Data to Knowledge to Results: Building an Analytic Capability...Results: Building an Analytic...

Data remains one of our most abundant yet

under-utilized resources. “Data to Knowledge to

Results: Building an Analytic Capability” pro-

vides a holistic framework that will help compa-

nies maximize this resource. By outlining all the

elements necessary to transform data into

knowledge and then into business results, the

paper helps managers understand that human

performance elements must be attended to in

addition to technology. The experience of over

20 companies who were successful in their data-

to-knowledge efforts helps us present both the

critical success factors that must be present as

well as specific, tangible advice for companies

seeking to build a robust analytic capability.

June 2000 · ©2001Accenture · Institute for Strategic Change · www.accenture.com/isc

Data to Knowledge toResults:Building an Analytic Capability

Thomas H. Davenport

Jeanne G. Harris

David W. De Long

Alvin L. Jacobson

Working Paper

Data to Knowledge to Results · ©2001 Accenture · Institute for Strategic Change · www.accenture.com/isc

Working Paper

Table of Contents

Data to Knowledge to Results: Building an Analytic Capability 3

Appendix A: Skills Matrix 24

Appendix B: Transformation Technologies 25

Case Study: The Earthgrains Company 28

Case Study: Harrah’s Entertainment 32

Case Study: Hewlett-Packard 34

Case Study: Kraft Foods 38

Case Study: Wachovia Bank 43

Case Study: U S West 46

About the Authors 49

About the Accenture Institute for Strategic Change 49


how it could save over $100 million by changing its pur-chasing patterns.

� Wachovia Bank uses customer transaction data to sup-port a modeling process that evaluates each branch’s cur-rent and long-term profitability. In Atlanta, the bank’slargest market, Wachovia showed significant performanceimprovements when it used the outputs of the modelingprocess as a basis to decide which of its 96 branches toclose and in which locations to open new ones.

The problem is that turning data into knowledge just doesn’thappen often enough. And even if a company does manage totransform data into knowledge and then into results, it is rarelya sustained or widespread process. In our experience, the com-panies mentioned above were exceptions. In the rush to ensurethat employees and suppliers are paid, orders are taken accu-rately, debits and credits are posted to the right ledgers, and thedata warehouse is full, most organizations have neglected whatwe consider to be the most important step in the data transfor-mation process: the human realm of analyzing and interpretingdata and then acting on the insights.

Throughout our research, we found that while companies haveemphasized important technology and data infrastructure ini-tiatives, they have virtually ignored the organizational, cultur-al, and strategic changes necessary to leverage their invest-ments. They lack the broad capabilities needed to performhigh-level data-based business analysis and the cultures, busi-ness processes, and performance measures needed to makeand implement data-driven decisions.

The purpose of this paper is to present a framework thatclearly identifies and articulates the primary success factorsthat must be present in order to build broad organizationalcapabilities for transforming data into knowledge and theninto business results.4 Throughout the paper, we will not onlydescribe our findings, but we include specific, tangible advicegarnered from our research. The final section of the paper willprovide managers with a general, action-oriented guide as tohow to implement the model as a whole.

Our Research Method

We were persuaded that companies have a problem in thisarea by our participation in previous studies of certain typesof data environments.5 To begin our research, we conductedsome initial exploratory interviews in order to better under-

One of the most enduring traits of the information age is thatwe have focused too much on mastering transaction data andnot enough on turning it into information and knowledge thatcan lead to business results. The information systems in orga-nizations gather zillions of bits and bytes of data from busi-ness transactions in order to serve operational or record keep-ing needs. We are awash in data on topics ranging from cus-tomer purchases, supplier payments, loan repayment sched-ules, work hours by charge code, and the amount of educationand training received by employees. Despite our growing abili-ties to collect all this data, however, most of us are stillstruggling to develop the very capabilities that prompted us togather data in the first place: the ability to aggregate, ana-lyze, and use data to make informed decisions that lead toaction and generate real business value.1

It’s not that data doesn’t get turned into knowledge andresults at all, nor that we don’t really know how to do it.Many of the tools and skills that can help organizations lever-age transaction data have been around for years. Decisionsupport, executive information systems, online analytic pro-cessing, and data warehousing and data mining were alldesigned to help us meet these needs. In fact, one of us co-authored a book on executive information systems more thanten years ago, and other books on decision support existedwell prior to that.2 The problem is not that sufficient moneyisn’t being spent on analytic technologies. Today the marketfor “business intelligence and data warehousing” tools andservices, according to one market research organization, isgrowing at an average rate of more than 50% and is expectedto reach $113 billion by 2002.3

And it’s not that some companies aren’t making significantprogress in turning data into knowledge. Our research uncov-ered the following:

� Earthgrains, a $2 billion bakery products company, usesSAP data in its Refrigerated Dough Division to analyzeproduct profitability. This has led to the elimination of 20percent of the division’s product line and a 70 percentjump in the division’s earnings during the first year thatprofitability data was available.

� Owens & Minor, a $3 billion distributor of healthcareand medical supplies, has the goal of reducing supplychain costs by providing customers Web-based access to transaction data that can be used to analyze purchasingpractices. The company has already secured at least onemajor long-term contract by showing a hospital chain

Working Paper

being used to change management and decision-makingprocesses.7 In other words, is ERP data being turned into infor-mation and knowledge, or is it simply being used to processbasic business transactions? We found that some companies(e.g., Microsoft and Dow Chemical) had developed new perfor-mance measurement contexts in which their ERP data fit andthat other companies (e.g., Earthgrains and Amerada Hess)had specific examples of how ERP data had led to money-sav-ing and money-making business decisions. However, thereweren’t enough of these success stories. More than 60 compa-nies with ERP systems were canvassed on the issue. Fewerthan five had made substantial progress at turning data intoknowledge, or could give any specific examples of how deci-sions and actions had been affected by ERP data.

The use of customer data also provides evidence of a generallack of analytic capabilities. Take “customer asset manage-ment” systems that track customer financial data in broker-age houses, for example. Although they could be used tosupport knowledge management initiatives such as customerproblem resolution, the forwarding of customer and productknowledge to other parts of the organization, and trendanalysis of customer interactions and product problems, theyare generally only used to support simple service transac-tions. Two multi-company studies confirmed that eventhough companies gather considerable customer data intheir transaction systems, few summarize or synthesize itinto a coherent picture of a customer across multiple differ-ent transactions and observations.8 Even when companies didsucceed in transforming customer data into knowledge, thetransformation was generally within one customer-facingfunction (e.g., sales or customer service), rather than acrossthe entire organization.

There are other anecdotal examples of the data-to-knowl-edge deficit. Scanner data in retail stores, for example, israrely turned into knowledge. One CIO of a grocery chainthat is highly regarded for its IT use confided to us that hiscompany analyzed at most only 2% of their collected data. Adifferent grocery chain in the Midwest recently decided tosimply throw its scanner data away after selling parts of itto third-party information services. The data had been savedfor years in the hope that someday it would be analyzed, butit never was.

We sense that Web transaction data will follow the same pat-tern. We interviewed managers from several small Internet-based firms with Web sites that generate substantial amountsof transaction data. Most say they plan to analyze the data at

stand the conditions under which transaction data is success-fully turned into knowledge and then into business results. Wenext developed an initial model that reflected our experienceand tested it by interviewing managers in 20 companies thathave demonstrated some success in using transaction data toimprove performance. The companies we studied in detail arelisted in Exhibit 1. A variety of industries, decision-makingdomains, and data sources were represented.6 Finally, afterpresenting our initial results to many of the companies at aworkshop, we further refined our framework by integratingtheir feedback.

Evidence of the Problem

What is the evidence that companies have difficulties turningdata into knowledge? In both systematic study and casualobservation, we have observed the lack of data-derived knowl-edge and action across a number of different situations andenvironments. Nearly everywhere we looked, we found man-agers that planned to eventually make effective use of transac-tion data, but hadn’t seemed to have gotten around to it yet-insome cases despite more than a decade of activity. Although welooked at companies using many different types of data, thefollowing technological contexts were most pervasive:

� Enterprise resource planning (ERP) systems� Customer relationship management efforts� Point-of-sale scanner data in retail stores� Web and e-commerce transaction data

The first context involves enterprise systems, transaction tech-nologies that are costing many companies billions of dollars toinstall. One of us recently completed a research project seek-ing evidence that the new, high-quality data supplied by ERPpackages-integrated, cross-functional, real-time, and global-is


Working Paper

American Century InvestmentsChase Manhattan BankThe Dow Chemical CompanyJ.D. Edwards & CompanyFleet BankHarrah's Entertainment, Inc. Kraft Foods, Inc.Owens & Minor, Inc.J Sainsbury plcU S WEST Inc.

Bank of AmericaCitgo Petroleum CorporationThe Earthgrains CompanyFirst Union CorporationHarley Davidson, Inc.Hewlett-Packard CompanyNational SemiconductorPentair, Inc.ShopLink Inc.Wachovia Bank

EEXXHHIIBBIITT 11:: LLIISSTT OOFF CCOOMMPPAANNIIEESS RREESSEEAARRCCHHEEDD

The model can be used at several different levels of an organi-zation. At the simplest level, it can be applied to describe orunderstand a particular, one-time attempt to convert datainto knowledge and results. For example, we used the modelto understand an attempt by J Sainsbury, a U.K.-based retailer,to increase sales of cat food in its stores through more effec-tive use of scanner data. In such instances, the presence orabsence of the critical elements described in the model canhelp managers understand why business value was or was notcreated. Secondly, the model can be used to develop an ongo-ing capability in a particular area of a business, with particu-lar types of data and business objectives. Most of our researchsites, for example, were working primarily in one area of theirbusinesses-retail services in banks, marketing in casinos, cata-log response in a high-tech firm. Finally, it’s possible to usethe model as a way to understand an organization’s broadcapability for turning data into knowledge and results, as afew of our research sites such as Dow Chemical and Earth-grains were attempting to do.

The relationships between components of the model must beviewed differently, however, with these different applications.Context, transformation, and outcomes can thus either beviewed sequentially in the context of a particular data-basedanalytic decision, or more holistically in the case of building aspecific or general analytic capability. We will now describeeach element and its respective components below, givingexamples wherever possible.

