Before studying how IS/IT can support the managerial decision making process, we need a good understanding of the managerial decision making process itself. Below is the Harrison model of managerial decision making, developed by E. Frank Harrison. It illustrates the process nature of decision making.
The Harrison model is particularly suited to our purposes because it is an "information driven" process. As you can see from the illustration, the process consists of six phases and the process is iterative in nature. The process begins with management setting objectives. One of the primary functions of management is to set objectives which must be specific, measurable, and (realistically) attainable. Once set, the objectives form the basis for the decisions to be made. Managers next search for alternative methods of achieving the objectives. Once a number of alternatives have been identified, managers evaluate each of them. After the evaluation process is complete, managers have a relatively easy choice to make - the "best" alternative. This is the "Act of Choice" phase that most people think of when the topic is discussed. A major point here is that the "Act of Choice" cannot be validly completed without the three previous phases being completed. To do otherwise is to make a "SWAG". Once a choice has been made, the manager must then implement the decision. Failure to complete this phase renders all previous phases meaningless, an exercise in futility. Finally, the manager must monitor the implemented decision to verify that the decision is producing the desired results. This is the "Follow-up and Control" phase. Managers must ensure that they have an IS that will provide the information necessary to monitor the decision made. With the information managers can control the situation, making subsequent decisions as necessary. This phase of the decision making process "defines" the information requirements for managers.
Look at this (relatively simple) Cost-Volume-Profit example (also known as Break Even Analysis). A manager is considering the production of a product that sells for $6.00, costs $2.00 to produce, and will have a $35000 fixed startup cost. The marketing department tells the manager that at the projected selling cost the expected sales volume is 8000 units. Does the manager make the product or not? Cost-Volume-Profit analysis can assist in the decision. The accompanying Excel Spreadsheet illustrates the decision support. At the projected figures, the breakeven volume, 8750 units, exceeds the marketing department's sales projection, so the product should not be manufactured.
What if, however, the engineering department tells the manager that a simple redesign of the product can reduce the production cost to $1.50 per unit, and the fixed startup costs can be reduced to $20000. Under these conditions should the product be manufactured?
Here we can see that the breakeven volume is now 4444.4 units, which is below the anticipated sales of 8000 units, so the product will generate a profit and should be made.
From our Harrison decision model perspective, management sets an objective to produce a product that has the potential to generate a profit. Several products (alternatives) are proposed and various pricing and costing alternatives are examined. Finally, the "best" alternative is the one with the potential to produce the largest profit. This alternative is identified through "what if" analysis. The decision to produce the product is implemented and production begins. Now the IS must provide the production manager with production costs and volumes. Are costs actually $1.50 per unit? Do they remain at $1.50 for the entire production run? Were startup costs actually $20000? Did they increase during startup? Are sales as high as anticipated? What if the sale price is dropped to $5.50? What impact will this have on profit? The production manager must ensure that his/her IS can provide this information so the production process can be monitored and any corrections can be made as soon as possible. It is the ability to monitor and control that is facilitated through a well designed IS.
Now that we have gained a better understanding of the managerial decision making process, we can examine the support that IT/IS can provide.
In the E-Business Decision Support Trends section on page 350, we see that the Internet is expanding the information and decision support uses and expectations for everyone in a business. The trend is accelerating with the Internet and the E-commerce revolutions. Figures 9.2 an 9.3 illustrates this point.
Figure 9.4, page 352, illustrates the types of information required by managers when making decisions. Top level managers typically make strategic decisions, so their information requirements are much less specific and wider in scope than the other levels of management. Top managers consider data generated by the company as well as the information about current and potential competitors, population characteristics, economic conditions, and any other sources of information that may help them with the decision making process. These types of management decisions can be characterized as third order feedback loops (remember them from chapter 1?). These types of decisions are typically unstructured. The rules for making the decision cannot be written down. Managers use expertise, experience, consultants, and any other knowledge they can get to help with the decision making process. An example of this type of decision could be to enter into a new market, introduce a new product or service, or acquire a major competitor.
The tactical management level is characterized as (relatively) short term in nature. Monthly or quarterly decisions are typically made by these managers, and the standing reports from the MIS is the primary source of information. We can characterize most of these decisions as second order feedback loops. These types of decisions are semistructured. Managers have guidelines for making decisions, but they must also utilize expertise and experience to guide the decision making process. An example of this type of decision could be a decision to put a slow selling item on sale, call in salespeople who are not performing well, or increase the credit limit for a good customer.
