EFTA01221384.pdf

at Northern Arizona University A Perspective on Data Mining July 1998 Authors: Dr. Kenneth Collier Dr. Bernard Carey Ms. Ellen Grusy Mr. Curt Marjaniemi Mr. Donald Sautter EFTA01221384  Table of Contents Executive Summary 1 1.0 What is Data Mining? 2 1.1 Data Mining and Knowledge Discovery in Databases (KDD) 2 1.2 Relevance to Corporate Enterprises 5 2.0 The Emergence of Data Mining: Hype or substance? 6 2.1 The Emergence of an Industry 6 2.2 Recovery front the Artificial Intelligence Hype of the 80's 7 3.0 The Proof Of Concept Period 9 4.0 What the Critics Say 11 4.1 The View of the Business Analyst I1 4.2 Views of the Press 12 4.3 Views of Those Who Have Used Data Mining 14 5.0 Early Adopters: An Industry Overview 17 6.0 Data M g Algorithms 19 7.0 Data Mining Companies: Pieces of a Larger Puzzle 27 7.1 History of Data Mining - Software Vendors 27 7.2 History of Data Mining - Service Providers 31 8.0 A Move Toward Vertical Applications 33 8.1 The Tendency of Companies to Verticalize applications 33 8.2 Who is Specializing in What' 33 9.0 The Analysts' view of the future market 34 EFTA01221385 10.0 Business Models for an Integrated Solutions Company or Consortium 36 10.1 A Collection of Software Offerings 36 10.2 Full Service Data Mining 36 10.3 Vertical Solutions Providers 37 Bibliography 38 2 EFTA01221386  Executive Summary This document presents an overview of the subject of data mining. Data mining is a stage in the overall process of Knowledge Discovery in Large Databases (KDD). Data mining is a semi- automated process for finding undiscovered patterns and/or relationships in large databases. Data mining finds its roots in the Machine Learning community whereby academicians invented and developed artificial intelligence algorithms as a basis for machine learning. Such algorithms were the basis for the interest in artificial intelligence in the 1980s, a disappointment to the corporate world due to an overselling of an immature technology at the time. The good news is that such technologies kept maturing and became the basis for an industry with solid technological and business foundations in the 1990s. In parallel with the development of data mining products, very powerful computers, networks, and database systems also came into existence that permit storing, formatting, accessing, and analyzing large operational data stores to improve decision support operations in businesses. Such systems, in combination with data mining tools, now permit the development of new knowledge management processes to apply to meeting both corporate and scientific objectives. This document rust describes data mining and the overall knowledge discovery process. It presents opinions of industry analysts concerning the benefits of data mining. 'Me document then gives examples from corporations who have used data mining technologies to meet a variety of business objectives. For those interested in the actual algorithms embedded in the data mining tools, a section is provided that summarizes the main data mining algorithms now being employed. KDD is a new industry. There are a number of companies that arc providing data mining tools and other products to support the KDD process. A section of this report presents a summary of companies providing products in this industry as of the date of the writing of this report. Last, some thoughts arc presented as to the future strategies for the use of such technologies. EFTA01221387 1.0 What is Data Mining? 1.1 Data Mining and Knowledge Discovery in Databases (KDD) "Data mining is the exploration and analysis, by automatic or semiautomatic means, of large quantities of data in order to discover meaningful patterns and rules." [4] While there are many other accepted definitions of data mining, this one captures the notion that data miners are searching for meaningful patterns in large quantities of data. The implied goal of such an effort is the use of these meaningful patterns to improve business practices including marketing, sales, and customer management. Historically the finding of useful patterns in data has been referred to as knowledge extraction, information discovery, information harvesting, data archeology, and data pattern processing in addition to data mining. In recent years the field has settled on data mining to describe these activities. [9] Statisticians have commonly used the term data mining to refer to the patterns in data that are discovered through multivariate regression analyses and other statistical techniques. As the evolution of data mining has matured, it is widely accepted to be a single phase in a larger life cycle known as Knowledge Discovery in Databases or KDD for short. The term KDD was coined in 1989 to refer to the broad process of finding knowledge in data stores. [10] The field of KDD is particularly focused on the activities leading up to the actual data analysis and including die evaluation and deployment of results. KDD nominally encompasses the following activities (see Figure 1): I) Data Selection — The goal of this phase is the extraction from a larger data store of only the data that is relevant to the data mining analysis. This data extraction helps to streamline and speed up the process. 2) Data Preprocessino — This phase of KDD is concerned with data cleansing and preparation tasks that arc necessary to ensure correct results. Eliminating missing values in the data, ensuring that coded values have a uniform meaning and ensuring that no spurious data values exist are typical actions that occur during this phase. 