6 chapter's summary

Chapter 5: Advanced Information Systems

1 of 31

ACCOUNTING INFORMATION SYSTEMS: A DATABASE APPROACH by: Uday S. Murthy, Ph.D., ACA and S. Michael Groomer, Ph.D., CPA, CISA

Advanced Information Systems

Learning Objectives

After studying this chapter you should be able to:

• describe and discuss the hierarchy of decision making that occurs in transaction processing systems, decision support systems and executive information systems

• describe the input, processing and output characteristics of a decision support system

• describe the various components of decision support systems

• describe the benefits and drawbacks of a decision support system

• discuss concepts of the data warehouse, data mart, and data mining, which are methods of garnering intelligence from the volumes of data in transaction processing systems

• explain the characteristics of expert systems

• differentiate between transaction processing systems, decision support systems, and expert systems

• explain the various components of expert systems

• describe the knowledge representation methods used in developing expert systems

• discuss the process of developing expert systems with specific reference to the knowledge acquisition and system validation steps

• explain the benefits and drawbacks of expert systems

• discuss the criteria for identifying applications that might benefit from expert systems technology

• discuss the accountant's role in expert systems

Having discussed the technology of information systems in the previous two chapters, we shall now examine the different categories of information systems, from routine

Chapter 5: Advanced Information Systems

2 of 31

transaction processing systems to advanced decision support systems and expert systems. The database approach to accounting systems, which will be presented in detail in future chapters, can provide the foundation for more advanced decision support systems. The purpose of this chapter is to contrast the relatively straightforward transaction and event recording role of transaction processing systems with the more strategic role of advanced decision support systems. Transaction processing systems record all transactions and events, while decision support systems are aimed at garnering business intelligence from the volumes of data recorded by transaction processing systems. The chapter also discusses the components, characteristics, advantages, and disadvantages of decision support systems and expert systems.

Hierarchy of decisions made in organizations Recall from our discussion in chapter 1 that the purpose of accounting is to provide information for decision making. Information systems are designed to convert data into information. The kinds of decisions that need to be made, and correspondingly the complexity of the information system, can vary considerably depending on the decision making level in the organization. The following figure on the next page describes the three levels of organizational decision making along with the types of systems used at each level.

As the figure suggests, transaction processing systems are used at the lowest level in the organization, expert systems at the highest level, with decision support systems falling in the middle. Let us now discuss the types of decisions made at each level in an organization and the type of information system that ought to be developed to provide users with relevant and timely information for decision making.

Chapter 5: Advanced Information Systems

3 of 31

Transaction processing systems At the lowest level, decisions are straightforward and involve relatively simple data processing needs. How to account for a credit sale? How sales returns are to be handled? What actions need to be taken to process a back order? These are some examples of decisions that need to be made at the lowest level in an organization. Such decisions are routine and repetitive. They are also highly structured in that well defined rules of data processing exist. For example, accounting for a credit sale is a simple process of increasing the balance in the sales account and the customer's account receivable. The manner in which such transactions are to be accounted for does not call for judgment. Rather, it is simply a matter of applying the appropriate processing rule. The data processing requirements for such systems can very easily be elicited by systems developers and are also easily articulated by users. Information systems for such applications are easy to develop because well defined processing rules are easily programmed using languages such as COBOL, C or Visual Basic. Such information systems, for processing routine and repetitive tasks, are classified as transaction processing systems (TPS). Sales order processing, payroll, and purchase order processing systems are examples of TPS. Recall from our discussion in chapter 2 that all information systems have three essential elements: inputs, processing, and outputs. The inputs to TPS are transaction data from routine, recurring transactions. These transactions occur internally and also as exchange transactions with external parties such as customers, vendors, and other institutions such as banks. Data input into and processed by TPS are classified as internal data. External data that does not relate to routine transactions is typically not processed by or stored in TPS. The processing of these transaction inputs involves the application of well defined rules with no ambiguity. Outputs of TPS are routine reports conveying the results of transaction activity for a certain period. The formats of such reports are standard, and there is typically little need for extensive customization of TPS reports. TPS are therefore designed for recording and processing routine transactions and generating standardized reports on a periodic basis.

The actual form of TPS can vary from older COBOL-oriented batch processing systems to on-line real-time systems such as an airline reservation system. Recall from our discussion in chapter 1 that organizations are moving towards enterprise-wide database systems. In particular, enterprise resource planning (ERP) systems such as SAP R/3, PeopleSoft, and BaaN have become very popular in large organizations. Even in such ERP systems, routine transaction processing must be performed. However, the raw data and processed transactions all feed into the database which can be accessed by users at all levels in the organization. Older COBOL TPS were stand-alone systems, and the data generated in those systems were not easily accessible by other systems in the organization. The differences between the older COBOL-based TPS and the newer database oriented systems will be explored further in the next chapter.

Decision Support Systems The next level of decision making involves situations that call for some degree of human judgment. Decisions that do not have well defined processing rules and therefore benefit from managerial intuition and experience are categorized as semi-structured

Chapter 5: Advanced Information Systems

4 of 31

decisions. Budget setting, cash flow forecasting, inventory control, and managing an investment portfolio are examples of semi-structured decisions. These decisions frequently require decision makers to consider a large number of data inputs, some of which may be imprecise. In fact, the ability to deal with multiple inexact criteria is what distinguishes expert managers from novice decision makers. Unlike the environment of highly structured decision making situations, semi-structured decisions frequently require data inputs unrelated to routine transactions from sources external to the organization. For example, a manager responsible for allocating funds in an investment portfolio would want to keep track of interest rates, bond yields, and stock prices. It is highly unlikely that such data would be routinely captured within any of the organizations' existing transaction processing systems. Thus, systems to support semi- structured decisions must be capable of capturing or accessing data from a variety of external sources.

In the context of the above discussion on the difference between structured and semi- structured decisions, several characteristics of decision support systems are evident. First, decision support systems consider data from both internal and external sources. Second, the rules for processing these data are complex and not as well defined as in the case of structured decisions for processing routine transactions. Third, managerial experience and intuition are frequently brought to bear in making semi-structured decisions. Given seemingly similar sets of data, managers may arrive at different conclusions based on their prior experiences in that decision making environment. Fourth, managers may not be able to precisely articulate their methods of processing the inputs and arriving at conclusions. In fact, based on their personal styles and preferences, different managers may require slightly different systems to assist them in very similar semi-structured decision making situations. Furthermore, managers' needs and preferences may change over time, and perhaps from one situation to the next. Therefore, flexibility is a key design criterion in developing systems to support decision making for semi-structured problems.

Information systems support for semi-structured decision making situations is provided in the form of decision support systems (DSS). A definition of DSS that encompasses its characteristics in terms of inputs, processing, and outputs follows:

A DSS is a flexible user-friendly information system that incorporates data from internal and external sources, applies models and complex processing rules to the data, and produces information in a variety of formats for the purpose of decision making in semi-structured situations.