Contextual Factors in the Model

some point, but have not yet done so. We did find a few com-panies, however, that are attempting to utilize their data inorder to meet business objectives. We already know thatOwens & Minor is Web-enabling transaction data to help cus-tomers analyze and cut supply chain costs. Other firms areusing software to automatically customize product and pro-motional offerings on their Web sites. Just how useful theseWeb interface applications will be without greater humanintervention and involvement remains to be seen. Whenhuman attention is necessary, we believe there is still preciouslittle significant knowledge and business results emergingfrom Web-driven transaction data.

Regardless of the source of the transaction data-whether it beERP systems, marketing data warehouses, point-of-sale scan-ners, the Web, or just about any other system-the emergingpattern is clear. Companies are investing billions of dollars intechnology that generates huge volumes of transaction data.But these investments will not live up to their potential unlessfirms are able to address the strategic, organizational, andcultural barriers to building broad capabilities that make itpossible to convert data into knowledge and then into busi-ness results. The balance of this paper will describe a modeldesigned to help managers identify and diagnose the elementsinvolved in building a strong analytic capability.

A Model for Turning Transaction Data into Knowledge and Results

By understanding and implementing the critical success fac-tors presented in the model below, organizations can vastlyimprove their analytic capabilities. The model consists of threemajor elements: context, transformation, and outcomes.Together, these elements produce value for an organization.Context includes the strategic, skill-related, organizational andcultural, and technology and data factors that must be pre-sent for an analytic effort to succeed. They may be viewed asthe prerequisites of success in this process, though they arecontinually refined and affected by the other elements. Thetransformation element is where the data is actually analyzedand then used to support a business decision. Finally, out-comes are the events that change as a result of the analysisand decision-making. They include changes in behaviors,processes and programs, and financial conditions.

Working Paper

Data to Knowledge to Results ·

The business strategy at Harrah’s Entertainment also had cleardata implications. When the explosion of newly legalizedgaming jurisdictions in the mid-1990s ground to a halt, Har-rah’s managers realized that growth could no longer comefrom the construction of new casinos. Rather, growth wouldneed to come from existing casinos and from an increase ofcross-market visitations between Harrah’s 18 properties. Har-rah’s new strategy is to drive growth through customer loyaltyand data-driven marketing operations. The implementation ofthis strategy requires the use of Harrah’s extensive transactiondata amassed on the gaming behaviors and resort preferencesof existing customers.

Some firms may have defined formal performance metrics thatare used to track the success of a defined strategy. If, forexample, a company has developed its own version of a “bal-anced scorecard9,” it will be easily apparent what data needsto be compiled and summarized to assess the strategy. Otherfirms may employ another model for creating strategic value,e.g., the concept of “value-based management.” At DowChemical, for example, the concept included a focus on mea-suring and maximizing shareholder value, understanding andallocating activity-based costs, and monitoring the profit con-tributions of different business segments (e.g., products, cus-tomers, or business units). Such pre-defined measurementmodels ease somewhat the task of ensuring the fit betweendata and strategy.

When the organization’s strategy has not been clearly articulat-ed and communicated by senior executives, it’s still possible tobuild other aspects of a firm’s analytic capability in the hopethat eventually a compelling strategy will emerge. It will rarelybe productive, however, to gather substantial amounts of datawith the hope that some of it will eventually be useful.

Skills and Experience

Senior management often overlooks the unusual combinationof skills and knowledge needed to transform data into action-able decisions. Not even the most sophisticated data miningsoftware can obviate the need for a high degree of humanskill and experience in the successful analysis and use of busi-ness data. The skills and experience needed to analyze trans-action data are vastly different than the skill and experienceneeded to record transaction data. Moreover, thinking aboutdecisions in analytic terms is very different from thinkingabout them based on prior experiences. These differences areprofound enough that at least two companies in our study

Crucial to the data-to-knowledge process are the contextualfactors-those background elements that inform every decision.Decisions are not made in a vacuum, after all. They are madewithin the context of a particular business strategy, a particu-lar experience and skills level, a particular culture and organi-zational structure, and a particular set of technology and datacapabilities. Most companies, however, tend to focus on justtwo elements-technology and data-or none at all. By con-sciously addressing and improving each and every contextualelement, however, managers can expect significant improve-ments in their analytic capabilities.

Strategy

Experience has taught us that strategy is often not addressedin the context of data access and analysis. Some may evenquestion the need for it in this domain. However, without astrategic context, a company will not know which data tofocus on, how to allocate analytic resources, or what it is try-ing to accomplish in a data-to-knowledge initiative. Providinga strategic business case for the data-to-knowledge initiativecan also create support and be used to obtain funding. Finally,many of companies we spoke with, such as Kraft, First Union,and Citgo, emphasized that they actually gained deeperinsight into the value drivers of their business as a result oftying data to strategic objectives. An analytic initiative cantherefore actually improve the firm’s insights into its ownstrategic position over time.

Managers should thus ask themselves the following strategicquestions: What are our core business processes? What keydecisions in those processes, and elsewhere, need support fromanalytic insights? What information really matters to the busi-ness? How will knowing something help the business performbetter? What are the information and knowledge leveragepoints of the company’s performance? These types of choicesconstitute the strategic context for turning data into knowledge.

When an organization’s senior executives have clearly articu-lated what the strategy is, then the informational implicationsof it may be readily apparent. For example, a strategy that’sheavily focused on improving profitability will likely lead to anorientation to customer, segment, and product profitabilityanalysis. It was such a profitability-oriented strategy that leadEarthgrains to use its ERP data to understand the profitabilityof its retail customers and product lines. Besides cutting 20percent of its product line, the company also stopped sellingto about 25 percent of its lowest margin retailers who wereunwilling to change their purchasing patterns.


Working Paper

Our research identified five key competencies that must bedeveloped to varying degrees for each of these roles if a firmwants to build strong analytic capabilities. The competenciesare outlined below. For a more detailed description of the spe-cific skills and experience needed for each role, see Exhibit A.

Technology Skills

Database administrators must have deep knowledge of the coretechnical system. Business analysts and data modelers, whowork with decision-makers to frame business problems and pro-duce analytic outputs, need working knowledge of a variety ofsoftware tools, such as analytic packages and presentation soft-ware. Those making and implementing decisions will need to becomfortable with desktop browsers, query applications, anddata interpretation and presentation software.

Statistical Modeling and Analytic Skills

Firms that want to use causal or predictive business modelsfor planning or forecasting decisions will find that the abilityto select and apply appropriate analytic and statistical tech-niques to business decisions is critical. Also important is the ability to interpret and present findings and limitationsof specific statistical analyses. Data modelers will needsophisticated modeling and statistical techniques such as data visualization, regression analysis, categorical data analy-sis, and neural networks in order to design and build models.Business analysts, while needing less sophisticated modelingskills, must nevertheless be able to understand the model’sconstraints, run models, and assess the results. Decision-mak-ers, while not needing to be proficient in statistics, need tounderstand the underlying analysis so that they can properlyinterpret and act on the findings.

Knowledge of the Data

Understanding the sources, relationships, and meaning ofdata elements is often an overlooked but highly critical com-petency that is central to getting business value from elec-tronic data. Although more important for the business ana-lyst and data modeler, the decision-maker should be awareof these issues. A deep understanding of how the data isproduced and transformed often only comes from experiencein working with the database. Much of this knowledge istacit, changeable, and idiosyncratic, but without it the value

found it easier to recruit and build staff from the outsiderather than try to retrain existing staff.

Recruiting, developing, and retaining highly skilled employeeswith analytic capabilities is in fact a major human resourcechallenge for organizations that seek to transform transaction data into knowledge and results. Almost two-thirds of thecompanies we studied specifically raised this issue. Unfortu-nately, retaining analysts within a company’s bureaucraticcompensation structure is often difficult. To overcome this,many managers spend a lot of time building strong personalrelationships with their analysts and modelers.

The specific skills and experience which we found to be essen-tial for transforming data into knowledge depend, of course,upon individual roles and responsibilities. They also vary withthe scope and sophistication of the analytic capability. No oneindividual can possess all the skills and experience necessaryto transform data into knowledge. Rather, specialized roleswork together to achieve this transformation. The four keyroles we observed are as follows:

1. The Information Technology (IT) Database Administratorextracts data from multiple databases, loads it into a datamart, and presents it in a useable form to business analysts.

2. The Business Analyst and Data Modeler both work withdata to convert it into information and knowledge that canbe used by a decision-maker. Although sometimes the busi-ness analysis and data modeling functions are combinedinto one, they are often separated. In such cases, businessanalysts frame the analysis in terms of business needs whiledata modelers build computer models based on statisticalknowledge.

3. The Decision-Maker uses the analysis to make a decisionthat will impact business performance. In our research, deci-sion-makers always came from the business unit. They couldbe senior executives, middle managers, or individual contrib-utors such as salespeople.

4. The Outcome Manager, although rarely a formal role,ensures that the decision is implemented and that outcomesare achieved. Sometimes the decision-maker is also the out-come manager, but not always. Sometimes there isn’t even aneed for an outcome manager. Product pricing decisions, forexample, require little involvement from stakeholders andthus might not require an outcome manager. When manystakeholders are affected by a decision, however, an out-come manager will most likely be needed to work closelywith the decision-maker, with those affected by the deci-sion, and sometimes with the data analyst.

Working Paper


ness. This can help reduce the cost of turnover by building inredundancy.

Communication/Partnering Skills

The most sophisticated analyses in the world are worthless iffindings can’t be communicated to decision-makers in a waythat will encourage their use. Likewise, if decision-makerscan’t communicate their needs to analysts, modelers, and out-come managers, or if IT database administrators can’t commu-nicate with data modelers for that matter, then the entire

that can be derived for decision making is very limited. Onemarketing analyst in a credit card firm demonstrated thiskind of knowledge by saying:

There are 40 different types of activities recorded in ourdatabase, and I probably know 10 very well. That means I’mfamiliar with how the data is stored, the types of data fieldsused for the activity, how they are updated, and the nuances.For example, is the field populated? Is it reliable? By lookingat our ‘address change’ field, for instance, you can tellwhether is the card is active or cancelled. And by looking atanother field in the customer’s master record you can alsotell. But I need to know which one is more reliable becausesometimes they conflict.10

It can take an analyst a year or longer to know the data in alarge system. “Roger has been here 15 years and has livedthrough all the data transitions,” said one manager of mar-keting information. “He knows the tribal lore. We rely on himto understand the data, and it’s all in his head.” Such anemployee would obviously be difficult and costly to replace.

