Today the pressures on an organization stem from many sources: competition, regulations, economics, public opinion, as well as the more traditional financial goals (whether shareholder or managerial wealth maximization). The traditional focus in America has been on maximizing single measures (e.g., sales growth, asset growth, return on sales). Other cultures view the interests of the shareholder as relatively unimportant, while the interests of the society dominate. If U. S. firms are to effectively compete against foreign firms, the U. S. firms must not only understand the forces which motivate the foreign firm, but also compete under those same sets of pressures.
To achieve this, more advanced tools for making decisions are needed so that managers in other countries can make decisions based on the local conditions while still communicating effectively with managers in the home country. This means that subsidiaries in each country must have independent sets of goals while maintaining the information needed by corporate headquarters. The computerized spreadsheet is one such tool, but it has many limitations, particularly in expansion of capabilities and tailoring the tool to the individual manager.
The increased complexity of managing in a matrix organization involving international partners increases the importance of keeping a large part of the long term planning at the local level. The complexity of the international environment requires knowledge which is not usually in the manager's background (e.g. currency rates, bartering, trade restrictions, different laws, etc.). The specialized knowledge and computational tools can be provided by combining a natural language interface and the use of an expert system devoted to the specific needs of the manager.
The purpose of this thesis is to examine the development of computer based models in corporations, identify the trends and the factors involved in the development of the models, and to suggest some ways in which AI technology can be used to reduce the difficulty in getting managers to use new technologies.
This paper is divided into two parts. Part one attempts to identify the development of the corporate model and some of the trends and problems involved in employing corporate models. These trends focus on changes in the organizational environment, organizational dynamics, and the available technology. Part two is a simple implementation of a personal decision support system. The proposed implementation utilizes a simple natural language interpreter and an expert system. The discussion and summary at the end of part two contains some commentary on what other extended uses can be made of Artificial Intelligence (AI) technology in the future.
The following section is by no means an exhaustive survey of the expanding field of computer based modeling. The diversity of terminology and methods of analysis are indicative of a field which has not yet reached the point of intellectual exhaustion. This diversity is an impediment to a synthesis of the information into commonly agreed upon constructs. In the following sections, I have attempted to bring together the elements of organizational behavior, computer science, and finance which form the driving force behind the use of interactive decision making tools in the corporate environment.
-1- THE NEED FOR A PERSONAL DECISION SUPPORT SYSTEM (PDSS)
-1.1- A DEFINITION OF MODELING
It may be useful to begin with a definition of a model.
Models represent an abstraction of reality - they represent what things were like , what they could be like, or what they should be like. Models are designed to clarify certain aspects of a problem or problem area; they are supposed to highlight certain important relationships and certain key interactions. A well-defined model, in abstracting the essence of the model should clarify it, facilitate analysis of the constraints and potentials, and indicate some of the information needed. [46 p(5]
A more general definition of a model is a subset of relationships in the real world. Since the subset does not represent all of reality, the validity of the model is limited. There are two types of validity, content validity and construct validity. The former measures the predictive validity of the model, while the latter measures the stability of the construct. Many authors consider only the content validity of a model. However, the importance of the construct validity lies in the fact that any construct is a collection of individual variables which are placed under one heading. The relationships of these variables change over time, and hence the validity of the construct changes over time. An example might make the importance of construct validity clearer.
The basket of goods used in the consumer price index had been stable over a period of years. However, goods typically consumed by the average member of the population changed as the average age and income of the population changed; they were also affected by changes in technology and individual tastes. The construct validity of the consumer price index (i.e., that the goods in the index actually represented general consumption) had changed although the content validity of the index (the price of the goods in the index varied against the base period) remained unchanged.
The importance of the recognition of both types of validity by the modeler is that, at any moment in time , it is not possible to determine, by purely statistical tests, whether the change is due to changes of the variables comprising the construct or a change in the relationship of the variables within the construct. As accountants often say, in the long run, all costs are variable. The isolation of relevant variables and their relationship requires an intimate knowledge of the system being modeled.
What purpose does a model serve then? It provides a formal method for transforming data into useful information for decisions. The five classes of information specified by Oxenfeldt [42, p.27] are descriptive, explanatory, predictive, evaluative, and innovative. The utility of a model lies in how much information it provides in one or more of these categories. The flexibility of the model lies in how easily it can be altered to increase the information provided in one or more of these classes. The model is therefore a decision support system.