3) Data Transformation — This phase of the lifecycle is aimed at converting the data into a two-dimensional table and eliminating unwanted or highly correlated fields so the results are valid. 4) Data Mining — The goal of the data mining phase is to analyze the data by an appropriate set of algorithms in order to discover meaningful patterns and rules and produce predictive models. This is the core element of the KDD cycle. S) Interpretation and EV211.116011 — While data mining algorithms have the potential to produce an unlimited number of patterns hidden in the data, many of these may not be meaningful or useful. 11is final phase is aimed at selecting those models that arc valid and useful for making future business decisions. 2 EFTA01221388  Tramforming Target Dala Dam Warehouse Figure 1The Traditional KDD Paradigm The result of this process is newly acquired knowledge formerly hidden in the data. This new knowledge may then be used to assist in future decision making. The Center for Data Insight has extended this process model in the following ways: Framing the Question(s) - One of the common misconceptions about data mining is that one can blindly set the algorithms loose on the data to find all interesting patterns that may be present. Data mining is not a magic panacea for curing all business ills. Rather it is a decision support tool that, when used in conjunction with business understanding, can provide a valuable means of gaining new business insights. Therefore, the first step in the KDD cycle must be to determine between one and three questions or goals to help direct the KDD focus. Actionable Results - The current KDD cycle ends with the evaluation and validation of analytical results. The difficulty with this conclusion is it does not provide a prescription for what to do with these results in business decision support. For example, a market basket analysis which tells you that people who purchase eggs tend to also purchase bread does not tell you what to do with this newly gained information. Should a supermarket put bread and eggs side by side or should it place them at opposite ends of the store to send the buyer past other goods on his/her trip from the eggs to the bread? Therefore, we feel that the important final phase in the KDD cycle is to identify a set of actionable results based on the validated models. 3 EFTA01221389  Iteration - While the current KDD cycle supports the return to previous phases to improve the data mining results, experience has shown that iteration is much more an integral element in the cycle than implied by the traditional model. The CDI has adapted for the KDD process an concept widely recognized in the software engineering community. That process is shown in Figure 2. Note that the model illustrated in Figure 2 also incorporates the additional phases of "framing the questions" and "producing actionable results". Under this model a prototype study is conducted to determine if the data will support the objectives. Successive iterations serve to refine and adjust the data and the algorithms to enhance the outcome. START Define the Ihp objectives Select Relevant Business Data An et rn ri vN EDD rota:R.' Clean and Transfor Data Data Mining Figure 2 A Refined ROD Paradigm EFTA01221390 1.2 Relevance to Corporate Enterprises Small businesses rely on their knowledge of the customer to inspire loyalty and confidence. For example, a businessman might continue to purchase business suits from the same tailor since that tailor has grown to understand his particular body shape, features, fabric likes and dislikes. Similarly, one returns to the hairstylist who has developed a personalized understanding of the client's preferences. While the small tailor or hairstylist do not necessarily have more knowledge about clothing or hair than larger clothing or beauty chains, it is their personal knowledge about their customers that gives them a business edge. Companies of all sizes can benefit by emulating what small, service-oriented businesses have always done well - creating one-to-one relationships with their customers. (4J In every industry, companies are trying to improve their business by individualizing its customer interactions. Small businesses build these customer relationships by NOTICING customer needs, REMEMBERING their preferences, and LEARNING from past interactions. Large enterprises have great difficulty accomplishing something similar since customers may never interact personally with company employees. Or if there is such interaction, it may be with different sales people or an anonymous call- center. While nothing can replace the ability of the small business owner to recognize customer preferences and patterns, data mining provides a mechanism that simulates this ability for larger enterprises. Most large companies have collected customer and transactional data for many years into operational data stores for billing and auditing purposes. Some companies have discovered that this transactional data is one of their most valuable assets. Consumers create a constant stream of data during daily business transactions. Consider placing a single telephone order to a catalog retailer. A transaction record is created by the local telephone company showing the time of the call, the number dialed, and the long distance company used to connect the call. The long distance company