the fundamental steps of systems engineering and are discussed at some length in Sage (1991,

1992, 1995), from which much of this chapter is derived. It is very important to note that the purpose of a decision support system is to support humans in the performance of primarily cognitive infor- mation processing tasks that involve decisions, judgments, and choices. Thus, the enhancement of information processing in systems and organizations (Sage 1990) is a major feature of a DSS. Even though there may be some human supervisory control of a physical system through use of these decisions (Sheridan 1992), the primary purpose of a DSS is support for cognitive activities that involve human information processing and associated judgment and choice. Associated with these three steps must be the ability to acquire, represent, and utilize information or knowledge and the ability to implement the chosen alternative course of action.

The extent to which a support system possesses the capacity to assist a person or a group to formulate, analyze, and interpret issues will depend upon whether the resulting system should be called a management information system (MIS), a predictive management information system (PMIS), or a decision support system. We can provide support to the decision maker at any of these several levels, as suggested by Figure 2. Whether we have a MIS, a PMIS, or a DSS depends upon the type of automated computer-based support that is provided to the decision maker to assist in reaching the decision. Fundamental to the notion of a decision support system is assistance provided in assessing the situation, identifying alternative courses of action and formulating the decision sit- uation, structuring and analyzing the decision situation, and then interpreting the results of analysis of the alternatives in terms of the value system of the decision maker. In short, a decision support system provides a decision recommendation capability. A MIS or a PMIS does not, although the information provided may well support decisions.

In a classical management information system, the user inputs a request for a report concerning some question, and the MIS supplies that report. When the user is able to pose a ‘‘what if?’’ type question and the system is able to respond with an ‘‘if then’’ type of response, then we have a

predictive management information system. In each case there is some sort of formulation of the issue, and this is accompanied by some capacity for analysis. The classic MIS, which needs only to be able to respond to queries with reports, is composed of capabilities for data processing, structured data flows at an operational level, and preparation of summary reports for the system user. The predictive management system would also include an additional amount of analysis capability. This might require an intelligent database query system, or perhaps just the simple use of some sort of spreadsheet or macroeconomic model.

To obtain a decision support system, we would need to add the capability of model-based man- agement to a MIS. But much more is needed, for example, than just the simple addition of a set of decision trees and procedures to elicit examination of decision analysis-based paradigms. We also need a system that is flexible and adaptable to changing user requirements such as to provide support for the decision styles of the decision maker as these change with task, environment, and experiential familiarity of the support system users with task and environment. We need to provide analytical support in a variety of complex situations. Most decision situations are fragmented in that there are multiple decision makers and their staffs, rather than just a single decision maker. Temporal and spatial separation elements are also involved. Further, as Mintzberg (1973) has indicated so very well, managers have many more activities than decision making to occupy themselves with, and it will be necessary for an appropriate DSS to support many of these other information-related functions as well. Thus, the principal goal of a DSS is improvement in the effectiveness of organizational knowledge users through use of information technology. This is not a simple objective to achieve as has been learned in the process of past DSS design efforts.

Frameworks for the Engineering of Decision Support Systems An appropriate decision support system design framework will consider each of the three principal components of decision support systems—a DBMS, an MBMS, and a DGMS—and their interrela- tions and interactions. Figure 3 illustrates the interconnection of these three generic components and illustrates the interaction of the decision maker with the system through the DGMS. We will describe some of the other components in this figure soon.

Sprague and Carlson (1992), authors of an early, seminal book on decision support systems, have indicated that there are three technology levels at which a DSS may be considered. The first of these is the level of DSS tools themselves. This level contains the hardware and software elements that enable use of system analysis and operations research models for the model base of the DSS and the database elements that comprise the database management system. The purpose of these DSS tools is to design a specific DSS that is responsive to a particular task or issue. The second level is that of a decision support system generator. The third level is the specific DSS itself. The specific DSS may be designed through the use of the DSS tools only, or it may be developed through use of a generic DSS generator that may call upon elements in the generic MBMS and DBMS tool repository for use in the specific DSS.