Knowledge of the Business

Designing, producing, and presenting analytic outputs is also dependent on extensive contextual knowledge of theparticular industry involved and the business issues that thedecision-makers are concerned with. Although knowledge of the business is obviously essential for the decision-makerand outcome manager, it is also imperative for the datamodeler and business analyst. Such knowledge enables themodeler and analyst to effectively access data as well astranslate and refine decision-makers’ queries. “You can’t getthe data out of the data warehouse until you know the busi-ness,” said a vice president for one major bank. “You need akind of deep knowledge of our business processes, like howthe ATM system works, for example, in order to know howthat data is created.”

Sometimes companies have to sacrifice business expertise incertain analytic positions. Because analysts and modelerswith technical and analytic skills are in such short supply,companies often have to hire people without industry knowl-edge. First Union, for example, hired analysts with manufac-turing, not banking, experience. Also, some companies suchas Bank of America encourage analysts and modelers todevelop a broad, rather than deep, knowledge of the busi-


Working Paper

IIDDEEAASS FFOORR AACCTTIIOONN::DDeevveellooppiinngg CCrriittiiccaall SSkkiillllss aanndd EExxppeerriieennccee

The following are some suggestions on how to develop the skills andexperience needed to transform data into knowledge and then intoresults.

� CCoonnssiiddeerr hhiirriinngg ffrroomm tthhee oouuttssiiddee rraatthheerr tthhaann rreettrraaiinniinnggffrroomm tthhee iinnssiiddee.. Since data-to-knowledge initiatives are sodifferent than other types of initiatives, it is often helpful tohire people already experienced in such processes. Some com-panies however, such as Kraft, were able to successfully reas-sign and educate existing IT professionals to work in their newtechnical environment.

� EEdduuccaattee bbuussiinneessss--oorriieenntteedd ppeeooppllee aabboouutt ddaattaa,, aanndd ddaattaa--oorrii--eenntteedd ppeeooppllee aabboouutt bbuussiinneessss.. Bank of America successfully edu-cated its marketing staff about data modeling through holdingcarefully crafted educational forums. Other companies havetaken a less formal approach by intentionally creating environ-ments where intense informal communication can take place.“My team spent a year sitting next to the people developing ourdata warehouse, so they could understand how it goes together,”said the vice president of customer data management and analy-sis at Fleet Bank. This group of reporting analysts still has a 20-minute daily conference call with the technical team thatupdates the database every night.

� TTeeaacchh ccoommmmuunniiccaattiioonn aanndd pprreesseennttaattiioonn sskkiillllss,, nnoott jjuusstt tteecchh--nnoollooggyy sskkiillllss.. Knowing how to communicate analytic insightseffectively may require training in consultative selling and datapresentation techniques. At Kraft Foods, a “business template” inslide format was constructed to help sales representatives com-municate the results of complex analyses to grocery managers ina clear, compelling and concise manner.

with analytic resource constraints in their own units than dealwith constraints imposed by a centralized group.” Anothermanager at First Union Bank foreshadowed this evolving rela-tionship between analysts and decision-makers, saying: “Thewhole objective of training managers to access the data is toget people to do their own analysis and queries, so we canfocus on deeper analysis. Then when they do come to us forassistance, their use of the system means they have a betterunderstanding of the data.”

Both centralization and decentralization have their trade-offs.On the one hand, decentralization allows the analyst to betterunderstand the business, the data, and the decision-makersthemselves. On the other hand, if companies go too far withdecentralization, they can lose the cutting edge knowledgethat may come from a more centralized group. This mayreduce an organization’s chance of attracting top flight ana-lysts who are inevitably drawn to working with the mostsophisticated groups. Some companies have establishedprocesses that help counter the negative aspects of oneextreme or the other. Unofficial “knowledge expediters” candirect knowledge to areas of the business that will benefit. AtHarley Davidson, for example, a marketing executive helpsexternal researchers studying customer behavior link up withdifferent parts of the business, such as product development,that would benefit from his research. Other firms rotate theiranalysts over time through particular lines of business. Thisbenefits the firm because insights can be shared and lever-aged across business units. It also often benefits the analysts,as a variety of assignments can help shape or advance theircareer opportunities.

An alternative to decentralizing or centralizing analyticresources is to simply outsource or partner to gain analyticcapabilities. Kraft Foods managers, for example, decided thatthe most effective way to develop a strong analytic capabilitywas to enter into a partnership with a leading market infor-mation supplier. This relationship is carefully structured as apartnership rather than a simple outsourcing arrangement,and includes performance metrics, evaluation processes, andan advisory board composed of senior executives from bothcompanies. For Kraft, the partnership has successfully stream-lined analytic processes and enabled new analytic capabilities.The information supplier has also benefited from the partner-ship, for it has lead to the creation of new products such asthe Hispanic data panel.

Creating a data-oriented or fact-based culture

data-to-knowledge process is at risk. A director of decisionsupport for a consumer goods company says his biggest prob-lem is getting business analysts to present their findings toproduct managers in ways that they will be understood andaccepted as useful. The analysts tend to present their fore-casting numbers as “the truth” and the only numbers man-agers should use, he said. Rather, he would have liked them topresent a range of numbers, explain the probabilities, andleave it to the managers to decide.

Managers of firms seeking to build analytical capabilities shoulddiagnose the level and structure of skills needed to supporttheir organization’s data analysis capabilities. If the skill levelsof the data analysts in an organization don’t meet needs, then afirm can’t be getting full value from its transaction data.

Organization and Culture

By far the most neglected aspect of extracting value fromtransaction data-and one of the most important factors inbuilding a firm’s analytic capabilities-is the organizational andcultural context needed to support the new activities. In fact,in one informal survey we did, over 62% of the managersresponding indicated that organizational and cultural factorswere the greatest barriers to achieving a significant return ontheir ERP investment. Although difficult to change, we foundthat the most successful companies had both created organi-zational structures and a culture and value system that sup-ported data-based analytic decisions.

Structuring Analytic Resources

Where should analytic resources be located in a company?Should they be centralized, decentralized, or even outsourced?About 45% of the companies in our study had predominatelycentralized analytic groups. Almost a third of the firms hadanalysts working primarily in decentralized business units orfunctions, while about 25% of the companies had analysts andmodelers working in both centralized and decentralized units.

We suspect that the current emphasis on centralized groups isa reflection of the fact that most firms are in the early stagesof developing data-to-knowledge capabilities. Eventually, wepredict that the distribution of analytic resources will be morebalanced. One vice president of database marketing andanalysis explained: “Over time the pressure will be to decen-tralize these resources because managers would rather deal

Working Paper


zational, cultural, and experiential factors that can help acompany maximize its technology and data quality initiatives.

The successful companies that participated in our researchhad many different technologies and unique approaches totheir technical architecture. These technology products andtools are amply discussed by vendors and the media. Since wefeel that technology and data are already overemphasized atthe expense of other important contextual factors, we willfocus this section on the human performance implications ofthe technology and data context. It is these, we found, thattruly set some companies apart from the pack.

Technology Context

Technology includes the specific hardware and software usedin data capture, cleaning, extractions, and analysis, as well asthe networking and infrastructure capabilities needed totransfer data and provide end-user access. Appendix B pro-

Creating a culture that values data-based decision-making isan ongoing and highly challenging task. Yet it is vitally impor-tant to maximizing an organization’s analytic capabilities.Without such a culture, an organization will almost certainlyfail somewhere in the process of transforming data to knowl-edge and then into results. Without solid values underlyinganalytic efforts, an organization will all too easily neglect thealready difficult to sustain capabilities necessary for success.

In an ideal situation, a data-to-knowledge effort will be seen asa natural extension of the way a company does business. Nowhere was this more apparent than at Wachovia Bank. Oneexecutive vice president there explained how data and informa-tion were part of the intrinsic value system of the firm, saying“Wachovia’s competitive position depends upon our ability touse information faster and smarter than our competition.”

Not all companies value data-based analysis and decisions asWachovia does, however. Senior management in one firm, forexample, wanted its sale force to analyze purchasing historydata to help their customers improve ordering and inventorymanagement practices. The salespeople, however, were muchmore comfortable building and maintaining personal relation-ships with the customers’ purchasing managers, so they resist-ed using the data to help customers make better supplychain-related decisions. Thus even if one area of an organiza-tion values data (the senior management in this case), thevalue system rarely permeates all levels of an organization. Foran analytic capability to really succeed, the entire organiza-tion will need to value data-based analysis and decision-mak-ing, and the outcomes that result from it.

Technology and Data

Two related aspects of the context for turning data intoknowledge and results are the technology and data environ-ments of the organization. During the course of our study, weconsistently found that companies spent the majority of theirtime, energy, and money in these two areas. This should comeas no surprise, however, for it is only after an organization’stechnology has produced high-quality data that an organiza-tion will be able to analyze and use the data as a basis forsound decision-making.

Despite the heavy focus on technology, however, most organi-zations tend to neglect two important technology and data-related areas. First of all, they tend to put too much emphasison transaction-oriented technologies and not enough on ana-lytic technologies. Secondly, they neglect some of the organi-


Working Paper

IIDDEEAASS FFOORR AACCTTIIOONN::SSttrruuccttuurriinngg AAnnaallyyttiicc RReessoouurrcceess

The following questions can help firms decide how to structure theiranalytic resources.

� HHooww ssoopphhiissttiiccaatteedd iiss tthhee aannaallyyssiiss iinnvvoollvveedd?? Complex analyticmodeling is more effectively done by a centralized group, or evenoutsourced, because of the cutting-edge skills required.

� HHooww mmuucchh llooccaall kknnoowwlleeddggee iiss nneeeeddeedd ffoorr eeffffeeccttiivvee mmooddeelliinnggaanndd aannaallyyssiiss?? Making decisions requiring extensive knowledge ofa local market context or specific product suggests decentralizinganalytic capabilities. A key factor here is how closely analystsmust interact with business unit managers. If high levels of inter-action are required, then a decentralized or matrix structure ismost suitable.