creates its own similar transaction record as well. Once connected, the catalog company generates a transaction record including all customer information, the catalog number, products purchased, etc. When the customer provides the requisite credit card information to pay for the order, the credit card company logs die transaction in its own operational data store. Finally, when the order is shipped the shipping company records the source and destination points of the package as well as its weight and other salient information. The amount of data collected during any given business day is staggering. Although this data is a rich source of potential knowledge about customers and business practices, many companies make the mistake of discarding their transactional data after a certain period of time. During die 1990s many companies reached the conclusion that their data is a valuable asset. These companies moved quickly to build data warehouses and data marts. Furthermore, companies such as Wal-Mart recognized the benefit of applying data mining to these rich stores of historical data. 5 EFTA01221391 2.0 The Emergence of Data Mining: Hype or substance? 2.1 The Emergence of an Industry Is there a real data mining industry? Or is data mining a lot of hype as unfortunately occurred with artificial intelligence in the 1980s? While the AI "boom" of the 1980s turned out to be a "bust", there were many beneficial outcomes of those efforts. Among these were the advancements in machine learning technologies and algorithms. Through these efforts, along with advances in database technologies, data storage capabilities, and parallel computing, a data mining industry has emerged. First as academic research, then as a collection of algorithms on the public domain, finally as a set of robust commercial products, the industry has matured in the past fifteen years to provide real advantages to those who employ it. This maturation has taken on a number of forms. • The past twenty years has seen our economy make a transition into the information age. Computers, data and information have become the basis for decision making in many industries. • Companies have and are collecting very large amounts of information about their customers, their products, their markets, their employees, their manufacturing processes, their distribution processes, and their marketing processes. This historical information can be "mined" to develop predictive models to guide future decision- making. • The field of machine learning has continued to evolve in the academic communities. New concepts, new algorithms and new computer structures and systems have been invented and applied to real world scientific and business problems. These ideas are being transitioned to industry in the form of new products. They have also become the basis for start-up companies developing entire new businesses. • Through experience, an understanding has developed that data mining is a step in a larger knowledge discovery, process. A systematic methodology has evoked to take raw data and transform it into information and to take information and transform it into knowledge to help solve real business problems. It is now understood that the larger process requires the careful use of different computer technologies in different stages. Raw operational data can be transformed into predictive models that support meeting major business objectives. Data mining plays a critical role in this overall process. • As the field of machine learning has developed, there has been a constant evolution of fragile Al technologies from the 1970s and 1980s, such as expert systems and neural computing, into mature products. These products correctly used can be successfully deployed into the business environment. • There has been an evolution and development of very powerful and efficient data base systems and associated access and query tools. • The development of fast, powerful, networked computer systems to provide massive amounts of storage and the rapid execution of data mining software has occurred in the 1990s. • There is an intense intellectual ferment occurring now in the knowledge discovery and data mining field. New divisions of existing corporations such as Angoss and IS!, or new corporations such as Unica, Cognos and Evoke have been created. These companies are developing, adapting and extending Al and other algorithms to provide data mining products targeted at specific business applications. They are 6 EFTA01221392  also developing the tools necessary to efficiently perform each step of the knowledge discovery process. Examples of such tools for intermediate steps arc data cleansing, data compression, data access, and data visualization. • There is a significant commitment by high technology companies such as "'Uniting Machines and Silicon Graphics Incorporated to provide data mining products as extensions of their core expertise in parallel computing and advanced computer graphics systems. • There is a significant commitment by large, stable corporations such as SAS, ORACLE and IBM to provide data warehousing and data mining products as extensions of their existing, core expertise in statistical data processing, on-line analytical processing, large data bases, data warehousing , powerful workstations and networked servers. • There is a significant commitment by large, stable corporations such as KPMG Peat Marwick LLP to provide complete knowledge discovery systems (building upon