Chapter 7 has shown that corporate databases and files have varying levels of inaccuracy and incompleteness, which

in turn can degrade the quality of decision making. When the underlying data are poor, the quality of decision making suffers. For instance, until recently, the United ingdom!s "oyal #avy was plagued by inaccuracies and inconsistent data for the $%&,77& items of supply in its inventory systems. 't found $(,&)$ incorrect or missing entries for the attribute *+ackaging Code, alone. -anagement could not make accurate decisions on what items needed to be ordered or how many items should be kept in inventory ./ata Warehouse 'nstitute, 0&&12. Management Filters 3ven with timely, accurate information, some managers make bad decisions. -anagers .like all human beings2 absorb information through a series of filters to make sense of the world around them. -anagers have selective attention, focus on certain kinds of problems and solutions, and have a variety of biases that isolate them from reality. 4hey filter by turning off information they do not want to hear because it does not conform to their prior conceptions.For instance, Cisco 5ystems Corporation, one of the most advanced users of online decision6support systems, nevertheless was forced to write off as a loss 7).1 billion in e8cess inventory in 0&&%. Cisco!s managers built inventory in response to the output from the company!s online sales order entry system, which throughout %999 and 0&&& showed e8ceptionally strong orders. Unfortunately, Cisco managers did not pay attention to the quality of those orders. Customers, perceiving a shortage of routers and other networking equipment, were placing orders with multiple manufacturers, awarding the business to the first one who could deliver, and canceling other orders. Cisco!s systems were recording high levels of order cancellation, but management ignored this *bad news, and emphasi:ed the *good news,; new orders were piling up .<audon and <audon, 0&&)2. Organizational Inertia Chapter ) shows that organi:ations are bureaucracies with limited capabilities and competencies for acting decisively. When environments change and businesses need to adopt new business models to survive, strong forces within organi:ations resist making decisions calling for ma=or change. /ecisions taken by a firm often represent a balancing of the firm!s various interest groups rather than the best solution to the problem. Consider the record label industry. /espite declining sales since 0&&& and the e8plosion of songs downloaded for free over the 'nternet, ma=or recording companies did not move quickly into online music retailing. 4his required a much leaner business model. 4he recording companies resisted selling music over the 'nternet for years. Until very recently, nearly all e8ecutives and most employees at record label companies continued to believe that music should be distributed on physical devices .records, tapes, or C/s2 and sold through retail record stores. 5tudies of business restructuring find that firms tend to ignore poor performance until threatened by outside takeovers, and they systematically blame poor performance on e8ternal forces beyond their control such as economic conditions .the economy2, foreign competition, and rising prices, rather than blaming senior or middle management for poor business =udgment .>ohn, <ang, #etter, et. al., %9902. Trends in Decision Support and Business Intelligence 5ystems supporting management decision making originated in the early %9(&s as early

-'5 that created fi8ed, infle8ible paper6based reports and distributed them to managers on a routine schedule. 'n the %97&s, the first /55 emerged as standalone applications with limited data and a few analytic models. 355 emerged during the %9?&s to give senior managers an overview of corporate operations. 3arly 355 were e8pensive, based on custom technology, and suffered from limited data and fle8ibility. 4he rise of client@server computing, the 'nternet, and Web technologies has made a ma=or impact on systems that support decision making. -any decision6support applications are now delivered over corporate intranets. We see si8 ma=or trends; Detailed enterprise-wide data. 3nterprise systems create an e8plosion in firmwide, current, and relatively accurate information, supplying end users at their desktops with powerful analytic tools for analy:ing and visuali:ing data. Broadening decision rights and responsibilities. As information becomes more widespread throughout the corporation, it is possible to reduce levels of hierarchy and grant more decision6making authority to lower6level employees. Intranets and portals. 'ntranet technologies create global, company6wide networks that ease the flow of information across divisions and regions and delivery of near real6time data to management and employee desktops. Personalization and customization of information. Web portal technologies provide great fle8ibility in determining what data each employee and manager sees on his or her desktop. +ersonali:ation of decision information can speed up decision making by enabling users to filter out irrelevant information. Extranets and collaborative commerce. 'nternet and Web technologies permit suppliers and logistics partners to access firm enterprise data and decision6support tools and work collaboratively with the firm. Team support tools. Web6based collaboration and meeting tools enable pro=ect teams, task forces, and small groups to meet online using corporate intranets or e8tranets. 4hese new collaboration tools borrow from earlier B/55 and are used for both brainstorming and decision sessions. SYSTEMS FOR DECISION SU ORT 38actly how have these trends affected systems for business decision makingC What can today!s decision6support systems do for the firmC <et!s look more closely at how each ma=or type of decision6support system works and provides value. T!e Di""erence #et$een MIS and DSS -anagement information systems .-'52 provide information on the firm!s performance to help managers monitor and control the business. 4hey typically produce fi8ed, regularly scheduled reports based on data e8tracted and summari:ed from the firm!s underlying transaction processing systems .4+52. 4he formats for these reports are often specified in advance. A typical -'5 report might show a summary of monthly sales for each of the ma=or sales territories of a company. 5ometimes -'5 reports are e8ception reports, highlighting only e8ceptional conditions, such as when the sales quotas for a specific territory fall below an anticipated level or employees have e8ceeded their spending limits in a dental care plan. 4raditional -'5 produced primarily hard6copy reports. 4oday, these reports are available online through an intranet, and more -'5 reports can be generated on demand. 4able %)6) provides some e8amples of -'5 applications.