-1.2- TYPES OF MODELS
Decision support systems can be classified into seven distinct types [2] as follows:
1.file drawer systems
2.data analysis systems
3.analysis information systems
4.accounting models
5.representational models
6.optimization models
7.suggestion models
The first three focus on providing descriptive information and are data oriented, while the last four are model oriented and are intended to provide explanatory, predictive, evaluative, and innovative information. Meyer [38] divides models into two categories -stochastic and optimizing- based on the major technique employed. Meyers focuses on computational (accounting) models for financial planning and suggests integrating optimizing and/or stochastic techniques in the accounting model. According to Alter [3], the key usage problem in the latter three models is user understanding, while the accounting models are difficult to integrate into the planning process.
Which type of model to employ depends on the information needs of the decision maker and may change over time. However, the model must be able to be integrated into some standardized whole for higher level planning. The modeling tool must be capable of handling all four types of models and should be modifiable by the decision maker.
The types of applications are clearly related to the characteristics of different models. Bryant [10] suggests four key features which distinguish models:
1. time horizon
2. level of detail
3. desired accuracy
4. ease of calculation
These four features determine the goodness of fit between the method used and the problem situation.
-1.3- LINKING THE MODEL TO THE TYPE OF JOB PERFORMED
The Critical Success Factor (CSF) analysis developed at MIT was used by Rockart [48] to identify the factors which were critical to the achievement of the objectives of CEO's and then to identify the prime measures of those factors (e.g., an objective might be making an acquisition of a firm through a stock trade, a CSF might be the image of the corporation in the financial markets, and the prime measure might be the PE ratio). Rockart divides CSF's into two categories, monitoring (associated with pressure for current results) and building (associated with long term adaptation). This approach recently was applied successfully at one large company [54].
The CSF analysis is job-specific. The factors and measures can be specified for every position in an organization. The information needed for any given position may draw on existing databases or may require the development of a specialized database which may not contain information useful to others in the organization.
-1.4- MODELING AND THE DISTRIBUTED PROCESSING SYSTEM
Scherr describes a distributed processing system as the implementation of a related set of programs across two or more processing nodes [51]. The individual positions of responsibility within an organization and the CSF's which define the responsibility of each position can be defined as a node (referred to as a responsibility center in the accounting literature). This node can be examined in terms of the information processing which occurs at that center. To the extent to which this node is personalized (e.g., with the personal computer), the node can be examined as a DDP center. This would require that access to central databases is possible, as well as the maintenance of a local database. The processing software used in each system as well as the hardware chosen would have to be compatible not only with the central system, but also directly with other DDP nodes.
Why distributed processing? Ackoff [1] describes both organisms and organizations as purposeful systems, but notes that organisms do not contain purposeful elements. It is misleading, therefore, to anthropomorphize organizations or to treat individuals as organs of the organization, possessing no wills of their own. Since one manager alone constitutes a complete system, the information or feedback system should exist at that level with a matching degree of autonomy.
-1.5- THE PROCESS OF DECISION MAKING
There are five distinct views of decision making, according to Keen and Morton [31]:
1. the rational manager view
2. the satisficing process oriented view (H.A. Simon)
3. the organizational procedures view
4. the political view
5. the individual differences perspective
Thus a flexible software tool for decision support should be able to accommodate all five perspectives. This is not without difficulties.
These five conceptions of the decision process range from entirely normative to entirely descriptive. Each approach highlights key issues and often directly contradicts some other one. It is obviously not easy for DDS designers to pick any one theory and they may well view this whole issue as theological casuistry. We suggest that the best way to use or to synthesize these very divergent viewpoints is by adopting a diagnostic perspective. [31, p.62]
Determining which type of model will serve the individual manager the best requires an intimate knowledge of the situation surrounding the decision to be made. This intimate knowledge is difficult for the manager to communicate in sufficient detail to the model builder in order for it to be designed into the system. The appropriate model may even vary from one time to the next if the circumstances surrounding the decision change. Major changes, such as reorganizations, which might change the dynamics of the decision, would affect a large number of managers. Even if the system could be changed as necessary, the experts could hardly be available to many managers simultaneously.