The DSS is user-friendly in that a variety of data should be easily accessible and should be displayed in an attractive manner. The use of multicolor graphs and well formatted tables is common in DSS. The graphical user interfaces of today's operating systems and applications software lend themselves to the development of DSS. The flexibility aspect of DSS allows it to be customized to the needs of different users. It also allows the same user to alter the look and feel of the system to better fit his/her needs. Let us consider some accounting examples of DSS. A sales and marketing DSS might present the sales manager with up-to-date information about sales, together with information about how the company is performing relative to the competition. The DSS might also

Chapter 5: Advanced Information Systems

5 of 31

provide forecasts on future sales performance to aid in planning. Such a DSS would incorporate sales data, competitor market share data, and data about economic trends. Another accounting oriented DSS might aid managers in planning inventory levels at the company's warehouses nationwide. The system could provide information about past inventory levels and demand in different regions as a basis for determining the appropriate inventory level for each regional warehouse. A variation of DSS is the executive information system (EIS) or executive support system (ESS). As their names suggest, EIS and ESS are designed to provide top executives with information relevant to their roles as high level managers. The information is provided in a very user-friendly manner. GUIs are almost always employed in designing EIS, with voice input and touch screen input systems also commonly used. The key assumption made in developing EIS is that the users (top management executives) have limited computer skills. These same executives, however, are responsible for tracking performance and making critical decisions to solve major problems as they arise. Thus, the challenge for designers of EIS is to make the system user-friendly enough for top executives to comfortably use and yet so information rich that they provide relevant and timely information. An EIS at a major airline might provide the CEO with up to the minute information about on-time arrivals at the company's major hubs.

Input, processing, and output of DSS In contrast to TPS, note that DSS utilizes both internal data and external data that does not pertain to transaction processing. DSS can be designed to interface with publicly available databases such as the Lexis/Nexis legal database or the national automated accounting retrieval service (NAARS). The Internet, and specifically the Web, is becoming another source of external data that could possibly be used in DSS. The processing of these data is performed by applying relatively complex rules that may need to be refined and modified over time. These processing rules may be difficult to identify and program because managers may not be able to easily specify the inputs required, the type of processing that needs to be performed, and the exact nature of reports that should be generated. In addition to these rules, DSS frequently use a variety of models such as statistical, financial, operational, and strategic organization- specific models. For example, statistical models such as the regression model and analysis of variance (ANOVA) are often used in DSS. In an inventory control DSS, the economic order quantity (EOQ) model might be programmed into the DSS. As an example of an organization-specific strategic model, the "Big Three" automobile manufacturers have probably developed sophisticated models that predict consumer demand for different types of vehicles for both short term and long term planning horizons. In the DSS definition, note that DSS produce information (reports) in a variety of formats. By contrast, TPS typically generate only standard reports adopting a "one size fits all" approach. In addition to standard and custom reports, DSS also allow the manager to pose "what if" questions, to perform goal-seeking analysis, or to run regressions or simulations on the data. Current spreadsheet software, for example, can

Chapter 5: Advanced Information Systems

6 of 31

be set up to facilitate "what if" and goal seeking analysis. In "what if" analysis, as the term suggests, the manager can determine the effects of changing certain variables on other variables or outputs. For example, the manager can determine the effect on net income after tax if the sales price per unit is increased by 10%. In goal seeking analysis, the manager sets a target for net income after tax and then examines how other variables such as the price per unit or cost of production would have to change to arrive at the desired net income figure. Simulation and regression analysis are useful for forecasting purposes, perhaps in a budget setting situation. Such analyses employ past data and apply a number of assumptions in an attempt to predict the values of variables such as cash flow or net income.

DSS by no means provide definitive and final solutions; there is considerable room for managers to apply their intuition and experience. However, DSS often suggest alternative courses of action perhaps with a rank ordering of the alternatives in terms of the degree of suitability to the problem at hand. By highlighting certain aspects of the output, and by prompting the manager for input, DSS can be designed to force the manager to consider relevant information.

Components of DSS

There are three distinct components of DSS: a data component, a processing component, and the user interface. The data component comprises a database of internal and external data. The processing component contains the rules and models that can be applied to the data as needed. As indicated earlier, a variety of financial, statistical, operational, and strategic models would be stored in the processing component of the DSS. Some of these models may be standardized models such as regression or EOQ models, whereas certain operational or strategic models might be unique in-house models tailored to the needs of the organization. Finally, the user- interface component allows the manager to generate customized reports, to perform "what if" or goal seeking analyses, or to perform tasks such as running simulations or regressions in an attempt to obtain answers to questions or problems that the manager might be faced with. The components of a DSS are shown in the following figure on the next page.

Chapter 5: Advanced Information Systems

7 of 31

Given the input, processing, and output components of DSS outlined above, it should be clear that spreadsheet and database software packages are well suited for building powerful DSS. The GUI environments of today's spreadsheet software such as Microsoft Excel lend themselves to the development of user-friendly DSS applications. Spreadsheet software in particular is the logical choice for creating "what-if" oriented and goal seeking DSS. Visual Basic macros, dialog editors, and a variety of add-in tools available in Excel are ideally suited for developing customized DSS. In addition to spreadsheet and database software, fourth generation programming tools such as SQL, FOCUS, and current flavors of Visual Basic can also be used. Object-oriented programming languages could also be used to develop DSS. The advantage of the object-oriented approach is that objects (or modules) developed for one DSS can easily be reused in other DSS.

A variation of a DSS is a digital dashboard. Like the DSS discussed above, digital dashboards are also aimed at providing decision support. As the term implies, a digital dashboard is a screen (usually a single screen) that shows key performance indicators (KPI) for an organization for the benefit of top management. Key features of a digital dashboard include (a) display of key mission-critical metrics, (b) data from multiple sources, (c) real-time information updates, (d) visually appealing with the use of graphs, charts, and colors, and (e) highlight areas in need of managerial attention. A sample digital dashboard for a restaurant chain is shown below.

Chapter 5: Advanced Information Systems

8 of 31

Benefits and drawbacks of DSS The benefits of DSS are difficult to quantify. In general, DSS are expected to result in improved decision making performance. However, "improved" performance can be defined in a number of ways. Faster decisions, fewer errors, and greater accuracy (when measurable) might be some attributes of decision making performance that can be interpreted as improvement. However, accuracy in decisions may be difficult to measure in certain situations. Further, increased accuracy may be attributable to causes other than a newly instituted DSS, especially in complex and dynamic environments. In addition to accuracy, other measures of improvement in decision making performance include consistency and consensus in decision making. Using a DSS, if a manager always makes the same or very similar decisions when confronted with similar inputs, then the use of the DSS contributes to increased consistency in decision making. From an organization's standpoint, consistency is desirable. If a manager makes drastically different decisions when confronted with essentially similar circumstances, the consequences of these differing decisions may prove harmful to the organization. Consensus in decision making is relevant when there are multiple decision makers who all make the same decision. For example, a company may have a number of recruiting managers who evaluate applicants for open positions in the company. Recruiting managers use a number of criteria to determine whether to make a job offer to an applicant, based on an interview with the applicant and information supplied by the applicant. If a DSS is developed to assist these recruiting managers, one measure of its success is if different recruiters arrive at the same decision (make an offer to or reject the applicant) when they use the DSS to assist them in their decision. Such consensus

Chapter 5: Advanced Information Systems

9 of 31

resulting from the use of a DSS might protect the company from law suits brought by applicants who feel they were wrongfully denied a job. In auditing, consensus is a much sought after attribute because concrete outcome measures are frequently not available. For example, one judgment auditors must make is the level of risk associated with a particular audit. Briefly, audit risk relates to the probability that the financial statements are materially misstated. The "true" level of audit risk remains an unknown so long as the client company does not go bankrupt. Thus, if different expert auditors all agree that the level of audit risk is "medium," then the risk level is determined to be medium in the absence of any independent "hard" measure of risk. It should be noted that the perceived amount of audit risk determines the amount of audit work that is performed and is therefore a very critical estimation from the viewpoint of the auditing firm.