� IIss tthhee oorrggaanniizzaattiioonn’’ss ssttrruuccttuurree aanndd ccuullttuurree hheeaavviillyy oorriieenntteeddttoowwaarrdd cceennttrraalliizzaattiioonn oorr ddeecceennttrraalliizzaattiioonn?? Firms such asHewlett-Packard and Kraft Foods, with a history of highlyautonomous business units, will find it very difficult to centralizeanalytic resources in ways that are not aligned with their culture.Companies with a legacy of more centralized operations, likeFleet Bank, will find it harder to decentralize these resources. Afirm’s culture may make certain tendencies inevitable.

ships they observed. The companies in our study werefocused on leveraging the effectiveness of skilled managers,not on eliminating them.

Data Context

Our interviews showed that creating and maintaining a high-quality data context is a difficult, costly, and never-ending

vides a graphic overview of the technologies necessary tobuild a strong analytic capability.

Since analytic technologies are fundamentally different thanthe transaction technologies that most organizations are inti-mately familiar with, establishing an integrated technicalarchitecture for analytic capabilities can be difficult. First ofall, the management activities that analytic technologies sup-port are very different from the activities required to processtransactions. Because management activities such as planning,analyzing, decision-making, and exploring new ideas are oftenunstructured and poorly understood by IT professionals, man-agers’ requirements are often ignored, distorted or minimized.Little consideration is given to management’s informationrequirements until long after the transaction system isdesigned. As a result, management information is often littlemore than a by-product created by summarizing data gath-ered for transactions.

Secondly, the methods, skills and assumptions needed todevelop the technical infrastructure for an analytic capabilityare different as well. The training and experience of most ITprofessionals is geared towards building efficient transactionsystems. For example, although an IT professional may be ableto interview an order entry clerk to determine the steps anddata needed to “place a order,” it will probably be much moredifficult for this person to determine the sequence of eventsand data needed to “analyze our customer base.” Unlike theorder entry process, there are several possible routes toachieve this end. Even more difficult is the need to antici-pate less structured analyses, such as “why are sales sovolatile?” Moreover, the ability to choose and integrate ana-lytic technologies is a rare and difficult skill, since a bewil-dering number of complex and unfamiliar hardware andsoftware products are available that differ considerably fromtransaction systems.

Another difference between transaction systems and analytictechnologies is that whereas the former generally require lit-tle human involvement, the latter most often require a lot.While some vendors might still argue that new technologiescan “automate” many analytic activities, the companies inour study disagree. Analytic tools such as data mining aremost effective when they are guided by human insight. Themanagers at the mutual fund company American CenturyInvestments (ACI), for example, used a variety of technolo-gies to explore their hypotheses regarding customer valuesand segmentation. Without that strategic orientation, ACIwould have been unable to make sense of the data relation-

Working Paper


IIDDEEAASS FFOORR AACCTTIIOONN:: DDeevveellooppiinngg aa DDaattaa--BBaasseedd CCuullttuurree

We found that the most successful firms in our study-those that hadthe most pervasive data-based culture, and those that were mostregularly making effective use of data in decision-making-hademployed similar tactics to build a data-based culture. They are asfollows:

� CCuullttiivvaattee eexxeeccuuttiivvee ssppoonnssoorrsshhiipp.. Obtaining high-level supportfor analytic initiatives is one of the best ways to create a data-based culture. U S WEST took an important step toward develop-ing such a culture when a database manager created executivesupport for a customer relationship management initiative bytying it to the firm’s overall strategy. The executives even begantouting their evolving data-to-knowledge capabilities to WallStreet analysts as a source of competitive advantage for the firm.

� SSttaarrtt wwiitthh aann uunnddeerrssttaannddiinngg ooff eexxiissttiinngg nnoorrmmss aanndd pprraaccttiicceess..Understanding what the prevailing values are before one beginsto try to change the culture can help immeasurably. If, for exam-ple, the sponsor of change in the firm trying to use purchasingdata recognized that the sales force valued personal relationshipsso highly, it would have been much easier for him to address theculture change.

� SSttaarrtt ssmmaallll.. Spurring culture change at a company like Bank ofAmerica, which has 180,000 employees11, is much more difficultthan trying to promote culture change at a small, Internet basedfirm like ShopLink, which employs less than 200 people. To maxi-mize efforts at large, more traditional firms, we suggest focusinganalytic initiatives in particular functions, such as marketing, orin specific units that are more receptive. By starting small, anorganization can build momentum over time.

� RReeccooggnniizzee tthhaatt ddaattaa--bbaasseedd ccuullttuurree cchhaannggee iiss aa lloonngg tteerrmm iinnii--ttiiaattiivvee.. The companies in our study who were most successful intransitioning to a data-based culture had been pursuing thesechanges for a long time – often for 2 1/2 to 10 years.

struggle that nevertheless must be waged. Not only mightdecision-makers shy away from making data-based decisionsif good quality data is not available to them, but end-usersmight lose confidence in data-based analyses and decisions ifthey believe that the data is suspect. This can completelyundermine the ability to achieve results from a data-to-knowledge effort.

Maintaining high levels of data quality - in terms of accuracy,completeness, reliability, accessiblity, currency, and compre-hensiveness - is also quite costly in monetary terms, however.According to industry estimates, 60 - 80% of the total cost ofa data warehouse project is spent on cleaning up and inte-grating data from mainframe legacy systems and externalsources. Upwards of three-fourths of the cost of a data ware-house project is associated with just transforming the data sothat it is acceptable for analysis and reporting.

Some of the problems that lead to poor data quality includethe following:

� Inconsistent Definitions: To paraphrase an old semanticsargument, a “customer” is not a “customer” is not a “cus-tomer.” Within a large bank, for example, the retail cus-tomer is the individual or household; the institutional trustcustomer is often an institution or group; the corporatelending customer is a division or corporation; and the trea-sury or investment customer is very likely another financialintermediary. Cross-selling and cross-servicing in this typeof environment becomes a near impossibility. Identifyingand fixing inconsistent definitions is a major part of thedata preparation process. The labor intensive process ofmapping inconsistent data often taxes valuable corporateresources, for the process requires employees who havesignificant knowledge of the data and of the business.

� Incomplete and Missing Data: One of the goals of datawarehousing is to build a composite picture of your cus-tomer, yet one of the realities of life in the warehouse isthat source data is notoriously incomplete. Transactionfeeds from mainframe applications, for example, typicallylack an historical perspective, and rarely contain behav-ioral or demographic data related to life styles. Filling inthe missing blanks, identifying external source files, andmerging records is a very time consuming, very complex,and very expensive proposition.

Whether an organization is currently involved in a data-to-knowledge initiative or just beginning to think about it, it is


Working Paper

IIDDEEAASS FFOORR AACCTTIIOONN:: IImmpprroovviinngg tthhee TTeecchhnniiccaall EEnnvviirroonnmmeenntt

Here are some concrete suggestions to create a technical environmentwhich will support, rather than impede, transforming your data intoknowledge and results:

� CCrreeaattee aann IITT ssttrraatteeggyy tthhaatt ssuuppppoorrttss ddeecciissiioonnss aass wweellll aass ttrraannssaacc--ttiioonnss.. By incorporating the needs of decision-makers into initial userrequirements and by creating an IT infrastructure that supports ana-lytic as well as transaction capabilities, organizations will have astronger technology context to support data-to-knowledge initia-tives. Questions such as whether or not to standardize on analytictools and applications will need to be addressed.

� MMaaiinnttaaiinn aa ssttaabbllee tteecchhnniiccaall eennvviirroonnmmeenntt ttoo tthhee eexxtteenntt ppoossssiibbllee..Many companies in our study cited the rapid pace of technologychange as being a major obstacle to learning and embracing newtechnologies. Some organizations we studied sometimes intentional-ly held back new capabilities until users asked for them. Othersresisted the temptation to continually upgrade their technical envi-ronments in order to give their users a more stable environment.

� MMaattcchh tteecchhnniiccaall ssoopphhiissttiiccaattiioonn ttoo bbuussiinneessss nneeeeddss.. Resist the temp-tation to give every user infinite access and unlimited options for everybusiness problem. Although performing a complex analysis mayrequire a sophisticated and highly integrated analytic environment,communication of the analysis must be done with a technologyappropriate for decision-makers and the issues at hand. At Kraft Foods,for example, highly skilled analysts used one tool set to perform cate-gory management analysis but standard slide presentation template tocommunicate the analysis.

� IInntteeggrraattee ddeecciissiioonn--mmaakkiinngg wwiitthh bbuussiinneessss pprroocceesssseess.. Linking data,analytic tools and a transaction-based process into a single “applica-tion” can help ensure that decisions are executed consistently andthat the desired outcomes are achieved. For example, if sales fore-casts and manufacturing plans share common data and applications,a manufacturer can do a much better job managing supply anddemand. Involving end-users early in the design process is critical tosuccessfully tying decision-making into an operational process.

� LLeett tthhee ssttrraatteeggiicc vvaalluuee ooff ddeecciissiioonnss ddrriivvee ‘‘mmaakkee oorr bbuuyy’’ cchhooiicceess..Deciding whether or not an analytic application lends your organiza-tion a competitive advantage can greatly simplify the ‘make or buy’choice. Although some companies in our study were unwilling to buystandard software because they believed that proprietary softwaregave them significant advantage, many other companies with com-mon decision activities decided to use pre-built applications (e.g.,credit scoring, capital budgeting, and profitability analyses) andtailor them to specific business requirements.

The analytic process involves the means by which databecomes knowledge. The process is essentially a transforma-tion of quasi-finished data (quasi because there is often morecleaning, massaging, and augmenting of data necessary duringthe process) into useful insights and knowledge. It may involvstatistical analysis, generation and testing of hypotheses, con-struction of mental or mathematical models, relating data-based knowledge to that derived from human interactions,and many other approaches and techniques.

The analytic process varies considerably based on the level ofstructure of the question that the data is designed to answer.12

Variables that change depending on the question’s level ofstructure include the decision-maker’s level of involvement,the need for highly skilled analysts, and the degree of technol-ogy and automation that can be employed in the process.What may initially seem to be a highly structured question(such as “to whom should we mail the new product promo-tional materials in order to get the biggest response?”) may bemuch less structured than it appears. What do we mean by“biggest response?” Is it the number of calls and/or returnmail items from the promotional offering, the number of firsttime buyers, the dollar amount of purchases, or the totalnumber of purchases? How might other ongoing promotionsdistort the response to the mailings? Are we talking aboutdirect channels only, or indirect as well? The ambiguities insuch questions can multiply quickly.