-1.6- THE PRACTICING MANAGER AS MODELER
The most appropriate modeler is therefore the decision maker. This leads to several problems. The first is the relative modeling sophistication of the decision maker. Frequently decisions are made using informal decision rules which the decision maker is hard put to identify, let alone put in quantifiable f'orm. Also, the decision maker may feel his job security and autonomy may be reduced if he reveals too much about how he makes decisions. For this reason the decision maker may avoid giving entirely candid answers to the questions the model builder may ask.
The conflict is identified by Briscoe [9] as:
Employees see themselves as individuals... They make decisions about how to act at work based on expectations that certain behavior will lead to satisfaction of their needs, desires, and goals... Organizations see people as resources... The major problem for organizations lies at that interface between employees as humans and employees as resources. [9, p.79]
The implication for the DSS design is that the locus of responsibility and decision making should be as close to the individual level as possible. Job satisfaction, in itself, is not the goal of the decision support system (DSS). The focus should be on the decision.
-1.7- IMPROVING THE MANAGER/ORGANIZATION INTERFACE
The traditional employee as resource approach is demonstrated by Williamson [55]. He classifies skills contributed by employees to the organization in terms of the degree of specialization and the ability of the firm to measure the output of an individual. Williamson ignores the individual entirely in his analysis. He presents an intellectually appealing approach to organizational theory, but fails to deal with individuals who comprise the firm. The transaction cost approach is used to explain the motivation behind the forward integration of firms in terms of governance costs and production costs. Thus, as the specificity of job skills and the difficulty of metering the output from those jobs increase, the organization is motivated to maintain those skills within.
If the environment changes rapidly, the skills that are economical to maintain within the organization change. The result is that costly highly- skilled labor must be hired or the current employees must be retrained, which in turn lowers productivity. Where change is constantly occurring, the result is higher overall labor costs and diseconomies of scale.
According to Ouchi [41] there are three basic mechanisms of control: markets, bureaucracies, and clans. The information required by markets is prices, by bureaucracies is rules, by clans is traditions. The decision support system for market based systems is mechanistic and requires no authority figure because the information is extracted directly from the environment. Markets fail when bounded rationality affects the judgments. Bureaucracies provide the additional information in a stable form, but themselves fail when the tasks become unique or integrated under single individuals. Information is again the critical element of competing effectively. The information must be tailored to the environment in which the organization competes.
Changes in the environment might affect several subsystems that are competing for the same scarce resources of the organization. This would result in conflicts that could not be resolved by any centralized authority. This is perhaps just another way of saying that the result of the failure of mass marketing will lead to the growth of matrix organizations and result in the politicization of behavior. Again this indicates that the appropriate DSS for individual organizations must be tailored to the environment, and the structural and administrative elements of the organization.
-1.8- A PERSONAL DECISION SUPPORT SYSTEM FOR MNC CONTROL
Hofstede [27] developed four dimensions which vary between organizations in different countries. These dimensions are: power distance, uncertainty avoidance, individualism-collectivism, and masculinity-femininity. Groups of countries tend to cluster together along several dimensions. These similarities and differences should riot be ignored in choosing the DSS which is appropriate for each country. For example, the extreme score of individuality and small power distance toleration indicates that in the U. S., the DSS should be individualized to reduce the conflict in the interface between the employee as individual and the employee as resource.
In less developed countries (LDC's), the power distance tolerated is much larger and the collectivist attitude is much stronger. An aggregate DSS would be appropriate in LDC's, relying on the social structure to provide the appropriate rewards and punishments; thus, the cultural assumptions in MBO programs, according to Hofstede, make this type of control mechanism inappropriate. The differences in the information requirements of organizations based in different cultures demonstrates the need for a flexible DSS.