Operational management is characterized by very short term decisions, usually daily or weekly. The TPS provides the data to the programs that make most of the decisions. Inventory management and customer credit limits are examples of programmed decisions that are first order feedback loops. These decisions are structured, and therefore programmable. Examples could be inventory quantity management and issuing of purchase orders.
Figure 9.5, page 353, provides additional examples of the three classifications of decisions.
The following table (from Davis & Olson, page 35-36) illustrates the seven characteristics of information. Figure 9.4 in the text provides insight into the levels of management and types of decisions managers make at the levels.
| Characteristic of Information | Operational Management | Tactical Management | Strategic Management |
|---|---|---|---|
| Source | Largely internal | both internal and external | External |
| Scope | Narrow, well defined | narrow to wide | Very wide |
| Level of Aggregation | Detailed | both detailed and summarized | Aggregated |
| Time Horizon | Historical | both historical and future | Future |
| Currency | Highly current | current to recent past | Quite Old |
| Required Accuracy | High, precise | relatively accurate | Low, imprecise |
| Frequency of Use | Very frequent | both frequent and infrequent | Infrequent |
Managers need to be familiar with the characteristics of information and recognize the differences in the characteristics over the various levels of management. Any IS can be made to satisfy any information requirement, but it the responsibility of each manager to specify his/her information requirements that are (generally) consistent with the management requirements. For example, a grocery store manager requires very current sales information in order to manage inventories, whereas the vice president for operations examines sales for the past five years in order to identify trends in customer purchasing habits. The store manager has no need for the trend projections since he/she doesn't have the authority to act upon them. Similarly, the vice president doesn't need specific store sales since he/she would be swamped with the volume of information from all stores in a large chain. The specification of information requirements becomes more important when we consider that the TPS captures all the data that is used to produce the information for all managers. It is the specific processing that determines the specific characteristics of the information.
Online Analytical Processing (OLAP) is a hot new area of IS/IT to support managers. As the author says, OLAP supports real time decision making. The book provides a brief introduction to OLAP. You can download my preliminary research into OLAP and gain some further insights into what it is, how it works, and how it supports the decision making process.
This DSS framework was developed by Sprague and Carlson, two pioneers in the DSS arena. You will most probably work with a DSS Builder in support of a top level manager, who is a DSS User. Microsoft Office is a classic example of a DSS Builder tool set. You are learning how to use the tools in your projects. Figure 9.10 provides a good comparison between an MIS and a DSS. They both (ultimately) do the same thing, but the DSS is more interactive in nature while the MIS focuses on scheduled reports as discussed on page 353.
The Data Visualization Systems, introduced on page 360, represent (IMHO) the most powerful DSS tool available to managers. Being able to quickly and easily visualize vast quantities of data makes decision making much easier. ALWAYS draw a graph instead of simply reporting column after column of data.
These lead to research to develop programs to enable computers to:
Understand natural languages
Solve problems by reasoning
These programs do not pretend to employ the same mechanisms as humans in carrying out such activities (neural networks notwithstanding). They do produce results which are (more or less) comparable to what we would expect humans to do.
Natural Language Systems: The objective of NLS is to allow humans to interact with computers in their own natural languages, as with the latest model of the famous "HAL 9000." Some example interactions may include:
One of the most difficult obstacles to overcome with NLS is the requirement for "context." For example, the question, "Who are in Florida?" can have many meanings, all depending upon who is asking the question and about whom the question is being answered. For example, is the person asking the question referring to all salespeople who work for the company, or all the citizens of Florida? This question must first be answered before the actual question can be answered. Ambiguities can be resolved only through the development of context. The NLS program must have the ability to ask questions that define the context, or the context must be programmed into the NLS. Either endeavor is quite difficult and time consuming. Notice that once the context necessary to understand and answer the first question has been established, the next three questions are easy to answer. One of the greatest problems with AI is getting past the first question which defines the scope of the question.
Relational DBMS have query languages, such as SQL. These serve to perform the functions that NLS are intended to perform. It is far easier for people to learn the syntax of a query language, and use it to define context, that it is for a NLS to ask questions necessary to define context. (This is part of why you have the Access RDBMS assignment)
Expert Systems, often referred to as "Reasoning Systems," are introduced on page 381. The section discusses the benefits and limitations of XS. Figure 9.29 illustrates the processes and components of an XS. The following discussion provides another perspective on XS.