The drawbacks of DSS hinge on the semi-structured nature of the decisions they are intended to support. To the extent that the system developer does not fully understand the variables needed and processing required, the DSS will not fully meet the needs and expectations of users. DSS should be thoroughly tested before they can be relied upon since the consequences of reliance on a faulty DSS can be disastrous. Finally, there may be a tendency for managers to blindly follow the DSS and not apply their expertise in identifying variables and approaches to solve problems in a manner superior to that programmed into the DSS. The benefits and drawbacks of DSS are summarized in the following table.

Benefits and Drawbacks of DSS

Benefits of DSS Drawbacks of DSS

Can improve decision making performance resulting in more accurate decisions.

Semi-structured nature of problem environment may result in poor fit between DSS capabilities and user's needs.

Improves consistency and consensus in judgmental decision making (where criteria for "right" or "wrong" answers may not be available).

Consequences of errors in models or processing rules can be disastrous.

Forces managers to consider relevant information.

Managers may focus too heavily on DSS and may fail to apply their expertise.

As noted earlier, transaction processing systems record data about transactions and events. For large organizations, several gigabytes of data may be recorded in a week or even within a day! How can an organization make sense out of large volumes of data? How can one identify the "gems" of information from within gigabytes of data? Organizations must be able to "mine" or "model" current and historical data via decision support systems to obtain the necessary business intelligence to compete effectively.

Chapter 5: Advanced Information Systems

10 of 31

Data warehousing, data marts, and data mining are three concepts related to the issue of modeling and of providing a robust set of data elements which could support DSS.

A data warehouse is a repository of historical business transaction data, organized in a manner to facilitate efficient querying of data for reaching marketing, tactical, and strategic decisions. The key point is that the data warehouse is separate from the organization's "live" database that captures and stores current transaction data. Why is the data warehouse separate from the live database? A data warehouse can be justified on several grounds, as follows: (1) the live database is missing old data -- the operational database system may be designed to automatically rid itself of old data, which may be useful from a strategic and marketing viewpoint, (2) data in the live system is constantly being altered, and it is necessary to keep track of the historical state of certain data items, (3) it is difficult to find information in the databases -- the data may be arranged in a manner that is most efficient for day-to-day operations but that arrangement may not facilitate analysis of the data to reach the intended strategic, tactical, and marketing decisions, (4) queries to answer strategic, tactical, and marketing questions are a burden to the operational system, and (5) the data needed for these decisions come from different systems -- an enterprise whose subsidiaries use different computer systems may find it difficult to gather data from all of the subsidiaries in order to see aggregate and individual statistics.

A data mart is a closely related concept. It can be defined as a repository of data gathered from operational data and other sources that is designed to serve certain users' needs. The data in a data mart may actually come from a data warehouse, or it may be more specialized in nature. The emphasis of a data mart is on meeting the specific demands of a particular group of users in terms of analysis, content, presentation, and ease-of-use. The terms data mart and data warehouse often imply the presence of the other in some form. However, there seems to be consensus that the design of a data mart tends to start from an analysis of user needs and that a data warehouse tends to start from an analysis of what data already exists and how it can be collected in such a way that the data can later be used. A data warehouse is a central aggregation of data (which can be distributed physically); a data mart is a data repository that may derive from a data warehouse or not and that emphasizes ease of access and usability for a particular purpose. In general, a data warehouse tends to be a strategic but somewhat unfinished concept; a data mart tends to be tactical and aimed at meeting an immediate need.

Data mining can be defined as the analysis of data for relationships that have not previously been discovered. For example, data mining of sales records for a large grocery store may reveal that consumers that buy diapers also tend to buy ice-cream and apple juice. Data mining, also referred to as knowledge discovery, looks for associations (correlated events -- beer purchasers also buy peanuts), sequences (one event leading to series of related events -- someone buying tile for a new home later buys other home products), classification of data (generating profiles of consumers showing their interests), clustering of data (finding and visualizing groups of related facts), and forecasting (making predictions about future events based on an analysis of patterns of existing events). Companies collect vast quantities of data about their

Chapter 5: Advanced Information Systems

11 of 31

customers and their preferences through the sales order transactions that are recorded. For example, Walmart processes upwards of 20 million point-of-sale transactions every day! Apart from simply recording these transactions, companies like Walmart have realized that data mining techniques can be employed to learn more about their customers, which products sell better under what conditions, and how to make their operations more streamlined. Incidentally, Walmart has about 460 terabytes of data stored in its data warehouse.

A host of tools are available to bring these notions of data warehouses, data marts, and data mining into reality. You are encouraged to search the Web for further information about these concepts. A very useful web site that provides answers to "what is" questions like "what is a data mart" is whatis.com -- check it out!

Expert Systems Since the early days of computing there have been ongoing efforts, both successful and unsuccessful, to make computers perform tasks performed by human beings. The branch of knowledge investigating the ability of computer-based systems to simulate human reasoning is called artificial intelligence (AI). The most successful application of computer technology has been in automating mundane, routine, and repetitive tasks. Some of these tasks were relatively simple, but others were rather involved requiring humans to expend considerable time and effort to master. For example, as an accounting student you undoubtedly spent at least a year or two to learn various accounting rules and procedures. However, the vast majority of these rules can very easily be programmed into a computer system! In fact, many off-the-shelf software packages exist that can perform most accounting tasks even for relatively large businesses. Herbert Simon, one of the pioneers in the field of AI, has very astutely observed that the abilities we human beings acquire naturally in the first two or three years of life are those that are hardest to program into a computer system. Speech, recognizing different shapes and sounds, distinguishing between objects, and the sense of touch and smell are some of the capabilities we acquire in our infancy. However, despite considerable research and expense, we still do not have a computer system that can faultlessly recognize speech. In the field of robotics, which is one of the primary areas of AI research, considerable progress has been made in developing machines that can recognize different shapes and maneuver around objects.

AI research has resulted in the development of (1) intelligent robots, (2) voice and pattern recognition systems, (3) natural language query systems, and (4) expert systems. Robots, voice, and pattern recognition systems have already been briefly discussed above. Natural language query systems aim to allow the user to type in queries in plain English, without having to conform to a strict syntax, and still provide accurate and reliable answers. For example, one manager might type "List all customers with balances above $10,000" while another might type "Display those customers who have outstanding balances greater than 10000." These two queries are identical and should yield the same answer from the computer system. A natural language query system is programmed to understand synonyms and to interpret query words according to the context in which they are used. Thus, the system would

Chapter 5: Advanced Information Systems

12 of 31

recognize that "display" and "list" really mean the same thing, and that "above" and "greater than" are synonymous when the target is a numeric field - the customer's current balance. The system would also recognize that "balance" and "outstanding balance" both refer to the "current balance" in the customer file. The “Siri” system on the latest Apple iPhone is an example of expert systems technology at work. You can simply push the button on your iPhone and ask Siri virtually any question, including “What is the meaning of life?” (The answer Siri gives can vary, from “I don’t know, but I think there’s an app for that,” to “42”).