/ecision6support systems support decisions in a different way. Whereas -'5 primarily address structured problems, /55 support semistructured and unstructured problem analysis. An -'5 provides managers with reports based on routine flows of data and assists in the general control of the business, whereas a /55 emphasi:es change, fle8ibility, and a rapid response. With a /55 there is less of an effort to link users to structured information flows and a correspondingly greater emphasis on models, assumptions, ad hoc queries, and display graphics. T%pes o" Decision&Support S%stems 4he earliest /55 primarily used small subsets of corporate data and were heavily model driven. "ecent advances in computer processing and database technology have e8panded the definition of a /55 to include systems that can support decision making by analy:ing vast quantities of data, including firmwide data from enterprise systems and transaction data from the Web. 4oday, there are two basic types of decision6support systems, model driven and data driven ./har and 5tein, %9972. -odel6driven /55 were primarily standalone systems isolated from ma=or corporate information systems that used some type of model to perform *what6if , and other kinds of analyses. 4heir analysis capabilities were based on a strong theory or model combined with a good user interface that made the model easy to use. 4he voyage6estimating /55 described in Chapter 0 and /aimlerChrysler!s 4ransportation 3fficiency 5upport 5ystem described in the chapter6opening case are e8amples of model6driven /55. 4he second type of /55 is a data6driven /55. 4hese systems analy:e large pools of data found in ma=or corporate systems. 4hey support decision making by enabling users to e8tract useful information that was previously buried in large quantities of data. Dften data from transaction processing systems are collected in data warehouses for this purpose. Dnline analytical processing .D<A+2 and data mining can then be used to analy:e the data. Companies are starting to build data6driven /55 to mine customer data gathered from their Web sites as well as data from enterprise systems. 4raditional database queries answer such questions as, *Eow many units of product number 1&) were shipped in #ovember 0&&1C, D<A+, or multidimensional analysis, supports much more comple8 requests for information, such as, *Compare sales of product 1&) relative to plan by quarter and sales region for the past two years., We describe D<A+ and multidimensional data analysis in Chapter 7. With D<A+ and query6oriented data analysis, users need to have a good idea about the information for which they are looking. Data mining, which we introduce in Chapter 7, is more discovery driven. /ata mining provides insights into corporate data that cannot be obtained with D<A+ by finding hidden patterns and relationships in large databases and inferring rules from them to predict future behavior. 4he patterns and rules then can be used to guide decision making and forecast the effect of those decisions. 4he types of information that can be obtained from data mining include associations, sequences, classifications, clusters, and forecasts; l Associations are occurrences linked to a single event. For instance, a study of supermarket purchasing patterns might reveal that when corn chips are purchased, a cola drink is purchased ($ percent of the time, but when there is a promotion, cola is purchased ?$ percent of the time. With this information, managers can make better decisions because they have learned the profitability

of a promotion. lIn sequences, events are linked over time.We might find, for e8ample, that if a house is purchased, a new refrigerator will be purchased within two weeks ($ percent of the time, and an oven will be bought within one month of the home purchase 1$ percent of the time. lClassification recogni:es patterns that describe the group to which an item belongs by e8amining e8isting items that have been classified and by inferring a set of rules. For e8ample, businesses such as credit card or telephone companies worry about the loss of steady customers. Classification can help discover the characteristics of customers who are likely to leave and can provide a model to help managers predict who they are so that they can devise special campaigns to retain such customers. Clustering works in a manner similar to classification when no groups have yet been defined. A data6mining tool can discover different groupings within data, such as finding affinity groups for bank cards or partitioning a database into groups of customers based on demographics and types of personal investments. Although these applications involve predictions, forecasting uses predictions in a different way. 't uses a series of e8isting values to forecast what other values will be. For e8ample, forecasting might find patterns in data to help manager sestimate the future value of continuous variables such as sales figures. /ata mining uses statistical analysis tools as well as neural networks, fu::y logic, genetic algorithms, and rule6based and other intelligent techniques .described in Chapter %02. 't is an important aspect of knowledge discovery, which includes selection, preparation, and interpretation of the contents of large databases to identify novel and valuable patterns in the data. 4he Window on Drgani:ations describes a data6driven /55 used for analy:ing customer data at Earrah!s hotels and casinos. 4his massive /55 mines customer data gathered by Earrah!s when people play its slot machines or use Earrah!s casinos and hotels. Earrah!s carefully analy:es these data to identify profitable customers and find ways to encourage them to spend more. 4he system has been so successful that it has become the centerpiece of Earrah!s business strategy. Components o" DSS Figure %)61 illustrates the components of a /55. 4hey include a database of data used for query and analysisF a software system with models, data mining, and other analytical toolsF and a user interface. 4he /55 database is a collection of current or historical data from a number of applications or groups. 't may be a small database residing on a +C that contains a subset of corporate data that has been downloaded and possibly combined with e8ternal data. Alternatively, the /55 database may be a massive data warehouse that is continuously updated by ma=or corporate 4+5 .including enterprise systems and data generated byWeb site transactions2. 4he data in /55 databases are generally e8tracts or copies of production databases so that using the /55 does not interfere with critical operational systems.