While the level of structure in the question is clearly a contin-uum, it is typical to break the continuum into three points, aswe have done with the sample questions in Exhibit 2 and inthe discussion below.

Highly-Structured Questions

Some questions are highly-structured, unambiguous, andstraightforward. Such questions tend to have few and clearlydefined variables. An analyst addressing a highly-structuredquestion, for example, will most likely assume that the vari-ables associated with the question (e.g., sales, state, quarter,or product X) have been defined by everyone in the organiza-tion in the same way. The data associated with such questionsalso tends to be easily accessible and easily interpreted.Answers are simply presented in the form of a list or report.

Analytic problems that pose highly-structured questions maystill be very time consuming for analysts, but they are rela-tively uncomplicated and require few interactions with the

not too early to start to get data fields in shape. Work donenow will pay big dividends later in terms of reducing the costsand complexity of transforming raw data into useful results.Most importantly, a high-quality data environment will helpbuild confidence and trust in actually using the results ofsophisticated decision support tools.

The Transformation Process

At the center of the overall model are the processes by whichdata actually becomes knowledge and is applied in decisionsand actions. The processes have two iterative and intertwinedphases: the analytic process and the decision-making process.

The Analytic Process

Working Paper


IIDDEEAASS FFOORR AACCTTIIOONN:: CCRREEAATTIINNGG AANNDD MMAAIINNTTAAIINNIINNGG HHIIGGHH QQUUAALLIITTYY DDAATTAAFirms can get started building a high-quality data environment bydoing the following:

� BBuuiilldd aa ddiiccttiioonnaarryy ooff kkeeyy tteerrmmss.. Key terms such as customer,revenues, and expenses should be defined and understood interms of the differences that may occur by product, division,geography, and time. The lack of a common language was so per-vasive and troublesome at Texaco, for example, that the companyembarked on a special initiative to develop consistent definitionson a global basis. The result was a dramatic reduction in mis-communication problems.

� CCoonnssttrruucctt aa ddaattaa mmaapp.. An organization should have a map thatshows what information comes from what sources and applica-tions, how often the information is updated, and where to go tofind what you are looking for. Hoffman-LaRoche, a major phar-maceutical firm, mapped all the information that is used inbringing a new drug through the regulatory approval cycle. Theresult for one new drug was a significant decrease in time tomarket.

� AAppppooiinntt aa ddaattaa qquuaalliittyy tteeaamm.. A dedicated data quality team canhelp to establish data quality standards as well as assess theconsistency, accuracy, and completeness of data - including datafrom external sources. Although Hewlett-Packard did not estab-lish a formal data quality team, the company did execute aworldwide initiative to establish customer data quality standards.This initiative resulted in a significant increase in overall accura-cy, coverage, and completeness of information.

about interpretations and explanations of the data. Further-more, the less likely it is that the process of analysis can beautomated or turned over completely to computer systems.Analytic results of unstructured questions thus often come inthe form of a presentation or memo that presents a project orstudy.

Semi-Structured Questions

But how does the analytic process work when the desiredquestion falls in between these extremes? This characterizesmany analytic situations, and has long been the focus of the“decision support” literature and technology. Our researchsuggests that the semi-structured analytic process is bestdescribed as a series of iterative steps that successively refine

decision-maker. In some instances, the decision-maker alonedirects the relatively structured analytic process. Decision-makers often prefer to access the data themselves and tospecify the form of analysis. Using a growing number of user-friendly software tools, managers today can instantly analyzedata in ways that previously might have taken months toaccomplish using complex queries and code. For routineanalyses that support initiatives such as a balanced scorecardor a marketing campaign, a growing list of software specificapplications will now guide managers step by step throughthe entire analytic and decision-making process.

Unstructured Questions

Other questions are more complex, exhibiting nearly the oppo-site attributes of structured questions. The decision-maker maynot even have a specific question in mind, as he or she isinstead looking for lessons or trends that can be gathered fromthe data. If there is a specific question to be answered, there isconsiderable ambiguity to the question itself and to definitionsof its key variables. An analyst who receives an unstructuredquestion must spend considerable time interacting with thedecision-maker trying to clarify his or her real informationneeds and the nature of the decision involved.13 The analyst willalso spend time with the IT staff, asking such questions as: Howdo we define how a customer was acquired in the database? Dowe track responses to our specific magazine ads? How long dowe store those responses in the database?

With unstructured questions, it may be difficult and time-con-suming to access data in the database, or even to decide whatdata is relevant. Multiple variables are usually involved, andextensive interpretation and analysis of the quantitative out-puts is required to make them meaningful for decision-mak-ers. Usually this involves the use of more sophisticated statis-tical tools; only the most sophisticated “data mining” tools,for example, can identify patterns in data without the specifi-cation of a hypothesis in advance. Even then, we found that ahighly skilled analyst must tell the software where in the datato look, must be able to reject spurious relationships or pat-terns, and must interpret the findings for reasonableness andimplications.

Analysis of unstructured questions can also involve summariz-ing and interpreting textual information (e.g., “Web farming”on the Internet14), or even mere detailed observation by thehuman brain. The more unstructured the analysis, the moreuseful it is to validate a finding by talking with real people


Working Paper

EEXXHHIIBBIITT 22:: SSAAMMPPLLEE QQUUEESSTTIIOONNSS BBEEIINNGG PPUURRSSUUEEDD BBYY CCOOMMPPAANNIIEESSIINN OOUURR RREESSEEAARRCCHH

HHiigghhllyy--SSttrruuccttuurreedd QQuueessttiioonnss

� Given current customer profiles, to which customers should wemail our new catalog?

� Which products did our customers purchase most frequently lastmonth?

� How much inventory do we have of product X?

� What were sales last week in the Southwest region?

UUnnssttrruuccttuurreedd QQuueessttiioonnss

� What is the demographic and psychographic makeup of our potential high value customers?

� Which hurts the bottom line more: inventory holding costs or hiring staff to handle frequent deliveries?

� How effective was our last marketing campaign?

� How do our customers migrate between segments?

SSeemmii--SSttrruuccttuurreedd QQuueessttiioonnss

� What is the most effective product and pricing mix for a specific product category?

� What is the next logical product we should offer to specific customers?

� How can we allocate our branch and ATM resources most effec-

anticipate resource needs as the organization’s use of databecomes more sophisticated.

In one division of Hewlett-Packard, it took several years formanagers and analysts to develop a solid working relationship.The relationship began when managers requested analysts’help in understanding their list of catalog subscribers. Cus-tomers (those who had bought from Hewlett-Packard in thepast) were indistinguishable from sales representatives’ sug-gestions for potential customers. And since the list of sub-scribers had grown but the budget had not, managers neededa way to identify and selectively target their customers. Oncethis business need was established, data files were collectedand compared so that customer profiles could be developed toenable marketing resource allocation decisions. The analystsand managers had to spend considerable time together inmeetings at the outset just to understand each other and howthey were going to approach the problem at hand. Once theresults of the analysis created successful new marketing cam-paigns, the profiling process was embedded in the division’smarketing operations, and the models were continually updat-ed. As decision-makers learned more about the database andits capabilities, they identified new problems and decisionsthat it could support. This increased the demand for analyticsupport. However, over time decision-makers realized thatthey could be more effective with direct access to the data toperform basic queries. And as they got to know the data bet-ter, they started asking more sophisticated questions. Analystsbegan to spend less time educating managers about the dataand the database, and more time discussing the implicationsof analytic findings for the business.

As analysts and decision-makers work to answer semi-struc-tured questions, tensions may arise, for the two groups mayhave different objectives, expectations, and time framesexpected with regard to the business question. We observedsuch conflicts in one of the banks that participated in ourresearch. After developing a retail customer product propensi-ty model, the bank’s analytic staff wanted to begin testing it.Their choice was to review the model in two or three countiesthat could be relied upon to represent the state’s overall con-sumer. The marketing department, in contrast, was undermanagement pressure to “get their numbers up.” They wantedto conduct the pilot in two of the state’s more populouscounties, counties that also represented a larger share of bankretail revenues. The analytic team went along with market-ing’s preferences. Following a highly successful pilot, market-ing managers were anxious to roll the model out on a broaderregional basis. The analytic group resisted, arguing that the

and approximate the decision-maker’s business needs. Theresult of such a process is usually represented in the form of amodel or simulation. Although these iterations may eventuallylead to a more structured and routine analytic process thatcan be automated to some degree, early on semi-structuredanalyses are labor intensive for both analysts and decision-makers. Indeed, one of the common concerns managers ofanalytic units shared with us was how best to allocate verylimited staff resources to the semi-structured analytic process.“Everything we do today is very ad hoc,” noted one manager.“The challenge is how to make a lot of our work routine. Whatwe want to be able to do is to make successful models a partof our business, and get out of this ad hoc thing.”

Typically, the semi-structured analytic process begins with ageneral understanding of the desired outcomes and businessobjectives. The decision-maker communicates these to thebusiness analyst. Based upon this understanding, the analystwill then perform a preliminary assessment of data availabilityand develop a “proof of concept” model to review with thedecision-maker. The preliminary model results in turn enablethe analyst and decision-maker to identify various constraintssuch as data availability, data quality, model specification andmodel options as well as to provide additional clarification ofthe business objectives. More intensive development work onthe model then follows. Throughout this latter step continuousinvolvement and interactions occur between the decision-maker, the analyst, database specialists, and other key staff(e.g., product managers and individuals involved with imple-mentation). The final steps in this process involve testing andvalidation, production (by this point the question has becomestructured and routine), and ongoing evaluation.

Effective analysts for this type of problem understand thatdecision-makers process information differently, depending ontheir orientation and cognitive styles. Thus, outputs must bepackaged with appropriate mixtures of graphics and text tosatisfy the particular audience. As one Fleet Bank analyst putit, “there are chart people and there are numbers people.” hadto design analytic outputs that satisfied both in order to com-municate her findings.