The last branch of AI research is expert systems (ES). Because this field has a number of accounting applications, we will discuss ES in greater detail. First, we present a general overview of ES technology after which we will discuss specific accounting ES that have been developed. The most basic definition of an expert system is that it is a computer-based system designed to imitate the reasoning process of a human expert. A more formal definition follows:

An expert system is a knowledge-based system that can consult with the user and applies heuristic reasoning to arrive at justifiable solutions to complex problems even with incomplete information.

Naturally, one goal of ES research is to develop systems that could replace human experts for certain types of decisions. Ideally, novices should be capable of using an ES to solve advanced problems which can ordinarily be solved only by experts in the area. However, ES need not (and perhaps should not) replace human experts; they could be used in a consultative role to assist rather than replace human experts. Expert systems are also referred to as knowledge-based systems (KBS) because they incorporate expert knowledge in some form.

Let us explore the components of the formal definition given above. An ES must be knowledge-based. The knowledge that is programmed into an expert system is usually acquired from one or more human experts. However, it is not essential that human experts are used. If the knowledge that is the subject matter of the system is codified in authoritative sources such as textbooks and specialized industry guides, then human experts need not be consulted. In developing an expert system to determine whether a business combination is a purchase or a pooling of interest, the developer can turn to the appropriate Statement on Financial Accounting Standard (SFAS) issued by the Financial Accounting Standards Board (FASB). In addition, guidelines published by the AICPA and in accounting journals might also provide useful information for developing the system.

The second unique characteristic of an ES, as indicated in the definition, is that it can "consult" with the user. ES are typically designed to engage in intelligent dialog with the user. Consultation refers to the ability of the system to explain why a question is being asked. Thus, the user can ask the system to explain the logic that underlies a question that he/she is asked. Intelligent dialog also means that the system asks only relevant and appropriate questions. One of the first questions a medical diagnosis expert system would ask is whether the patient is male or female. If the patient is male, then the system should not ask any questions that would be relevant only for a female patient

Chapter 5: Advanced Information Systems

13 of 31

(e.g., a pregnancy related question). This intelligent dialog ability is also evident in the so called "smart" questionnaires that do not ask every user all questions -- the system branches to different questions based upon answers already given to earlier questions.

The next component of the definition of an ES refers to heuristic reasoning. A heuristic is a rule of thumb. Applying judgment usually acquired over many years, experts approach problems very differently from novices. They apply decision rules that may not be obvious to novices in the area. These expert decision rules, which distinguish experts from non-experts, are called heuristics. Heuristic reasoning is also the ability to deal with inputs such as "controls are weak" or "risk is high." These types of inputs are non-numeric and are typical for application areas in Auditing. A key task in designing and developing expert systems is the codification of heuristics, or expert decision rules, into the system.

Regarding the outputs of an ES, the definition indicates that the solution provided is justifiable. In other words, the ES should be capable of explaining exactly how and why it arrived at the recommended course of action. The system essentially traces through the interaction with the user, detailing every question asked of the user and every answer provided along with an indication of the specific rules in the knowledge base that were applied. Such a trace enables the user to understand the process whereby the system arrived at the final answer.

Returning to the definition of ES, note that only complex problems are appropriate for ES development. Structured problems, and even most semi-structured problems, are not appropriate for the application of ES technology simply because they can be addressed by conventional computer programs. In terms of the hierarchy of decision making discussed earlier in the chapter, expert systems are meant to tackle unstructured, complex decisions. If the rules for solving the problem are well defined with no ambiguity, then a conventional computer program is sufficient -- ES technology need not be applied. Decision making tasks that are performed efficiently and with relative accuracy by human experts only after years of experience are tasks that are appropriate for ES.

The last component of the definition suggests that ES can arrive at solutions even when confronted with incomplete information. An ES can be programmed to apply a "certainty factor" (CNF) to a recommendation. If all needed inputs are provided by the user, then the ES applies a relatively high CNF to the suggested course of action. If certain inputs are not provided, the system attempts to arrive at a solution but applies a lower CNF. In essence, the CNF level indicates the extent of confidence the user can place in the system's recommendations.

The differences between ES, DSS, and TPS should be apparent in the above discussion. Whereas TPS are intended for routine structured decisions at the lowest level of the organization, ES are intended for unstructured decisions at the highest level within the organization. DSS fall between TPS and ES and are intended primarily for semi-structured decisions made by middle-level managers. TPS are designed to handle the input, processing, and reporting aspects of an organization's transaction processing. DSS use transaction data but also incorporate data from external sources such as

Chapter 5: Advanced Information Systems

14 of 31

industry databases. In contrast with TPS and DSS, ES are the only form of computer- based decision support capable of heuristic reasoning. Consultation, the ability to explain why a question is being asked, and justification, the ability to explain how a conclusion was reached are also two features unique to ES. Before we move on to a more detailed discussion of expert systems, consider the following table, which summarizes the differences between TPS, DSS, and ES.

Comparison Between TPS, DSS, ES

Feature TPS DSS ES Degree of structure in decisions addressed by system.

High Medium Low

Level within the organization where tasks are performed.

Low-level employees -- typically clerical accounting personnel

Middle-level management

Top-level management

Data which are the subject of the system

Internal only Internal and external Internal and external

Type of decisions Highly structured, fully programmable

Semi-structured, partially programmable

Unstructured, involves heuristic reasoning

User interaction

Prompting for input; displaying results of transaction processing

Prompting for input; guiding manager through a process; suggesting alternative courses of action

Consulting with user; justifying a conclusion

Processing Simplistic rules Complex rules and models Expert rules

Components of Expert Systems There are four primary components of an ES: the user interface, the domain database, the knowledge base, and the inference engine. The user interface in an ES is quite similar to that in a DSS with an emphasis on ease of use. Most ES rely on interaction with the user. The system prompts the user for input and perhaps presents a limited

Chapter 5: Advanced Information Systems

15 of 31

number of choices. As each question is being asked, the user has the ability to ask why the question is relevant (i.e., the "consultation" feature of ES). Today's GUI operating systems permit the development of very user friendly interfaces with pull down menus and toolbars. The second component of ES is the domain database. The domain database is designed to store the factual knowledge relevant to the decision making situation. Consider an ES for determining whether or not to grant credit to a loan applicant. The domain database would contain information about the number of monthly payments the applicant has missed, the number of late payments, the applicant's salary, years on the job, other debts, etc. Obviously, the specific values stored in the domain database would vary from one decision making situation to the next. For each new situation, the domain database variables would be reset and populated with the facts unique to that situation. Experts familiar with the decision making situation, for example bank loan officers for the credit granting decision, should be consulted to determine the type of factual knowledge to be stored in the domain database. The contents of the domain database are also referred to as declarative knowledge.