their extensive core expertise in business practices and systems). Such a company can understand the business objectives for a client and develop an integrated system solution. That solution will cover all aspects of the knowledge discovery process. The knowledge discovery results are deployed into businesses such as financial services, transportation, insurance, medical and many others. • Enough examples exist of the successful use of data mining to justify investing in the use of these technologies. It has been shown that data mining products have been put in an enterprise form and that very significant benefits accrue from their use. In summary, data mining and the associated computer technologies have now reached a mature point. Very important competitive advantages are being gained by a variety of companies that have used data mining. However, it must be emphasized that data mining must be used in a thoughtful and careful manner. 2.2 Recovery from the Artificial Intelligence Hype of the 80's There has been, and continues to be, a hype associated with data mining technologies that rightfully makes die prospective corporate user wary. The first cycle of hype for Al occurred in the late 1970s through the mid-1980s. The unfulfilled hype for artificial intelligence based products originally promoted in the 1980s rightfully has a negative resonance to this day. As an example, in response to the push to more automated manufacturing plants artificial intelligence approaches were hyped as a panacea for corporations seeking profits through productivity. The hype said, buy a computer vision system and the quality of your product will be doubled. The hype said, buy twenty robots and your defect rate will be halved and you can also reduce your labor force and costs. The hype said, install such and such an intelligent computer material handling system and you can turn out the lights at night and find your finished product waiting for you the next day. Neural networks were presented as an analog for the human brain even with little understanding of basic, human cognitive functions. It was thought that parallel computing should be employed on an assumption that if the software ran faster it must be smarter. Enormous investments were made. Great expectations were raised. Enormous investments were lost. Great disappointments occurred. However, even with these disappointments there were significant successes. 71te lesson learned in the 1980s was that success occurred when die expectations were reasonable and when the 7 EFTA01221393  technologies were properly used. An example of such a success is the expert system. Given a set of conditions the human expert would make an informed decision. Careful methodologies were developed to capture the knowledge of experts in certain problem domains and then encode that knowledge in rule-based computer software. The expert system was able to make similar or identical decisions to the human expert when faced with those same conditions. Examples of this were successful computer systems developed for doing medical diagnostics and repair of complex equipment such as locomotives. The data mining industry currently runs the risk of creating a second hype cycle. A data mining tools may be presented as a "magic box" in which you pour raw data and a business solution flows out. It does not work that way. Data mining can yield many misleading or unimportant findings. The user must understand which results have real meaning. Getting useful results from mining the data requires a careful understanding of the data. Data mining also requires a careful interpretation of the relationships or patterns uncovered in the data. Data mining therefore requires an extensive interaction between die person who is the business expert and the person who understands the data mining tools. The business expert must be involved. This is the person who understands the information contained in the data and can evaluate whether the output of the analytical or mining stages truly makes sense in the specific business domain. The proper use of predictive models must be carefully integrated into the actual business processes so the models can be properly evaluated and updated. The various data mining tools have certain strengths and certain weaknesses. The tool and its use must be properly matched to the expertise of the user and that person's objectives in using the tool. As part of the hype there is also a technical jargon used with data mining. Like all high tech products, such jargon makes it difficult for a new user to really understand how the data mining product really works, what it really does, what the practical results really arc, and which product is best used in which way to successfully meet a given business objective. The confusion resulting from such jargon can result in a potential client not using data mining at all. Even worse, the client may make a very expensive investment in following false paths or learning various tools just to understand how to best employ data mining in a given business. The Center for Data Insight at Northern Arizona University was created to provide a neutral environment where business clients can come and cut through the hype. They learn from staff already experienced in the knowledge discovery process and the leading data mining tools. In a low- risk way the new client considering the use of data mining can quickly get up to speed on data mining and gain insights into which tools are best to meet their business objectives. The first step in such a process is a proof-of-concept period to develop hard evidence that data mining can be cost-effectively employed to meet given business objectives. 