4he /55 software system contains the software tools that are used for data analysis. 't may contain various D<A+ tools, data6mining tools, or a collection of mathematical and analytical models that easily can be made accessible to the /55 user. A model is an abstract representation that illustrates the components or relationships of a phenomenon. A model can be a physical model .such as a model airplane2, a mathematical model .such as an equation2, or a verbal model .such as a description of a procedure for writing an order2. 3ach decision6support system is built for a specific set of purposes and makes different collections of models available depending on those purposes. +erhaps the most common models are libraries of statistical models. 5uch libraries usually contain the full range of e8pected statistical functions, including means, medians, deviations, and scatter plots. 4he software has the ability to pro=ect future outcomes by analy:ing a series of data. 5tatistical modeling software can be used to help establish relationships, such as relating product sales to differences in age, income, or other factors between communities. Dptimi:ation models, often using linear programming, determine optimal resource allocation to ma8imi:e or minimi:e specified variables, such as cost or time. A classic use of optimi:ation models is to determine the proper mi8 of products within a given market to ma8imi:e profits. Forecasting models often are used to forecast sales. 4he user of this type of model might supply a range of historical data to pro=ect future conditions and the sales that might result from those conditions. 4he decision maker could vary those future conditions .entering, for e8ample, a rise in raw materials costs or the entry of a new, low6priced competitor in the market2 to determine how new conditions might affect sales. Companies often use this software to predict the actions of competitors. -odel libraries e8ist for specific functions, such as financial and risk analysis models. Among the most widely used models are sensitivity analysis models that ask *whatif , questions repeatedly to determine the impact on outcomes of changes in one or more factors. *What6if, analysisG working forward from known or assumed conditionsGallows the user to vary certain values to test results to better predict outcomes if changes occur in those values. What happens if we raise the price by $ percent or increase the advertising budget by 7%&&,&&&C What happens if we keep the price and advertising budget the sameC /esktop spreadsheet software, such as -icrosoft 38cel or <otus %606), is often used for this purpose .see Figure %)6$2. Hackward sensitivity analysis software is used for goal seeking; 'f ' want to sell % million product units ne8t year, how much must ' reduce the price of the productC 4he /55 user interface permits easy interaction between users of the system and the /55 software tools. A graphic, easy6to6use, fle8ible user interface supports the dialogue between the user and the /55. 4he /55 users can be managers or employees with no patience for learning a comple8 tool, so the interface must be relatively intuitive. -any /55 today are being built with Web6based interfaces to take advantage of the Web!s ease of use, interactivity, and capabilities for personali:ation and customi:ation. Huilding successful /55 requires a high level of user participation to make sure the system provides the information managers need. Business 'alue o" DSS /55 can be used in many ways to support decision making. 4able %)61 lists e8amples of /55 in well6known companies. Hoth data6driven and model6driven

/55 have become very powerful and sophisticated, providing fine6grained information for decisions that enable the firm to coordinate both internal and e8ternal business processes much more precisely. 5ome of these /55 are helping companies with decisions in supply chain management or customer relationship management. 5ome take advantage of the companywide data provided by enterprise systems. /55 today can also harness the interactive capabilities of the Web to provide decision6support tools to both employees and customers. 4o illustrate the range of capabilities of a /55, the following sections describe some successful /55 applications. /aimlerChrysler!s 4ransportation 3fficiency 5upport 5ystem described in the chapter6opening case, 5hop o!s pricing optimi:ation system, and 5onoco!s inventory management system are e8amples of model6driven /55. 4he customer analysis systems used by Earrah!s and the /reyfus Corporation, as well as +arkway Corporation!s asset utili:ation system, are e8amples of data6driven /55. 5weden!s <aps Care system for managing elderly home care is an application of geographic information systems .B'52, a special category of /55 for visuali:ing data geographically.