The next component of expert systems is the knowledge base. Whereas the domain database contains the facts relevant to a particular decision making situation, the knowledge base contains the rules that determine how the facts are to be manipulated to arrive at a decision or conclusion. Such knowledge is referred to as procedural knowledge. In the credit granting decision described above, consider the following rule. If the applicant has missed more than three monthly payments in the past and if the applicant has been on the job for less than two years, then the application should be denied. This rule is an example of the contents of the knowledge base in an ES. There are a number of methods of representing knowledge, and these will be discussed in more detail a little later. The knowledge base and the domain database interact with one another - the knowledge base acts on the parameters unique to a particular situation as defined in the domain database. The difference between the two is that the domain database is dynamic - the values contained therein constantly change from one decision making situation to the next. The knowledge base on the other hand is more stable or static - knowledge represented by means of rules or other methods stays relatively constant over time. In other words, the way in which decisions are made in a particular decision making environment (e.g., credit granting) does not change constantly, but the variables that are dealt with do change from situation to situation. Given that the domain database constantly changes, it is best created using a database management system. The knowledge base, on the other hand, is typically created using tools provided within ES software.

The final and most complex component of ES is the inference engine. The inference engine is the logic component of the ES. It is also called the rule interpreter because it is this software that processes the rules stored in the knowledge base. The process of invoking rules is referred to as "firing" of rules. In a rule-based expert system, there are two methods by which the inference engine fires rules - forward chaining and backward chaining. As the term suggests, forward chaining involves working with available data to try to draw conclusions. It is a data driven approach. By contrast, backward chaining is a goal seeking approach. The system begins with a particular

Chapter 5: Advanced Information Systems

16 of 31

hypothesis (conclusion) and then attempts to find data to either support or refute that hypothesis. To recap, the components of an ES are the user interface, the domain database, the knowledge base, and the inference engine, as shown in the following figure.

Knowledge representation methods

As alluded to above, one method of representing knowledge in an ES is using rules. Rules have an IF-THEN structure, with the "IF" component referring to conditions or premises and the "THEN" component referring to conclusions to be drawn if the conditions are met. In a rule-based expert system the knowledge base would be populated with a series of related rules, possibly running into the hundreds or even thousands for complex systems! Returning to the credit granting application discussed earlier, a few rules for a credit granting ES follow:

Sample Rules for Credit Granting ES

RULE1 IF missed-payments > 3 AND late-payments > 2 THEN credit-history = POOR;

RULE2 IF missed-payments >=1 AND missed-payments < 3 AND late-payments <= 2 THEN credit-history =

Chapter 5: Advanced Information Systems

17 of 31

FAIR;

RULE3 IF missed-payments = 0 AND late-payments =< 2 THEN credit-history = GOOD;

RULE4 IF missed-payments = 0 AND late-payments = 0 THEN credit-history = VERY-GOOD;

RULE5 IF credit-history = POOR and years-on-job < 2 THEN decision = DENY-CREDIT;

RULE6 IF credit-history = GOOD and years-on-job > 2 THEN decision = GRANT-CREDIT;

RULE7 IF credit-history = VERY-GOOD and years-on-job =< 2 THEN decision = GRANT-CREDIT;

The above list of rules is certainly not exhaustive; many more rules would have to be created for a functional ES for the credit granting decision. Certain variables derive their values from other rules, while the remaining variables require user input to obtain values. The "credit-history" variable derives its values from rules 1 to 4. The variables in the "IF" part of rules 1 to 4, "missed-payments" and "late-payments," rely on user input for their values, as does the "years-on-job" variable in rules 5 through 7.

Many relatively unstructured decisions in accounting and business can be described as rule-based. In fact, consciously or unconsciously, almost all business decisions involve the application of rules. Many decisions in auditing, such as whether the client can continue as a going concern, or determining the level of audit risk, are rule-based. Thus, many of these decisions are suited for the application of ES technology. What distinguishes these decision rules from the structured transaction processing rules for TPS is that they can be articulated only by experts in the area. The elicitation of rules from experts is one of the major tasks in developing an ES and is one of the significant potential drawbacks of ES. Another concern is that most rules are "brittle" in the sense that changes in circumstances, procedures, or laws can result in rules being rendered invalid. The process of changing rules in an ES is cumbersome. As indicated earlier, to the extent possible, dynamic knowledge of the parameters of rules should preferably be stored in a database rather than being programmed into the knowledge base.

There are two primary methods of developing an ES. The first is by using pre- programmed ES "shells" such as Jess, The Java Expert System Shell, EX-SYS and Open Rules. These shells come with the user-interface, inference engine, and knowledge base structure already developed. The system developer need only populate the knowledge base with rules specific to the problem area in question. Effort in building the user interface and inference engine need not be expended. Of course, the drawback of this approach is that it limits the flexibility of the developer - the interface and the functioning of the inference engine cannot be modified. The second approach is to build the entire ES using programming languages such as LISP (List Processor) and Prolog. Both LISP and Prolog are powerful AI languages that allow complex ES to be custom developed. The downside of using these languages is the time and expense involved

Chapter 5: Advanced Information Systems

18 of 31

since every component of the ES including the user interface and the inferencing mechanism must be built from scratch.

Other than rule-based knowledge representation, other knowledge representation techniques include semantic networks, frames, and object-oriented methods. In addition, a relatively recent technique, referred to as "case-based reasoning," uses past cases as the basis for providing solutions to unstructured problems. All of these knowledge representation techniques are discussed further in the Appendix.

Development of ES Selecting a knowledge representation method that is most appropriate for the target application is only one step in the process of developing an ES. Determining whether to purchase off-the-shelf ES shell software or to build the ES from the ground up using LISP or Prolog is another decision that needs to be made. But perhaps the most crucial, time consuming, and expensive step in the development process is knowledge acquisition. This step involves working closely with one or more experts in the field to elicit from them all the knowledge required for a functional ES. The knowledge to be elicited from experts includes both declarative knowledge of facts pertinent to the application as well as procedural knowledge of how to manipulate the facts. As discussed earlier, the declarative knowledge is best stored in the domain database whereas the procedural knowledge should be stored in the knowledge base. The method of populating the knowledge base is a function of the knowledge representation method used - either rules, frames, semantic networks, or cases. Exactly how is knowledge elicited from experts? One method is called verbal protocol analysis. Experts are asked to verbalize their thoughts and decision making process as they evaluate and make decisions in a hypothetical case. The experts’ verbalizations are recorded and subsequently transcribed. These verbal protocols are then analyzed to discern both the declarative and procedural aspects of knowledge for the particular application. Identifying the "right" experts to use for verbal protocol analysis is not always easy - there may be a number of legitimate contenders who might qualify as "most valuable experts." Another critical decision is the number of experts to use. If only one expert is used, then the developer runs the risk of spending a lot of time and money only to end up with a system that is tuned to the idiosyncratic decision making process of that one expert. If multiple experts are used, the developer has the added problem of reconciling differing points of view between the experts (they may not all agree on the facts relevant to the decision and the process of using those facts to arrive at a decision). The developer who works with experts to elicit knowledge and develop the knowledge base is referred to as a knowledge engineer.