8 EFTA01221394 3.0 The Proof Of Concept Period Data mining, knowledge discovery and computer systems technologies have matured in the past twenty years. The disappointments in AI technologies from the I980s arc now behind us. A more careful use of mature technologies is being employed. Real examples exist of the benefits of using data mining. Organizations such as KPMG and the Center for Data Insight exist to assist businesses in cutting through the present hype and learn and apply these new technologies in a cost-effective manner. A mature industry now exists to give businesses confidence that investments in new products will be supported by a stable industry. All of these factors provide confidence that a business should invest in data mining to achieve a competitive advantage. However, the first step should be to conduct a proof-of-concept study. The purpose of this step is to ensure that the most important business objectives are being met and to ensure the investment in data mining is done in the most cost-effective manner. The proof-of-concept period is used to answer the following questions. • Mat is data mining? • What do the data mining tools really do? • How should my raw operational data be structured to be compatible with data mining? • Which data-mining tool, or suite of tools, is best suited to meet my business objectives? • Is there hard evidence that can be generated by mining my data that shows that my company should invest in data mining and deploy it in my business? The proof-of-concept process is as follows. • Define the business objectives. Start with at most three objectives in order to focus the study. • Identify the corporate data that contains information related to those business objectives. • Create a sample data set that contains all relevant information. • Identify a domain expert(s) to work with a group experienced in knowledge discovery systems. • Install the data in a facility that has the computational power to handle the size of the data being examined and which has a suite of knowledge discovery tools suitable to meet the business objectives. • The domain expert(s) works with the data mining expert(s) to determine which data mining tool(s) are best suited to meet the business objectives. • Extract relationships and patterns from the business data set. • The domain expert(s) works with the data mining experts) to determine which patterns and relationships are really relevant to the business objectives. Experience in the CDI on a number of data mining projects has shown that surprising results may occur at this stage. Underlying assumptions about how a business works, how the market works, or how the customer behaves may change. 9 EFTA01221395  • Develop models to predict how data mining results can assist in meeting business objectives. • The company then decides what level of investment to make in data mining consistent with their business plan. At this point, a company will have significant evidence of how data mining can be employed to achieve a competitive advantage, training in data mining, and the skeleton of a development plan for using data mining in a cost effective manner. 10 EFTA01221396  4.0 What the Critics Say It is useful to gain a sense of the public perception of this emerging technology. To assist in this we provide perspectives from three relevant communities: business analysts, the press, and data mining users and practitioners. 4.1 The View of the Business Analyst The following is a collection of brief excerpts from some of the most influential members of the business and technology communities: From the Aberdeen Group Enterprise Data Mining Buying Guide: 1997 Edition Data mining offers technology to systematize the excavation of the value from databases. However, because of several significant differences in the types of tools, levels of expertise, and financial health among the suppliers, enterprises and lines of business will need to carefully evaluate prospective data mining suppliers and partners. ...To unleash the power of what Aberdeen calls scalable, commercial-grade data mining tools, enterprises must select, clean, and transform their data, perhaps integrate data brought in from outside sources, and essentially establish a recognizable environment for data mining algorithms. Data mining is more about data preparation than the sometimes-wonderful possibilities of any given algorithm or set of algorithms. Approximately 75% of the work in data mining is in getting the data to the point where data mining tools can actually start running. ...By misapplying a scalable, commercial-grade data mining tool, an enterprise can squander potential and possibly millions of dollars. Prom the Data Mining Systems web page at www.datagems.eom: The Gartner group predicts that data mining will be one of the five hottest technologies in the closing years of the century. ...Data mining is finally coming to the forefront of information management initiatives. Our research indicates that the majority of Global 2000 organizations will find data mining technologies to be critical to their business success by the year 2000. Historically, data mining referred to the process of applying artificial intelligence (often through an automated process) to look for patterns in large data sets. We believe the most successful information architectures incorporate both traditional data mining tools and end-user DSS tools. 