The power distance, the degree of difference between the power held by a subordinate and his supervisor, affects the type of management control which is acceptable in an organization. The differences in the other dimensions identified by Hofstede identify the cultural context in which decisions are made and hence what information is needed to make decisions. It is the differences in acceptable methods of control in different cultures that makes MNC control such a complex problem. Evans [19] notes that the multinational corporation (MNC) is faced with the unstructured problems of control at a distance and the scale of investment in other countries. Investment in other countries is difficult to control when causal relationships (e.g., the degree to which the MNC affects the foreign government and the converse) are difficult to establish and the effects of changes are unpredictable (e.g., a political revolution). Decisions in the MNC are not well suited to numerical analysis, but might benefit from the use of an expert system, employing Williamson's transaction cost analysis methodology to examine such questions as make or buy decisions in the host country. Since this decision would have to be made at the individual product level , the decision support system would have to be available on an individual basis. General policy directives from MNC headquarters are hardly flexible enough to allow effective decisions to be made.
A taxonomy for MNC control of subsidiaries is given by Doz and Prahalad [16]. To facilitate rapid response to local changes the manager in the foreign subsidiary must have the autonomy to make strategic decisions locally. This means the effective local tool would have a low continuity cost and little top management ongoing support. The appropriate areas for local control are therefore in data management. Specifically, the uses of and changes in the information systems and the budgeting process should be localized while those control mechanisms that require top management involvement and high continuity costs, such as career paths and the patterns of specialization, should remain centralized at MNC headquarters. A personal DSS would facilitate strategic control over foreign subsidiaries by allowing rapid, flexible responses to changes in the local environment, while maintaining a formal process for decision making.
-1 .9- HUMAN LIMITATIONS LIMIT THE EFFECTIVENESS OF A PDSS
A flexible DSS is one that can be tailored both to the individual manager and the pressures of the environment in which he operates. But the practitioner/decision maker is at best suspicious of the DSS. In the past, the primary purpose of the corporate model was an accounting model allowing corporate headquarters to monitor and control subsidiaries. The corporate model represented external authority rather than a decision support tool. Thus its unpopularity.
Another problem is that the data produced in the corporate model can also be misleading for the individual manager because the data is frequently aggregated. The aggregation of data can obscure many relationships that might be significant to the performance of the subsidiary (or other subsystem of the organization). Academic researchers deal with aggregated data and due to the high level of aggregation, the fluctuations are averaged out. The practitioner/decision maker must deal with wide fluctuations of data. It is little wonder that the practitioner is suspicious of academic approaches to the analysis of data. The practitioner, to be a successful decisionmaker, must be able to deal with random fluctuations in such a way as to screen out inappropriate signals. In an unstable environment this ability takes precedence over rigorous data analysis, if for no other reason than to select the appropriate data for analysis for making local decisions. The focus of the DSS should therefore be on support.
It becomes clear that the limitation of a decision support system is ultimately a human limitation. The limitation stems, as Alter notes, from the :Limit of the system designer's ability to communicate with the user, and the user's ability to understand and use the system designed. The limitation lies in the medium used to communicate. The user and designer are not always accessible to each other and so opportunities for learning from each other are :Limited. Most implementations have limited, well-defined periods during which contact is the most intense. But technological issues are raised throughout implementation.
The technological aspects of the system comprise the medium through which the two must communicate. The flexibility of the DSS should therefore include some natural language interface that is capable of producing computations and at the same time acting as a consultant to aid the manager who is making an analysis with which he is not entirely familiar. The changes made by the manager should be recorded in some way for study by the system designers for implementing further improvements in the system.
Case studies of DSS implementations indicate that the expertise in the design of the system is concentrated in a few personnel who frequently go on to other projects once the DSS is implemented. They report frustration at not being able to communicate with users on an ongoing basis, especially when the users are ready for more answers (e.g., [3, ch.7, pg.177]). The individualized DSS should therefore be capable of adapting to the user as the user becomes more sophisticated and should evolve with the user's concept of what the support system should provide. An expert system tailored to the position would insure that the some of the job specific knowledge of the user will remain with the position even if the incumbent is either transferred or promoted.
With Williamson's transaction cost approach, the specificity of job knowledge would decrease as expert systems were developed for each position. At the same time, by tying the individual DSS to the CSF of each position, the ability to measure the output of a position would increase. Thus, the transaction costs would decrease and the firm would become more competitive. Whether the increase was due to improved productivity or to lower labor costs allowed by being able to employ less skilled workers would depend on the position. This is the economic driving force behind the fifth generation computer technology. The speed with which it is adopted will be moderated by government regulation.