After the knowledge base has been developed, the next critical step is to validate the knowledge base. In essence, the developer seeks to verify that the declarative and procedural knowledge embedded in the ES in fact allows the system to mimic expert decision makers. There are two methods of validating an ES. The first is to feed the system past cases for which the "right" decisions are already known (i.e., the decisions made by human experts). The system's decisions are then correlated with those that were actually made. A high degree of correlation would indicate that the system's

Chapter 5: Advanced Information Systems

19 of 31

knowledge base is in fact valid. The cases fed into the system should not be the same ones that were used to develop the knowledge base in the first place.

The second method of validation involves using a panel of human experts. A set of cases are fed to both the system and to the panel of experts. The degree of correlation between the panel's decisions and the system’s recommendations would indicate whether or not the system's knowledge base is valid and reliable. The experts used during the validation phase should not be the same as those used during the development phase. It is at the validation stage that many prototype ES have failed. The failure could be on account of using the "wrong" experts during the development phase, or it could simply be that the target application is in such a dynamic field that the knowledge cannot be adequately captured and programmed into a system.

Benefits and drawbacks of ES Regardless of the type of knowledge representation method employed, ES yield a number of benefits. First, they represent a cost effective alternative to human experts. ES provide access to expertise where human experts are not available. Consider a public accounting firm with multiple locations but with all corporate taxation experts concentrated in one location at the largest office. For every corporate tax problem that arises at the smaller offices, a corporate tax expert from the central office would have to be consulted. Complex problems might require the physical presence of the corporate tax expert. If an ES can be developed for handling the majority of corporate tax problems, then tax professionals who may not be very experienced in the corporate tax field may still be able to assist clients with the help of the expert system. Of course, successful ES can result in the displacement of human experts thereby lowering a firm's employee costs. A second benefit of ES is that it improves decision consistency. As with the DSS, an ES forces users to consider all relevant information and consistently applies rules in making unstructured decisions. Since ES for complex problems can encode thousands of rules, and since these rules can be executed very quickly, ES also promote efficiency in tackling unstructured problems. Finally, because of their consultation and justification capabilities, ES are valuable as a training tool. Novices can interact with ES to solve test problems at their own pace. By invoking the explanation features (for both questions and solutions) novices can understand how the system solves complex problems. ES thus have the potential to significantly lower training costs since employees do not necessarily have to travel to a training facility. Despite the promise (or threat) that ES would result in the replacement of human experts, expert decision making in most fields has not been taken over by ES. There are a number of problems associated with ES technology that present barriers to the widespread use of ES. First, the development of ES for complex problems can be a very expensive proposition. Even if ES shells are used, the most time consuming and therefore expensive aspect of building an ES is the knowledge acquisition phase. Other than the difficulty associated with identifying the "right" expert or experts to use, the time and expense involved in consulting with these experts and performing verbal protocol analysis can be astronomical for very complex systems.

Apart from the knowledge acquisition step, the other problem area in ES development is validation. Validation of the knowledge base, which involves verifying the accuracy of

Chapter 5: Advanced Information Systems

20 of 31

the rules that have been programmed into the system, can be a very difficult process. Validation efforts may ultimately reveal that all the effort expended in building the system was for naught, i.e., the knowledge base may prove to be invalid. Another problem with most ES is that they are not easily maintainable. When the knowledge for a particular application changes necessitating modifications in the ES, the knowledge engineer must be called upon to make the needed changes. Not only are such changes expensive, they may not be made on a timely basis and when the changes are ultimately made, the system may no longer be needed.

Aside from these problems in developing ES, there are a number of legitimate concerns regarding the use of these systems. The replacement of human experts with knowledge based systems raises ethical questions about the appropriate uses of information technology. The concern is that if human experts are largely replaced by ES in a particular area, at some point there may not be human experts available for the further development and refinement of ES for that area! Whether ES use results in the "de- skilling" of workers is another concern. Whether professionals who routinely use an ES come to rely on these systems to such an extent that they lose any skill they might have in the area is an open question. A related point is that users may blindly follow the system's recommendations even in the face of refuting evidence. Despite these problems and concerns with ES technology, it is apparent that many accounting applications are already benefiting from ES, and many more applications could benefit. The benefits of and problems associated with ES are summarized in the following table.

Expert Systems: Benefits and Problems Benefits of ES Problems with ES

Provides access to expertise where human experts may not be available

Difficult and time consuming process of identifying and eliciting knowledge from human experts

Greater accuracy in judgmental decision making

Difficulties with validating system's knowledge base

Improved consistency and consensus in decision making Expensive to modify and maintain

Useful as a training tool for novices because of the consultation and justification features

Potential de-skilling of users

Preserves expertise within the firm

Users may blindly follow system's advice even in the face of refuting evidence

ES in accounting Each of the three broad areas in accounting -- taxation, auditing, and management accounting -- have had a number of expert systems developed for various tasks. However, the vast majority of applications of ES technology in accounting have been in

Chapter 5: Advanced Information Systems

21 of 31

the auditing area. In auditing, a number of systems have been developed to assist the auditor at various stages of the audit process. PriceWaterhouseCoopers' Risk Advisor and PLANET (formerly PW-ARISK), and KPMG's Inherent Risk Analysis are examples of systems used for assessing inherent risk. A system called Internal Controls Expert developed by Deloitte & Touche, assists the auditor in assessing control risk. PriceWaterhouseCoopers’ Control Risk Assessor is a type of "smart" questionnaire for assessing the level of control risk. Systematic, Saville and AS/400 Expert are three expert systems developed by PriceWaterhouseCoopers for helping auditors identify internal controls that would minimize control risk in certain specific environments (e.g., for the IBM AS/400 environment). Audit Planning Advisor, developed by Deloitte & Touche, assists auditors in developing audit work programs. In the tax area, a system called ExperTax has long been used by PriceWaterhouseCoopers for tax planning and for determining the adequacy of tax accruals. Andrew Tobias' TaxCut system assists in individual tax return preparation and planning. Addressing the growing importance of tax minimization strategies for multinational corporations, Deloitte & Touche's International Tax Planning Expert System assists in providing tax planning advice to multinational corporations.

Apart from the audit and tax areas, a few expert systems have also been developed in management accounting. For example, Arthur D. Little has developed an expert system for analyzing cost variances. A system for assisting with capital investment decisions has been developed by Texas Instruments. Before it was acquired by Compaq, Digital Equipment Corporation had created a system called BUCKS which facilitated the analysis of divisions and projects in terms of their profitability. The following table indicates the major tasks in each of the three application areas (audit, tax, and managerial) for which expert systems have been developed.