A clear finding from our study is that managers and analystsaddressing semi-structured questions view their interaction asan evolving relationship. In virtually all of the companies welooked at there was evidence that data-to-knowledge capabil-ities grew as the relationships between analysts and decision-makers evolved. Understanding the implications of theseevolving relationships is important for executives who must

Working Paper


The decision-making process itself, largely invisible within the minds of managers, is difficult to understand, document,or improve. While it is at the center of the data-to-knowledgeprocess, it may be the aspect of the process that is least sub-ject to intervention. In our research we found only a fewexamples of any attempt to address data-driven decisionprocesses.

Decisions may be based on high-quality, well-analyzed data ora multitude of other factors. Management scholars over thepast several decades have documented the sometimes tenuouslink between data, decisions, and actions, and have evendescribed the decision process as being a “garbage can”model.15 Managers may commonly eschew data altogether,

model would have to be recalibrated for a totally differentpopulation. Commented one of the analysts on the team,“Marketing, however, did not want to hear this. They simplywanted to take the model and use it. They had no understand-ing or patience for our disciplined approach to the process.”

Decision-makers are often surprised to find that as analyticprocesses mature and greater insights into the business arehad, their sense of uncertainty and ambiguity increases. Oncethe easy wins and “low hanging fruit” are picked off, deeperinsights tend to be produced that may cause decision-makersto confront fundamental beliefs about the organization’s cul-ture. For example, one bank discovered that some of its mostprofitable customers were not doctors, as they had alwaysbelieved, but rather steady employees of a chain of pancakehouses who did all of their financial business with the bank.This insight required new management decisions that contra-dicted some long held beliefs.

Different types of analyses can lead to different segmenta-tions of users, process flows, and technologies. In First UnionBank’s customer relations marketing group, different analyticapproaches have been institutionalized into four distinct deci-sion-maker segments. Each of the segments is supported by acommon data mart (called SIGMA), but different applicationssoftware and capabilities are linked to each segment. At thehighest level, senior managers were given direct desktopaccess to SIGMA Executive, an application enables decision-makers to conduct their own simple queries (e.g., sales trends,profitability by geographical areas and products) and producecustomized, graphical reports. SIGMA Explorer supports thenext level down -product and marketing managers, for exam-ple. Like the Executive software, Explorer is relatively easy touse and is available from the manager’s own desktop. Thebrowser interface contains several decision support tools thatenable decision-makers and managers to drill down throughthe data in more detail than is possible at the executive level.The third level of targeted users are analysts and campaignmanagers. Sigma Prospector supports their needs for generat-ing targeted lists, assigning prospects to customized campaigngroups, and tracking responses over time. The final level ofusers are high end modelers and analysts, individuals who areinvolved in sophisticated and highly customized data analytictechniques, e.g., data mining. The application software at thislevel is referred to as SIGMA Analyst, which supports verydetailed manipulation and design of transaction files as wellas broad range of statistical procedures.

The Decision-Making Process


Working Paper

IIDDEEAASS FFOORR AACCTTIIOONN:: CCrreeaattiinngg aann EEffffeeccttiivvee TTrraannssffoorrmmaattiioonn PPrroocceessss

Here are some ways that successful companies have created astronger analysis and decision-making capability:

� CCrreeaattee pprroocceesssseess ttoo ccuullttiivvaattee aannaallyysstt//ddeecciissiioonn--mmaakkeerr rreellaattiioonn--sshhiippss.. Since tension can easily arise between analysts and deci-sion-makers, processes should be put in place to help make therelationship as smooth as possible. At least one of the companieswe spoke with, decision-makers routinely discussed ongoingcases with the analysts on a weekly basis. Partly technical innature, they also included case counseling reviews in which ana-lysts offer advice on how to deal with difficult situations.

� SSttaarrtt wwiitthh aa bbrrooaadd oorriieennttaattiioonn ttoo tthhee pprroobblleemm.. One analystrevealed that his “secret” in working with decision-makers wasnot to get too specific too quickly. His advice was to do an initialexploration of the business issue or decision at a high level, tryto fit some data to the problem, and then go back to the deci-sion-maker to get some feedback. As he put it, “...getting boggeddown in the details up-front is guaranteed to keep you in thestarting blocks.”

� CCoonnssiiddeerr sseeggmmeennttiinngg uusseerrss,, pprroocceessss fflloowwss,, aanndd tteecchhnnoollooggiieess bbyyaannaallyyssiiss ttyyppee.. This may help firms better use analytic resources, asthey can be specialized to best serve each type of user.

� TTaakkee ddeecciissiioonn--mmaakkiinngg oouutt ooff tthhee ““bbllaacckk bbooxx..”” By becomingconscious of the decision-making process and learning from suc-cesses and failures, managers may be able to increase theirchances of making effective decisions.

ity were determined but never fully acted upon. Take the caseof a bank we studied. After careful analysis, some bank man-agers realized that the bank would be more profitable if theyclosed some targeted branches. However, due to communityopposition, the hesitation of laying off employees, and theneed to end leases prematurely, they never acted on theirinsight. To avoid a situation in which a great deal of energy isdevoted to suggesting a change that can’t be made, desiredoutcomes and sources of constraints should be consideredfrom the very beginning of an analytic effort. Outcomes may,for example, influence what data is brought to bear, how it isanalyzed, or what decisions are made. Further, opposition orimplementation difficulties that may be encountered can beplanned for early in the process.

We’ve described the three types of outcomes in our model asbehaviors, processes and programs, and financial results. Formany companies, financial results may be all that matter inthe end. However, they probably won’t be achieved without

gather it and not use it, or gather it, analyze it, and makedecisions on other factors altogether.

However, if the results of data analyses are not used to informdecisions, then what is the point of capturing and managingthe data in the first place? With some well structured prob-lems, such as in the case of credit scoring models, this linkagemay be automatic. A “yes” or “no” decision is rendered basedon the model results. In other cases such as customer seg-mentation, the decision will require more active human inter-vention. The influence of the larger organizational and culturalcontext will play an important part in influencing just howthis decision-making process is configured.

There has been a tendency for organizations to treat manage-rial decision-making as a “black box,” subject neither to expla-nation nor review. But several of the companies we studiedwere beginning to consider more active intervention in deci-sions. One firm, for example, was trying to implement “deci-sion audits,” in which not only the result of the decision butthe information and data used to inform it would be evaluat-ed retrospectively. Another firm was planning to record deci-sions and later go back to evaluate them to consider whatlessons might be learned. Still, these companies are at thevery earliest stages of managing decisions, and the invisibilityand irrationality of managerial decision-making makes it avery difficult area to address.

Outcomes in the Model

Context and transformation count for little unless somethingchanges in the organization as a result. It may be tempting tothink of the end of the process as the point at which a deci-sion is made, or even the point at which the last piece ofanalysis is completed. After all, it’s not normally the responsi-bility of data access and analysis functions to make data-informed strategic decisions, much less to carry them out.Indeed, all visibility seems to go to the outcomes -thosechanges that occur when a decision is implemented and thefinancial results that follow. That there were ever foregoingsteps involving data and analysis may easily be forgotten.

Outcomes must be included in the model because they arewhat bring value to the whole endeavor. Neither analytic find-ings nor decisions themselves yield results; ultimately, it’s theprocess of implementing a decision that determines its effecton business performance. We’ve observed several instances incompanies where, for example, sources of increased profitabil-

Working Paper


IIDDEEAASS FFOORR AACCTTIIOONN::SSuubbssttaannttiiaattiinngg aanndd MMeeaassuurriinngg OOuuttccoommeess

Here are several techniques that can improve the chances thatsuccessful outcomes will be achieved from a data-to-knowledgeinitiative. They apply to all of the outcomes that are described below.

� IIddeennttiiffyy aanndd ddooccuummeenntt oouuttccoommeess ffrroomm tthhee bbeeggiinnnniinngg.. Manycompanies we spoke with found documenting progress signifi-cantly increased the organization’s commitment to achievingdata-to-knowledge initiatives.

� MMeeaassuurree tthhee bbaasseelliinnee lleevveell ooff bbeehhaavviioorrss,, pprroocceesssseess,, oorr ffiinnaann--cciiaall ppeerrffoorrmmaannccee.. Baseline measurements makes it possible todetermine improvements from the initiative and lend the mea-surements credibility.

� IIddeennttiiffyy ssoommee ttyyppee ooff ““ccoonnttrrooll ggrroouupp..”” Even if it’s only a businessunit or group to which the initiative won’t be applied, establishinga control group can help demonstrate value. First Union’s controlgroup, for example, helped establish its position that their the ana-lytic initiative contributed $100 million to the organization.

� MMeeaassuurree oouuttccoommeess iinn aa bbuussiinneessss aarreeaa wwiitthh rreellaattiivveellyy lleesssscchhaannggee.. The rapid rate of change in the external business envi-ronment makes it more difficult to understand linkages betweenactions and outcomes. It will therefore be easier to measure out-comes where change is less pervasive.

attention to intermediate outcomes such as changes in behav-iors and new processes or projects. It is these changes thatimplement the decision, or carry forth the decision into action.Financial outcomes ultimately measure the effectiveness ofthe decision and its implementation. These outcomes willoften exhibit a close relationship, with financial resultsdepending on new processes, which depend on new behaviors.However, this won’t always be the case.

Behaviors

New or different behaviors by individuals within your organi-zation (and sometimes those outside of it in the form of customers or suppliers) are usually necessary to getbusiness value from data-to-knowledge decisions and capabil-ities. A data-based decision to focus on cost control, forexample, can require thousands or millions of individualbehaviors that work to curtail spending. If these behaviorsdon’t occur, then the point of the analysis and the decision is questionable.

Acceptance or adoption of the results of an analytic model bymanagers alone does not deliver results. Ultimately, improvedfinancial outcomes depend upon the actions of the employeeswho do the work. New behaviors are most likely when thosewho need to perform them have a clear reason for doing so.Those people “downstream” from the analytic process shouldthus know the results of the data-based analysis. For example,any program of customer or market segmentation involves adetermination-based on data-of which customers are mostprofitable to serve. In order to achieve a successful outcomefrom segmentation, however, marketing and/or salespeoplehave to treat some customers differently than others. Themost valuable customers must be given preferential treat-ment; the least valuable must be served less expensively oreven “fired.” Those serving the customers need not know thedetails of the analysis, but they must believe that it is credi-ble. Other types of behavioral outcomes include reducingexpenditures in companies when financial data analysisreveals excessive costs, switching suppliers to a preferred ven-dor, or cross-selling certain products to customers whenanalysis of data reveals that they would be likely to buy.