11 EFTA01221397 From Herb Edelstein This past year has been one of real, albeit not spectacular, growth in the data mining industry. A survey Two Crows Corp. recently conducted provided strong evidence of corporate satisfaction... ...Of those organizations far enough along to have formed an opinion, ALL plan to continue or expand their present use of data mining. The survey also revealed that data mining is very much in its early growth stages. While many companies arc interested in the technology, only a few have active projects up and running. One of the most interesting findings was that while the process of implementing data mining is proving more difficult than expected, its results often exceed expectations. ...IT teams have bought into the myths of data mining's almost magical case of use. The myth is that if you simply put a data mining tool on your terabyte-sized database, truth, beauty, and valuable information will emerge. The reality is that successful data mining requires an understanding of the business, the data, the tools, and the process of data analysis. From David Stodder ...Many IT members grew up with the rise of Al - and many also remember what happened when the hype balloon popped and AI tumbled to the ground. Data miners know they can't afford another debacle. "Cautious optimism" defines the mood of the data mining industry. Magazine, April 1997: "Scientific Data Miners Make Use of All Tools Available" Extraction and analysis of patterns, trends, and clusters in large databases, however, can provide invaluable business information and forecasts. The general reasoning in the marketplace is that if you don't do the data mining, your competitor surely will. 4.2 Views of the Press Recently there has been much written in die general press regarding data mining. The following excepts from newspapers around the nation provide some insight into the perceptions of the press. From the San Jose Mercury News, October 6, 1997: "Dominating With Data - Dam Mining Emerges as The Future Of Business Analysis." Many...businesses built on data -- from the National Hockey League and its mostly paper-based statistics operation to retailers with their myriad daily transactions -- are beginning to realize that there arc big dollars hidden in their bulging databanks. Dozens of small and large data mining companies arc helping them go digging. While the industry is still young, practitioners believe that today's data mining trials will become tomorrow's essential tools. 12 EFTA01221398  ...Now that data storage is relatively cheap and computers can digest hoards of data with scarcely a burp, business is ready to pay attention. ...A Dataquest analyst says, "It's a fragmented market without an 800-pound gorilla at this point," she said. As happened with small data warehousing companies in the early 1990s, small data mining companies are likely to consolidate and be bought out over the next few years by larger database companies. "IBM has figured it out. Oracle hasn't figured it out yet, but they will soon enough." Flom the Minneapolis-St Paul Star Tribune, August 17, 1997 More than ever, businesses want to know as much as they can about their customers and potential customers. Using cutting-edge technology, firms such as Damark International Inc. and U.S. Bancorp are learning more about their customers through ... data mining. ... Modern data mining ... is more precise and less expensive to use than the mathematical computer models that preceded it, and as a result is more important to business. ... While die results ate not dramatic, Damark's Voda says even a small increase in orders from the 100 million catalogs Damark mails each year can make data mining worthwhile. He said cost reductions and increased sales attributed to data mining have allowed Damark to recover last year's data mining investment. "We're not looking for the Holy Grail. It's a game of inches, and it does not take a huge improvement in orders to drop a lot of money to the bottom line," Voda says. From The Washington Post, June g 1998: "Va. Tech Firm's Stock Surges 76% in Debut" Another Washington area technology company, MicroStrategy Inc. of Vienna, made a splash yesterday in its stock market debut, rising 76 percent in its first day of trading to close at $21.12 a share. Analysts said investors liked MicroStrategy because they believe it is pioneering a market that can only grow. The much-talked-about company, founded in 1989, specializes in "data mining" software that allows companies to synthesize scattered data into usable facts. With MicroStrategy software a clothing chain could, for instance, determine what color blouses sell best when it is cloudy in Washington. From the Boulder Camera, February 27, 199& IBM TAKING ON DATA MINING' International Business Machines, the world's largest computer company, launched a company- wide effort Thursday to capture a bigger share of the booming market for "business intelligence" tools. This market, which includes computers and software tools designed to enhance business decision-making, could grow to $70 billion by the turn of the century, according to analysts. 