Expert Systems in Accounting

Auditing Taxation Management Accounting

Audit planning (e.g., Deloitte & Touche's Audit Planning Advisor)

Corporate tax planning (e.g., PriceWaterhouseCoopers' Expertax)

Variance analysis (e.g., Arthur D. Little's system)

Risk assessment (e.g., PriceWaterhouseCoopers' Risk Advisor)

Individual tax preparation and planning (e.g., Andrew Tobias' TaxCut)

Capital budgeting (e.g., Texas Instrument's system)

Evaluation of internal controls (e.g., PriceWaterhouseCoopers' Control Risk Assessor)

International tax planning (e.g., Deloitte & Touche's International Tax Planning Expert System)

Profitability analysis (e.g., Digital's BUCKS system)

Chapter 5: Advanced Information Systems

22 of 31

While the examples given above suggest that ES technology is being used quite extensively in accounting, there are many applications that could benefit but which have largely been ignored for a variety of reasons. The early applications of ES technology have been in areas deemed to have the highest payoff. Accordingly, risk assessment systems were some of the first to be developed by the "Big Four" accounting firms. Firms have adopted varying strategies regarding ES, with some firms more reluctant than others to expend resources on what is seen by many as an untried, untested, and unproven technology.

Apart from resource and strategic considerations, there are a number of attributes of decision making situations which make them either more or less appropriate for ES development. First, the decisions should be relatively unstructured. Structured and semi-structured decisions can easily be tackled using more conventional systems like TPS and DSS. Second, there must be at least one identifiable expert in the field who must also be willing to expend the time and effort to assist with the knowledge acquisition process. Third, the expert must be capable of verbalizing how he or she goes about making decisions. If the expert is unable to articulate in detail how decisions are made, then knowledge acquisition will fail. A fourth prerequisite for the application of ES technology is that the decision environment should be somewhat stable. If the decision rules for a particular situation are constantly changing, then the frequent modifications that would have to be made to an ES developed for that application would likely make the system prohibitively expensive. Finally, the decision should be made with reasonably frequency. One-time decisions are obviously not good candidates for ES development. The decision for which the ES is being developed should be recurring with some periodicity. Unstructured decisions in auditing, such as setting the materiality level and determining whether the client can continue as a going concern, are certainly unstructured but must be made for every client. Therefore, such decisions would benefit from the application of ES technology, if the other criteria mentioned above are met.

Accountants can play many roles in ES among which could be (1) using a previously developed ES, (2) identifying applications that might benefit from ES technology, (3) assisting with knowledge acquisition and system validation, and (4) performing cost- benefit analyses for proposed and existing ES. The identification of potential uses of ES could be done by applying the criteria specified above. Accountants can make their expertise available either as subjects for the development of the ES knowledge base or in the validation stage to allow comparison of their decisions with the system's decisions on test cases. Regarding cost-benefit analysis, as with TPS and DSS, the benefits of ES are difficult to quantify. Associating benefits such as improved decision consistency and greater efficiency in decision making with dollars and cents is very subjective. Relative to TPS and DSS, the costs associated with ES development are somewhat harder to precisely quantify. The difficulty stems from the unpredictable nature of knowledge acquisition and validation activities - the extent of time and effort involved remains imprecise until the activities are actually undertaken.

Chapter 5: Advanced Information Systems

23 of 31

Summary

The chapter began by discussing the hierarchy of decisions made in organizations, from the highly structured transaction processing decisions at the lowest level to the highly unstructured decisions made by top management. Transaction processing systems, which have already been discussed in earlier chapters, were briefly described to provide a context for the discussion of decision support and expert systems. Decision support systems were then described in terms of the types of decisions they support, the components of such systems, and the benefits and drawbacks of such systems. An overview of artificial intelligence research was then provided. Expert systems represent one application of artificial intelligence technology. The unique features of expert systems in contrast with transaction processing and decision support systems were described. The components of expert systems were then described in some detail. The user interface, the domain database, the knowledge base, and the inference engine were identified as the key components of expert systems. Rule-based knowledge representation was discussed in the chapter, while the Appendix focused on other knowledge representation methods such as semantic networks, frames, and case- based reasoning. The benefits and drawbacks of expert systems were then discussed. Finally, expert systems in accounting were discussed. Existing applications of expert systems technology by the "Big Five" public accounting firms were identified. The characteristics of applications well suited for expert systems technology were described. The chapter concluded by discussing the accountant's role in identifying suitable applications, assisting with knowledge acquisition and validation, and performing cost- benefit analyses of proposed and existing expert systems.

Chapter 5: Advanced Information Systems

24 of 31

Key Terms Artificial intelligence Case-based reasoning Data component Data warehouse Data mart Data mining Decision support systems Declarative knowledge Domain database Expert systems Frames Inference engine Knowledge acquisition Knowledge base Knowledge engineer Knowledge-based systems Object-oriented knowledge representation Procedural knowledge Processing component Rule-based expert systems Semantic network Transaction processing systems User interface

Chapter 5: Advanced Information Systems

25 of 31

Key Web Sites • A terrific web site with a wealth of resources about Decision Support Systems

• MIT's Artificial Intelligence Laboratory -- one of the leading AI research labs

• The Yahoo directory listing on artificial intelligence

• An article by Harold Schwartz in the CPA Journal about expert systems in accounting.

• Another article in the CPA Journal about expert systems in accounting by Professor Murphy Smith

• This site has a listing of various expert systems in the medical field

• The Wikipedia entry on expert systems. Includes a link to expert systems vendors.

• The Wikipedia entry on artificial intelligence.

• The ALICE web site – the Artificial Intelligence Foundation. Here you can “chat” with ALICE – a “chat robot.” Visit this site to experience an interactive session with “ALICE.”

• The CLIPS pages give plenty of information about this programming tool for building expert systems using C, which was developed at NASA's Johnson Space Center in the mid-eighties to facilitate putting expert systems to work.

Chapter 5: Advanced Information Systems

26 of 31

Appendix -- Other Knowledge Representation Techniques The chapter discussed rule-based knowledge representation. Other, less popular knowledge representation methods are semantic networks and frame-based or object- oriented methods. A semantic network is a knowledge structure comprised of nodes and links. The nodes can represent physical objects or concepts while the links define the relationships between nodes (objects, or concepts). When the application domain is characterized by non-rule knowledge then semantic networks present a viable option. Consider that a financial institution is attempting to develop a knowledge-based system for distinguishing between financially healthy firms and firms that are in a weak financial position and therefore pose greater risk from a lending perspective. Experienced bank loan officers might be asked to provide some key characteristics of healthy and ailing firms. This knowledge is probably best represented by means of a semantic network of facts and concepts related to healthy and sick firms. Some of the facts and concepts might be "extremely-healthy-firms," "moderately-healthy-firms," "ailing-firms," "current- ratio-greater-than-two," "declining-trend-in-sales," "high-inventories," and "declining- trend-in-profit." Links might then be defined between "extremely-healthy-firms" and "current-ratio-greater-than-two" and also between "ailing-firms," "high-inventories," and "declining-trend-in-sales" and "declining-trend-in-profit." In essence, semantic networks attempt to mirror human expertise which often involves the association of related concepts. Another knowledge-representation method is frames, which is also referred to as object-oriented knowledge representation. A frame is a structure that uses fields called "slots" to store knowledge about a particular object. The fields define the characteristics of the object. Fields are also used to store statements that describe some aspect of the object. A "cash" frame might be defined as follows:

Cash Frame Example Slots Entries (values) account-type asset currency dollars normal-balance debit institution US National Bank

The slot entries define the normal expectations about the object. We expect a cash account to be an asset, to be denominated in dollars (in the U.S.), to have a debit balance, and to be located at some financial institution. When the system is confronted with a new cash account, it will be expected to have a debit balance in dollars and be associated with some financial institution. A unique aspect of frames is that the slots can also contain triggered procedures called "demons." For example, an inventory frame can have a demon that automatically triggers a back order when the on-hand quantity falls below zero. Frames are object-oriented in that they support the concept of inheritance which was described in Chapter 3. For example, the "cash" frame can be defined as a lower-level frame below the "asset" frame in which case it would automatically inherit the "asset" account type attribute.