Processes or Programs

In order for data, analysis, and decisions to yield businessvalue, most organizations will need to initiate process andprogram changes, or create new ones altogether. Often formedin part by an aggregation of behavior changes, process and


Working Paper

IIDDEEAASS FFOORR AACCTTIIOONN:: EEnnccoouurraaggiinngg aanndd EExxeeccuuttiinngg BBeehhaavviioorr CChhaannggee

It should come as no surprise then that many of the factors that makefor successful behavioral outcomes fall within the realm oforganization and culture. Whereas the organizational and culturalcontext refers to those factors that serve as effective preconditions toanalysis and decision-making, the following findings represent theorganizational and cultural factors that help to make use of thedecision and analysis once it has been made.

� EEdduuccaattiioonn aanndd ttrraaiinniinngg.. End-user behavior change oftenrequires a carefully crafted education and training program. Longbefore the decision results were put into practice, one companyspent 3-4 months rolling out an entire communications planinforming field staff and end-users about expected behaviorchanges. Such educational programs are often crucial to the suc-cess of data-to-knowledge initiatives.

� UUsseerr ddiissccrreettiioonn aanndd ffeeeeddbbaacckk.. User involvement in the “produc-tion” version of the analysis not only increases user adoption ofnew behaviors, but often helps build credibility into the out-comes themselves. This happens when learning or feedback loopswork to accommodate user suggested changes and refinements.In a banking model designed to identify “next logical products”to sell to targeted customers for example, managers found thatincorporating end-user personalized information into the modelwas extremely important in terms of utilization and acceptance.Field and analytic staff worked together to provide a way inwhich this information could be fed back into subsequent analy-ses and decision results.

� VVaalliiddaattiioonn ooff tthhee oouuttccoommeess.. Developing trust in the validity ofanalytic outputs is an important step in securing behavioralchanges from other stakeholders. Analytic models, for example,will need to be repeatedly tested so as to demonstrate to usersthat the assumptions underlying the model are valid and that theoutcomes can be believed.

� RReewwaarrddss aanndd iinncceennttiivveess.. Existing incentive and compensationstructures have to be properly aligned with the proposed behav-iors, particularly where processes have been altered. At Earth-grains’ Refrigerated Dough Division, the sales force had histori-cally been rewarded only for the quantity of products sold. Afterthe SAP system was implemented, however, the compensationwas changed to reward salespeople 50 percent on sales volumeand 50 percent on gross profit of their accounts. This significant-ly changed the behavior of the sales force, who became muchmore interested in using data to understand the current andpotential profitability of their customers.

� PPoolliittiiccaall wwiillll.. Sometimes reward systems, education, and othersuch tactics aren’t enough to motivate people to change theirbehavior, especially when the requested behavior changes aresignificant. Although difficult and politically risky, some organi-zations have actually removed those clinging to old behaviorsand replaced them with those who can perform as desired. Forexample, executives in two companies we studied had replacedtheir entire marketing departments because employees were seento be too trapped in the non-data oriented behaviors of the oldculture. Another executive encouraged the removal of several ITand marketing directors who resisted the firm’s data-to-knowl-edge initiative.

such cross-functional processes. The flow of work and infor-mation across different functions must be modeled andplanned in advance if it is to happen seamlessly.

Financial Impacts

It is important to specify the desired financial outcomes froman analytic initiative, for they can work to measure the suc-cess of a data-to-knowledge effort. However, financialimpacts shouldn’t be sought by themselves (because theywon’t be found), but only as the end result of a variety ofbehavioral and organizational changes.

The possibilities for specifying financial results are relativelyfew, and may include improved profitability, higher revenues,lower costs, or improved market share or market value.Although increased revenue is a better value proposition, it iscost savings that usually sells a data-to-knowledge initiative,for it is much easier to specify in advance how savings will beachieved. Revenue increases, on the other hand, are difficultto predict and measure. Moreover, they are difficult to

program changes may be seen as intermediate outcomes of aparticular data-driven decision.

A program is a project that seeks to improve the business insome regard. Analysis of customer transaction data may reveal,for example, that a promotion isn’t working, and that a newmarketing initiative is needed. The programs needed may involvetraining of employees, communications about |the promotion tocustomers, salespeople, or other affected parties, even develop-ment of new computer systems to automate the process ofusing data to take action. Alternatively, new insights may evenlead to the development of new products, such as a new invest-ment fund for a newly identified customer group.

Business processes may also need modification or redesign. Adetermination that an existing process is not working effec-tively can lead to a new process design and to its implemen-tation. If data analysis suggests that a new product develop-ment process takes too long, for example, decisions may betaken to shorten it incrementally or radically. In initiativesinvolving the objective of improved profitability, data analysismight also suggest the need to modify processes. Owens &Minor’s data analysis, for example, revealed that profitabilitycould be increased if supply chain processes were redesigned.

Like process reengineering efforts, analytic outputs resultingfrom the use of transaction data may require major changesin organizational structures, roles, and work processes in orderto use new insights and capabilities provided by the system.Organizations will be particularly prone to such change if aprocess demands new skills and knowledge and increasedcoordination between operating groups. One director of rela-tionship marketing we spoke with experienced an increasedwork load brought about by new analytic capabilities. An inte-grated data infrastructure and more efficient campaign man-agement software had boosted the number of outbound mar-keting campaigns by 85%; one division alone is now doing175 direct marketing campaigns a year. The director had tototally restructure her marketing team and the processes theyused to set goals, execute, and evaluate the performance oftheir campaigns in order to handle it.

Another similarity to reengineering is that the initiativesresulting from data analysis are often cross-functional. A newprocess of serving customers, for example, may need toencompass several customer-facing functions, such as sales,marketing, logistics, and customer service. Certainly initiativesbent on improving cross-selling opportunities or determiningcustomer segmentation based on profitability would involve

Working Paper


IIDDEEAASS FFOORR AACCTTIIOONN:: CCrreeaattiinngg SSuucccceessssffuull PPrroocceesssseess oorr PPrrooggrraammss

Companies can create more successful processes or programs as anoutcome of a data-based decision by doing the following:

� Anticipate the need to change business processes and programsin advance. By knowing ahead of time what changes may benecessary to take full advantage of data-based capabilities, man-agers will be able to anticipate any opposition that may arise aswell as plan for difficult to implement cross-functional processes.

� Create effective conflict-handling behaviors. Process changesthat require cross-functional collaboration between people withvery different types of experience may cause conflict. Institutingways to handle conflict ahead of time can significantly increasethe chance that a cross-functional process change will succeed.

� Obtain senior management and stakeholder support for your pro-gram or process change. Resistance to process or programchanges is common. For example, a process designed to improveoverall profitability may encounter opposition if it is instituted atthe expense of an individual channel or product. To counter anyopposition that may arise, and to ensure the successful imple-mentation of your process or program change, appropriate sup-port will be necessary.

Applying the Model in Practice

While the model described above is a useful tool for evaluat-ing and building a general analytic capability, we found in ourresearch that that there are generally several ways companiesevolve their data-to-knowledge capabilities. In a multi-factormodel such as this one, many companies need guidance indetermining where to start and which paths are the mosteffective.

Most companies begin with the technological dimension ofthe problem. Although we feel that technology is oftenoveremphasized in analytic efforts, we do believe that itmakes sense to address one aspect of technology early on.Firms need to do an early assessment of whether or not theyhave a transaction data environment that is sufficiently robustand of sufficient quality to provide data for decision-making.If the basic data platform is not good enough, it makes senseto begin putting a new one in place-a process that can easilytake several years. It doesn’t make sense to worry about thedetails of turning data into knowledge if the data isn’t worth-while to begin with.

For example, many companies have installed new enterprisesystems from SAP, Oracle, PeopleSoft, and other vendors overthe past several years. The goal of these systems is to provideexactly the sort of integrated, cross-functional, high-qualitydata that could eventually be used in management decisionsand actions. If a company is only at the beginning of installingsuch a system, or even earlier in the process, any thoughtsabout turning such enterprise data into knowledge should bepostponed. At Dow Chemical, for example, efforts to installone of the first SAP systems in the United States began in thelate 1980s. An initiative to turn the SAP data into knowledgedidn’t begin in earnest until the mid-1990s, when substantialamounts of the SAP transaction data had become available.

Assuming that the organization has some data with which towork, there is another critical factor to assess before gettinginto the finer points of data transformation. It is an aspect ofboth the strategic and organizational contexts-whether or notsenior executives are interested in data-based decision-mak-ing. If they’re not, no amount of data or knowledge is likely tochange their minds, and the process of using data to managethe organization will at best be tactical. Middle managers whowant to build the organization’s analytic capability, but wholack supportive executives, are advised to do small projectsthat demonstrate the value of data-based decisions, or to waitfor a more receptive climate.

attribute to an analytic effort since many people are eager totake credit for revenue generation.

However they are attributed, we found many examples of com-panies that had succeeded in pinning financial results to specif-ic data-to-knowledge initiatives. Fleet Bank saved $3-4 milliondollars when a new market optimization model relying on usagepattern and market demand variables helped executives decidewhich branches to close. Fleet also exceeded its costs savingsgoal of $12 million in 1999 as a result of its Changing Channelsprogram, which encouraged customers to switch from morecostly channels such as branch banking to less costly ones suchas ATMs and Internet banking. This program was an outcome ofcareful transaction data analysis. Earthgrains also tied substan-tial savings to its data-to-knowledge initiative: it revealed a $5million a year savings obtained by using transaction data analy-sis to resolve a greater share of invoice disputes with retail gro-cery chains in their favor.


Working Paper

IIDDEEAASS FFOORR AACCTTIIOONN:: TTyyiinngg FFiinnaanncciiaall RReessuullttss ttoo DDaattaa--ttoo--KKnnoowwlleeddggee IInniittiiaattiivveess

We found the following strategies useful in helping companiesattribute financial results to data-to-knowledge efforts:

� Narrow the financial analysis. Since improved financial results willusually have many competing explanations, it’s often useful tonarrow the financial analysis so that data-derived improvementsdon’t get lost in the shuffle. Revenue or profit analysis should berestricted to the business units, product or service lines, or geogra-phies to which the data-to-results initiative applies. Segmentationprograms should be measured with regard to the costs, profits, orrevenues from serving a specific segment. Initiatives that don’tdirectly involve financial outcomes, including human resourcemanagement programs, can still be measured in financial terms-e.g., a higher level of retention should translate into subsequentreductions in recruiting costs.