13 EFTA01221399 From the Arizona Republic, March 22, 199& "NO MORE SECRETS IN DATA MINING,' YOU'RE ORE" Acxiom is a leader among hundreds of companies around the country that now maintain vast electronic reservoirs. These companies include such retailers as Scars, Roebuck and Co., gift shop chains like Hallmark Cards Inc. and insurance companies like Allstate. "Technology has been the enabler," said [Donald] Hinman [an Axiom executive), who likens the advances to the invention of the printing press. "Today it's almost unbounded, our ability to gather, sort and make sense of the vast quantities of information." The number of data warehouses, large and small, using faster computers, the Internet and other networks now exceeds 1,000, a tenfold increase in five years. "They have gone on an information collecting binge," said Charles Morgan Jr., Acxiom's chief executive, describing the data mining explosion. 4.3 Views of Those Who Have Used Data Mining In addition to reports from the press and the opinions of analysts, there is an ever-increasing collection of case studies and anecdotal evidence that suggests that data mining technology has moved beyond the proof-of-performance phase of its evolution and is maturing into a mainstream business practice. The following are a few anecdotes from some significant corporations that have found value in data mining. 4.3.1 A Financial Institution The Center for Data Insight recently hosted analysts from a major financial institution considering data mining as a possible analytical tool. ite analysts brought their own company's data into the CDI to recreate an analysis that they had already conducted using traditional statistical methods. Unica's Model I produced a predictive model in less than one minute that had a lower margin of error than a similar model that had taken the analysts two weeks to build by hand. This result alone was significant enough to convince the company to adopt data mining as a standard business practice. 43.2 A Petroleum Company In another CDI engagement, a major petroleum company sought to profile, among other things, its most profitable credit card customers, and those customers who purchase super unleaded gasoline. The results of this study concluded that the company's best customers arc men with a middle income who drive trucks. The study also found that super unleaded customers tend to drive older cars and those driving newer cars tend to purchase regular unleaded. The evaluation of these results concluded that people buy super-unleaded gas because their cars need the higher octane to run properly. Since these customers will purchase super-unleaded in any case, the company was able to save over $500,000 by eliminating an unnecessary ad campaign. 14 EFTA01221400 4.3.3 A Financial Institution The following example was reported by the San Jose Mercury News, October 6, 1997, "DOMINATING WITH DATA - DATA MINING EMERGES AS THE FUTURE OF BUSINESS ANALYSIS." Mellon Bank, based in Pittsburgh, acknowledges using data mining techniques for everything from fraud detection to target marketing and loan underwriting. But officials tend to skirt talking about specific findings. "The fmancial service industry as a whole has been using this technology for quite a while," said Peter Johnson, vice president of Mellon's technology group. "The primary difference between what we're doing now and what we were doing in the 19130s was that we used to have to give the computer rules to help it make a decision. Now we have machine learning techniques that study the data and learn the rules themselves." "The consumer benefits from smarter, better packages at discounted rates that arc attractive to them," said Susan Maharis, Bell Atlantic's staff director of marketing information systems. "Product managers are saying, 'Gee, I never would have thought of putting those products together.' And consumers arc getting more of what they want." Consumers also are the focus of the data mining efforts inside Safeway's grocery stores in Great Britain. Safeway discovered through data mining that only eight of the 27 different types of orange juice in the stores sold enough to be profitable; the company responded by reducing the number of orange juice brands available. But when data mining revealed that only a dozen of more than 100 different types of cheese were profitable, the computer also noted that Safeway's most profitable customers were reaching for the least profitable cheese. The company kept all the cheeses rather than frustrating -- and possibly driving away -- its most valued customers. 4.3.4 A Financial Institution and a Direct Mail Order Company The following examples were reported by the Minneapolis-St. Paul Star Tribune, 9/17/97 At U.S. Bancorp, formerly First Bank System, in Minneapolis, computer specialists arc mining half a trillion bytes of computer data to fmd ways to minimize credit risks, cut customer attrition and boost the success sate for selling new bank services. "I think data mining is going to help us," said Richard Payne, vice president and manager of the customer decision analysis group at U.S. Bancorp. U S West, the 14-state regional telephone company that provides most of the local phone service in the Twin Cities, is sifting through a database of customer orders in Seattle to sec if data mining can help predict how customers will respond to new advertising campaigns. "By the end of the year, this should be deployed to 200 key people marketing U S West products," said Gloria Fatter, executive director of market intelligence and decision support for U S West in Denver. "We're going to give people who arc selling