Chapter 5: Advanced Information Systems

27 of 31

A relatively recent development in AI is case-based reasoning (CBR) systems. In fields such as medicine and law, experts often tackle new problems by recalling one or more solved problems or cases from memory. In essence, reasoning in these fields is "case-based" because experts (1) find similarities between the existing problem and solved cases and (2) apply the solutions from prior cases either directly or with modifications as needed to the new problem. Unfortunately, human beings are not infallible - even experts are sometimes unable to recall the most relevant solved cases for a problem. A computer-based CBR system builds a database of cases. Each new solution is added to the case database. The most critical step in CBR is to define the indexes or key words by which stored cases are cataloged in the databases. Inappropriately selected indexes would render the database of cases essentially useless.

When a new problem or case is presented to the CBR system, attributes relevant to the problem are elicited through interaction with the user. The system then searches through the case database and retrieves the most relevant cases. Without the assistance of the user (depending on the type of CBR system), one or more of the prior cases is applied to solve the new problem. The application of prior cases may be done either with or without modifications. As it becomes available, information about the success or failure of the CBR suggested solution is fed back to the system (e.g., case was won or lost, or patient was cured or not). The system then updates the indexes and in a sense "learns" over time because incorrect recommendations made in the past will not be repeated. By adding each solution to the database of cases, and by learning from successful and failed case solutions, the CBR system can truly become more powerful over time. The functioning of a CBR system is shown in the following figure on the next page. You are encouraged to read the Wikipedia entry on case-based reasoning systems.

Chapter 5: Advanced Information Systems

28 of 31

Discussion Questions 1. Explain the kinds of tasks for which transaction processing systems are best

suited.

2. In contrast to transaction processing systems, in what environments are decision support systems most appropriate?

3. Differentiate between transaction processing systems and decision support systems in terms of the features and capabilities of each type of system.

4. Explain the hierarchy of decision making from structured to unstructured decisions.

5. What are the components of a decision support system? Explain the functioning of each component.

6. Summarize the benefits and drawbacks of DSS.

7. What does the term "artificial intelligence" (AI) mean? What are some applications of AI?

8. Define an "expert system." Contrast expert systems from other forms of computer-based decision support.

9. Giving examples, explain heuristic reasoning.

10. Compare and contrast transaction processing systems, decision support systems, and expert systems in terms of (1) the data which is the subject of the system, (2) the nature of user interaction, and (3) the kind of processing performed by the system.

11. Briefly describe the various components of an expert system.

12. Distinguish between declarative and procedural knowledge. Explain where each type of knowledge is stored within an expert system.

13. Explain the "consultation" and "justification" features of expert systems.

14. Distinguish between forward chaining and backward chaining.

15. Other than rule-based systems, what are other kinds of knowledge representation methods for expert systems?

16. Explain the functioning of a case based reasoning (CBR) system. How are CBR systems in a sense able to "learn" from experience?

17. Discuss the knowledge acquisition process for a rule-based expert system. Indicate some of the difficulties that might be encountered during the knowledge acquisition process.

Chapter 5: Advanced Information Systems

29 of 31

18. Explain the two methods of validating expert systems.

19. Summarize the benefits and drawbacks of expert systems.

20. Discuss some applications of expert systems technology in accounting.

21. What role can the accountant play relative to the application of expert system technology in an organization? Illustrate using an example in the environment of a public accounting firm.

Chapter 5: Advanced Information Systems

30 of 31

Problems and Exercises 1. For each institution indicated, and for each of the following tasks/decisions, indicate whether the most appropriate form of computer-based support is a transaction processing system (TPS), decision support system (DSS), or expert system (ES).

• A parts distributor: Creating an aging schedule for accounts receivable

• An automobile manufacturer: Determining where to locate a new manufacturing plant.

• A financial institution: Determining the appropriate allocation between stocks, bonds, and liquid assets in an investment portfolio

• A university: Recording and processing employee expense reimbursement requests

• A CPA firm: Determining whether a client can continue as a going concern.

• A CPA firm: Establishing planning stage materiality levels.

• A bank: Identifying "high risk" customers who are likely to default on their loans.

• An industrial company: Processing maintenance requests from departments.

• A CPA firm: Performing tax planning for a client.

• A bank: Processing daily deposits from ATM machines. 2. Assume that you have been recruited by the College of Business Advising Office to assist in the development of a rule-based expert system for advising students regarding a choice of major. The system should obtain input from students regarding factors you feel are important in determining which business major best suits the interests, background, and aptitude of the student. Assume that the College of Business offers the following majors: Accounting, Management Information Systems, Finance, Marketing, and Management.

• a) Develop a list of factors that you feel are relevant to the "choice of major" decision.

• b) Develop questions to ask the student during a consultation session with the proposed system.

• c) Construct rules using these factors to arrive at conclusions regarding which major is most appropriate for the student, given his/her answers to the above questions (use the format shown in the "Sample rules for credit granting ES" table in the chapter).

3. A CPA firm is investigating potential applications of expert systems technology in its audit practice. The sequence of steps involved in auditing are: (1) understand client's business, (2) assess overall acceptable audit risk, (3) form preliminary estimate of materiality, (4) understand internal control structure of the client, (5) estimate inherent

Chapter 5: Advanced Information Systems

31 of 31

risk (susceptibility to error) of each financial statement assertion, (6) estimate control risk for each assertion (likelihood that the internal control structure will fail to prevent or detect a material error), (7) estimate allowable detection risk which drives the nature and timing of audit procedures to be performed, (8) obtain audit evidence by performing a variety of procedures, (9) evaluate the evidence, and (10) formulate an audit opinion on the client's financial statements. What advice would you have for the CPA firm regarding the suitability of expert systems technology for each of the above steps? Indicate the advantages and drawbacks of expert systems technology that the CPA firm should weigh in considering the adoption of the technology. 4. Search the World Wide Web using Google and find examples of the use of decision support systems (DSS) and expert systems (ES) technology in business settings. Write a brief report on your findings.

5. Assume that you are a senior manager with a major public accounting firm. You are considered one of the few experts in the firm on the topic of derivatives for oil and gas industry clients. Your firm’s area managing partner has invited you to become involved in the process of developing an expert system for assessing derivative contracts in the oil and gas industry. The system’s knowledge base will be built mainly using your expertise in the area. From your perspective, what are the pros and cons of your involvement in this project? From the firm’s perspective, assuming a successful development project, what are the pros and cons of deploying the ES for assessing derivative contracts throughout the firm? Last Updated: August 19, 2013

Copyright © 1996-2013 CyberText Publishing, Inc. All Rights Reserved