Often the best designers of a decision support system are not the specialists primarily familiar with DSS development tools. The principal reason for this is that it is difficult for one person or small group to be very familiar with a great variety of tools as well as to be able to identify the

requirements needed for a specific DSS and the systems management skills needed to design a support process. This suggests that the decision support generator is a potentially very useful tool, in fact a design level, for DSS system design. The DSS generator is a set of software products, similar to a very advanced generation system development language, which enables construction of a specific DSS without the need to formally use micro-level tools from computer science and operations re- search and systems analysis in the initial construction of the specific DSS. These have, in effect, already been embedded in the DSS generator. A DSS generator contains an integrated set of features, such as inquiry capabilities, modeling language capabilities, financial and statistical (and perhaps other) analysis capabilities, and graphic display and report preparation capabilities. The major support provided by a DSS generator is that it allows the rapid construction of a prototype of the decision situation and allows the decision maker to experiment with this prototype and, on the basis of this, to refine the specific DSS such that it is more representative of the decision situation and more useful to the decision maker. This generally reduces, often to a considerable degree, the time required to engineer and implement a DSS for a specific application. This notion is not unlike that of software prototyping, one of the principal macro-enhancement software productivity tools (Sage and Palmer 1990) in which the process of constructing the prototype DSS through use of the DSS generator leads to a set of requirements specifications for a DSS that are then realized in efficient form using DSS tools directly.

The primary advantage of the DSS generator is that it is something that the DSS designer can use for direct interaction with the DSS user group. This eliminates, or at least minimizes, the need for DSS user interaction with the content specialists most familiar with micro-level tools of computer science, systems analysis, and operations research. Generally, a potential DSS user will seldom be able to identify or specify the requirements for a DSS initially. In such a situation, it is very advan- tageous to have a DSS generator that may be used by the DSS engineer, or developer, in order to obtain prototypes of the DSS. The user may then be encouraged to interact with the prototype in order to assist in identifying appropriate requirements specifications for the evolving DSS design.

The third level in this DSS design and development effort results from adding a decision support systems management capability. Often, this will take the form of the dialog generation and manage- ment subsystem referred to earlier, except perhaps at a more general level since this is a DGMS for DSS design and engineering rather than a DGMS for a specific DSS. This DSS design approach is not unlike that advocated for the systems engineering of large scale systems in general and DSS in particular.

There are many potential difficulties that affect the engineering of trustworthy systems. Among these are: inconsistent, incomplete, and otherwise imperfect system requirements specifications; sys- tem requirements that do not provide for change as user needs evolve over time, and poorly defined management structures. The major difficulties associated with the production of trustworthy systems have more to do with the organization and management of complexity than with direct technological concerns. Thus, while it is necessary to have an appropriate set of quality technological methods and tools, it is also very important that they be used within a well chosen set of lifecycle processes and set of systems management strategies that guide the execution of these processes (Sage 1995; Sage and Rouse 1999a).

Because a decision support system is intended to be used by decision makers with varying ex- periential familiarity and expertise with respect to a particular task and decision situation, it is es- pecially important that a DSS design consider the variety of issue representations or frames that decision makers may use to describe issues, the operations that may be performed on these repre- sentations to enable formulation analysis and interpretation of the decision situation, the automated memory aids that support retention of the various results of operations on the representations, and the control mechanisms that assist decision makers in using these representations, operations, and memory aids. A very useful control mechanism results in the construction of heuristic procedures, perhaps in the form of a set of production rules, to enable development of efficient and effective standard operating policies to be issued as staff directives. Other control mechanisms are intended to encourage the decision maker to personally control and direct use of the DSS and also to acquire new skills and rules based on the formal reasoning-based knowledge that is called forth through use of a decision support system. This process independent approach toward development of the necessary capabilities of a specific DSS is due to Sprague and Carlson (1982) and is known as the ROMC approach (representations, operations, memory aids, and control mechanisms). Figure 3 illustrates the ROMC elements, together with the three principal components (DBMS, MBMS, and DGMS) of a DSS.