Expert Systems (XS) have three major components:
The reasoning system's objective is to examine relevant facts and assertions in a sequence that results in the derivation or discovery of a solution to a problem. Reasoning strategies generally fall into two categories:
Expert systems focus upon:
Elements of an Expert System: XS have three primary components:
When to Consider an XS:
Managers should look for problems of a reoccurring nature that require human experts who are in relatively short supply, overburdened, unavailable, or very expensive. Specifically, these four conditions must exist before an XS should be considered:
An XS opportunity may exist if the problem solving process involves reasoning for the following purposes:
Phases of XS Development
Problems with XS: The greatest problem with XS is that keeping them up to date has proven to be quite difficult and costly. Technological advances in all areas quickly obsolete knowledge captured in an XS. XS must be continually updated as the experts themselves update their knowledge and expertise. For example, an XS for configuring microcomputers for businesses from just five years ago is totally obsolete due to new, faster, cheaper microcomputer technology. XS work well in areas where the environment is relatively static, such as with the Statistical Advisor demonstrated in class. New and exciting statistical procedures don't come along as often as new computer technology.
How an XS Works: Below is an illustration of how an XS works. The system tries to replicate the conversation that a person may have with an expert in a specific area. The XS in this example is a "Statistical Analysis" expert that recommends specific statistical analyses to use to analyze data based on the responses to questions from the user. The ultimate result is a specific recommendation that the user can follow in order to properly analyze his/her data. The process begins with the XS "asking" the user what general type of analysis is to be performed. The user indicates the answer (is this specific XS) by moving the arrow to highlight the correct answer.
Let's assume that we select "Comparison of Two or More Populations." The XS next asks about the nature of the data - on what measurement scale was the data quantified?
Generally, all data quantified and measured on the nominal and ordinal scales can be validly analyzed using a non-parametric procedure. Data quantified and measured on the interval and ratio scales can be validly analyzed using a parametric procedure. The XS will determine which "family" of statistical analysis procedures is appropriate based on the answer to this question. This is an example of "backward" reasoning. The XS "backs" into an answer by continuing to narrow down the possible correct answers.
Let's assume that we select "Ratio" scale, the scale with the most information available for analysis. The XS next asks about the nature of the data collection procedure.
Let's assume that we select "Matched Pairs" as the data collection method. This is a popular method generally characterized as a "before and after" analysis. Data before a treatment (such as a training program) are matched with data after the treatment for the same person. The XS, based on the responses to the questions it asks, reaches a conclusion and makes a specific recommendation.
As you can see, the XS recommends that a "Single or multifactor ANOVA with repeated measures on the same element" be used. The user can now go to any statistics book, look up this analysis procedure, and conduct his/her research and valid analysis.
There are two important points to understand here. First, then XS makes a recommendation. The validity of the recommendation depends upon two factors: how well the expert's knowledge was captured by the XS, and how accurately the user answered the questions. This brings us to the second important point. The XS is no better than the user's ability to correctly answer the questions. For example, if the user doesn't know (or remember from his/her statistics class) what "Matched Pairs" or "Independent Samples" or "Randomized Blocks" means, he/she cannot correctly use the XS. An expert person could, at this point, provide an explanation of these sampling methods and help the user decide on one. This can, of course, be included in an XS, but the size of the XS explodes when these types of considerations are included. Providing background knowledge has proven to be very difficult when developing XSs.
One valuable aspect of an XS is the ability to ask it "WHY" it arrived at its conclusion. Below is illustrated the "chain of reasoning" for our example.
Notice that the first assumption is that a non-parametric procedure is to be used. The Knowledge Engineer, the person who is "capturing" the expert's knowledge, made this assumption as a starting point. Since there are analysis procedures for comparing two or more populations in both non-parametric and parametric analyses families, the answer to the first question did not contradict the initial assumption. However, when the second answer indicated "ratio" data, the XS shifted to a parametric procedure because this is the most appropriate family of procedures for ratio data. The "matched pairs" answer, combined with the ratio data and two or more populations caused the XS to recommend ANOVA. This is exactly (or close to it) the same "chain of reasoning" that an expert person would use to arrive at a conclusion.