telephone services a more precise way to do it." And at direct-mail catalog companies Fingerhut Cos. Inc. in Minnetonka and Damark International Inc. in Brooklyn Park, data miners are predicting consumer buying trends by segmenting millions of U.S. customers into groups who exhibit similar purchasing characteristics. Fingerhut uses the information to tailor mailings of the 130 different catalogs it sends to consumers each year. "There are about 3,500 variables we now study over the lifetime of a consumer's 15 EFTA01221401  relationship with us," said Andy Johnson, senior vice president of marketing at Fingerhut. "I can predict in the aggregate which customers will do similar things." Corporations such as Fingerhut have found that, by combining years worth of business transaction data with US. Census Bureau and privately gathered demographic information, they can predict consumer behavior well enough to profit from their knowledge. Fingerhut is mining a data warehouse with information about more than 10 million current customers to fmd out which arc most likely to buy certain types of products, and therefore should be mailed certain catalogs. Fingerhut mails catalogs to demographic niches such as home textiles, outdoor, home handyman, and holiday products. "Data mining is a low-cost way for us to assess the buying behavior of groups of customers," Johnson said. In a recent data mining effort, Fingerhut studied past purchases of customers who had changed residences, and found that thcy were three times more likely to buy items such as tables, fax machines, phones and decorative products, but were not more likely to purchase high-end consumer electronics, jewelry or footwear. Mining also showed that those buying patterns persist for 12 weeks after the consumer moves, but that purchases peak in about the first four weeks. As a result, Fingerhut has created a catalog aimed at people who have recently moved, and carefully tailored it to their purchasing patterns, Johnson said. He hastened to say the data mining results don't mean that all consumers who move will avoid buying computers or shoes, but that "people will not buy at a rate that would justify our investment in printing catalogs." Another data mining effort discovered a subset of Fingerhut customers who responded more favorably to catalogs printed in Spanish. Billing, customer service, and customer correspondence also arc provided in Spanish. Aside from the language in which they were printed, the Spanish catalogs varied only slightly from a standard Fingerhut catalog of the same type: They contained a somewhat greater emphasis on fine jewelry, which data mining showed to be particularly appealing to people who like Spanish language catalogs, lie said. 'Ile result: The Spanish catalog generates about 40 percent more orders than would normally be expected from those consumers, Johnson said. Data mining also helps Fingerhut identify customers who are not likely to purchase or not likely to purchase enough to make the catalog mailing profitable. This group, dubbed "potential attritcrs," is the target of special promotions to win them back, such as discount certificates. Customers identified as unlikely to be profitable enough to justify continued catalog mailings are likely to be dropped from Fingerhut's current mailing lists. ...Within the direct-mail catalog industry, data mining's biggest contribution to date has been in cutting costs but allowing companies to focus mailings on the customers most likely to buy, Johnson said. "The big carrot is to develop a way to fmd additional revenue. That's what everyone is seeking." He declined to say how much Fingerhut has saved as a result of data mining or how much additional revenue it has obtained. But, he adds, "We arc the second-largest catalog company in the United States, and we would not be in business without the market segmentation data that data mining produces." 16 EFTA01221402  5.0 Early Adopters: An Industry Overview This section presents an overview of the first industries to embrace data mining such as banking, insurance retail, telecommunications and direct mail order companies.. Many companies within these industries have benefited from their experiences as indicated by the previous section. Si Finance In 1988 Richard Fairbank and Nigal Morris shopped the concept around to 25 banks before Signet Banking Corporation decided to give data mining a try. Signet acquired behavioral data from many sources and used it to build predictive models. Using these models it launched the highly successful balance transfer card product that changed the way the credit card industry works. Now Data mining is at the heart of the marketing strategy of all the so-called monoline credit card banks: First USA, MBNA, Advanta, and Capital One. (5) Credit card divisions have led the charge of banks into data mining, but other divisions are not far behind. At First Union, a large North Carolina-based bank, data mining techniques arc used to predict which customers are likely to be moving soon. For most people, moving to a new home in another town means closing the old bank account and starting up a new account, often with a new bank. First Union set out to improve retention by identifying customers who arc about to move and making it easier for them to transfer their business to another First Union branch in the new location. Not only has retention improved markedly, but also a profitable relocation business has developed