� Substantiate financial results with behavioral and process out-comes. Although it may be easy to specify the financial outcomes,it isn’t always easy to convince people that faster or better deci-sions translate directly into financial outcomes. The credibility offinancial outcomes will be enhanced if there is evidence ofimprovement for all types of outcomes. It may be difficult to drawa direct chain of influence (from context to transformation to non-financial outcomes to financial outcomes), but establishing thatlinkage should always be the objective of an organization thatinvests effort and resources in using its transaction data.

very concerned about managing the organizational changesneeded to implement data-driven decisions and developingand maintaining a quality data infrastructure. Finally, seniormanagement pushed to expand the use of the bank’s evolvinganalytic capabilities into other parts of the business.

In contrast, several of the firms we studied were simply doingdata analysis, with other issues either left for line managers toaddress or not dealt with at all. We predict that this analytic-centric approach will limit positive outcomes for organizationsgoing this route; we saw evidence that some of these firmsare already running into implementation and organizationalchange issues that they had not anticipated.

For a firm to really capitalize on its analytic capability, itshould ultimately position itself in the upper right quadrant ofthe matrix in Figure 2. A firm would then have both a broad,generic approach to data-driven problem solving combinedwith a holistic approach to the strategic and organizationalconcerns raised by data-driven analysis. Perhaps because thisrequires long term support from knowledgeable senior execu-tives, we didn’t find many companies in this situation.

Summary

Since the beginning of computing in business, organizations ofall types have been generating transaction data. While a primarypurpose of the data was always supposed to be to inform man-agement decision-making (the function that managed technolo-gy thus evolved from “Data Processing” to “Management Infor-mation Systems”), this goal has never been fully realized. Firmshave purchased the technologies that enable transformation of

These two factors-a suitable transaction data environmentand supportive senior executives-are go/no go factors inactively pursuing a broad data-to-knowledge initiative. Ifthey’re both present, the organization should proceed withhaste. In our research we observed four different options forproceeding. The matrix of the options is displayed in Figure 2.

The options involve the breadth of implementation of theprocess and the primary focus of the effort. In terms ofbreadth, some firms we observed undertook a broad programof data-based analysis, building a general capability thatcould address multiple business problems and potentially servethe entire enterprise. Kraft Foods, in its alliance with a mar-keting information supplier, sought to build such a genericcapability that could address a variety of business issues relat-ed to marketing and customer relationships. Similarly, FleetBank built a corporate analytic group with more than 50 ana-lysts to deal with a variety of data-derived knowledge types.The danger, of course, is that these generalized groups will beunresponsive to business needs and will take on their ownobjectives that differ from those of the business. We did notobserve this problem in the companies we researched that hadbroad approaches, however. Instead, their biggest problemtended to be a shortage of human resources to meet thedemands of the business.

The other approach relative to breadth is to address a singlebusiness problem, with no immediate concern for the entireorganization’s data analysis needs. Hewlett-Packard’s Elec-tronic Measurement Division used the data-to-knowledgeprocess just to address the issue of maximizing catalog salesto customers. Harrah’s was focused on the single issue of cus-tomer loyalty and making effective promotions to customers.Of course, it’s possible that firms with a single focus on theanalytic function will later broaden their approach, or thatthose with a broad, general approach will narrow theirs. In ourresearch snapshot, however, we observed these two differentapproaches.

The other dimension of the matrix in Figure 2 involves the pri-mary focus of a data-to-knowledge initiative. Is it almostexclusively on data management and analysis, or is there a broader, more holistic approach addressing not only data-driven analysis, but also issues of business strategy, organiza-tional change, and behavioral and financial outcomes? AtWachovia Bank, for example, the focus was clearly on themore holistic approach. Wachovia’s analytic capabilities weretightly linked with its overall business strategy and were sup-ported by a data-oriented culture. In addition, executives were

Working Paper


data into knowledge, but the human component necessary forthis process has been given short shrift.

Our research suggests that the companies that succeed at data-derived analysis and decision-making make it part of the broad-er fabric of the organization. It is aligned with their strategies,their processes, and the behaviors of their employees. It is dri-ven by senior executives’ appetite for facts. And it is undertakenwith specific business problems and objectives in mind.

The need for turning transaction data into knowledge andresults is not going to go away. If anything, it will becomemore important. As basic business processes become moreautomated, we’ll have an even greater need to transform thedata they generate into meaningful results. Some of the dataanalysis can itself be done in an automated fashion, and someresults can even be turned into decisions and actions withoutdirect human intervention. These will be only the most routinedecision processes, however.

For the most important decisions, concerted human attentionwill continue to be necessary. As transaction data proliferates,the question of where to allocate scarce human attention willbecome a pressing problem. It is then that a broader analyticcapability will be crucial to success. The ability to frame datacollection and use within a larger strategic purpose will thusbecome more and more important. By applying our frameworkand starting now to develop broad analytic capabilities, com-panies can more readily prepare themselves for the deluge ofdata that is here and is bound to get worse.


Working Paper

8 The studies include “Transforming Customer Data Into Information,”

American Productivity and Quality Center, Houston, 1997, and “Managing

Customer Knowledge,” Concours Group, Kingwood, Texas, 1998.

9 Robert Kaplan and David Norton, The Balanced Scorecard: Translating

Strategy into Action (Harvard Business School Press, 1996).

10 From “‘My Job is in the Box:’ A Field Study of Tasks, Roles, and the

Structuring of Data Base-Centered Work,” by D.W. De Long, Dissertation

Thesis, Boston University School of Management, 1997.

11 This number does not take into account the recent merger with

NationsBank.

12 For more details on characteristics of analytic tasks, see “‘My Job is in the

Box:’ A Field Study of Skills, Roles, and the Structuring of Data Base

Centered Work,” by David W. De Long, Doctoral Thesis, Boston University

School of Management, Boston, MA, 1997.

13 One mistake managers and consultants often make is assuming decision-

makers always approach analysts with a well formed question related to

their business problem. In practice, the process of defining the analytic

problem is messy, ambiguous, and hindered by managers and analysts

who have very different language, ways of thinking, and views of the

world.

14 See Web Farming for the Data Warehouse by Richard D. Hackathorn

(Morgan Kaufmann Publisher, Inc., 1998).

15 For a more complete review of the extensive research on decision-making

see, for example, A Primer on Decision-Making: How Decisions Happen by

James G. March with the assistance of Chip Heath (New York: The Free

Press, 1994); and “Decision-Making in Organizations” by S.J. Miller, D.J.

Hickson, and D.C. Wilson in Handbook of Organizational Studies, S.R.

Clegg et. al. (eds.), (Thousand Oaks, CA: Sage Publications, 1996).

References1 It is the premise of this paper that data can significantly help inform

decisions and actions. Data-based decisions tend to reflect a reality that

is often overlooked when decisions are based on experience, intuition, or

other such factors alone. In general, data-based decisions tend to

minimize bias and unreliable human traits such as memory. They thus

tend to be more consistent, impersonal, and more cost efficient to

replicate, transfer, and leverage. Data can also be easily modeled to

predict the future, thereby enabling an organization to make decisions

that are timely and responsive to its environment.

2 One of the first books written on the subject of decision support was

Decision Support Systems: An Organizational Perspective by Peter G.W.

Keen and Michael S. Scott Morton (Addison Wesley, 1978). A subsequent

book written by one of the authors of this paper is Executive Support

Systems: The Emergence of Top Management Computer Use by J.W.

Rockart and D.W. De Long (Homewood, IL: Dow Jones-Irwin, 1998).

3 Market size figures from the 1999 Business Intelligence and Data

Warehousing (BI/DW) Program Competitive Analysis Report. World

Research, Inc., San Jose, California.

4 From here on, we will use the words “analytic” and “data-to-knowledge”

to describe the loosely structured process of using transaction data to

produce insights that impact decision-making and actions.

5 Previous studies included the following: a) an informal survey of about

100 practitioners that represented about 70 companies concerning the

use Enterprise System data; b) a study of how 20 companies used

Enterprise System data; c) a best practices survey of seven companies’

use of customer data; d) a detailed ethnographic study of the use of data

by one credit card company.

6 The only preponderance of companies from one industry in our study is

banking, a particularly intensive generator and user of transaction data.

Domains where the data was used for decision-making include customer

relationship management, channel management, brand management,

category management, promotion/pricing strategy, new product

development, customer segmentation, customer profitability, product

profitability, purchasing, product distribution and sales, inventory

forecasting, etc. Data sources included ERP systems, customer

relationship management systems, point of sale systems, and Web and e-

commerce related systems.

7 Results from this study, which was funded by the ERP vendor SAP AG, are

summarized in a report by Thomas H. Davenport, “Managing With R/3:

Effective Use of SAP Information,” September 1998, available from the

author. The study is also described in Chapter 7 of Thomas H. Davenport,

Mission Critical: Realizing the Promise of Enterprise Systems,” Harvard

Business School Press, forthcoming.

Working Paper



Working Paper

Appendix A: Skills Matrix

response to increased scope and scale of work. In other cases,the organizational structure had evolved in a way thatencouraged greater role differentiation and skill depth, such asthe creation of separate business analyst and data modelerroles.

The skills matrix below is meant to show an ideal set of skillsand experiences for each role in an organization. Less thanhalf the companies we studied, however, had such fully devel-oped roles. Many of the companies were in the process ofdeveloping these roles. In some cases, roles were evolving in

Working Paper

Appendix B:Transformation Technologies

levels to use effectively.The analytic software tools (cate-gorized by type of decision)are:

� Structured decisions (reporting and query tools)

� Unstructured decisions (data mining,neural nets)

� Semi-structured decisions (spreadsheet,decision sup-port,simulation models,statistical analysis, alertingagents)

4. Knowledge Development and Collaboration.Even the bestdata-based analyses need to be tempered with knowledgefrom other sources.To really create knowledge entails ahost of activities,some of which can be supported bycommunication and collaboration technologies. Someexamples of technologies that support knowledge devel-opment include:

� Email,videoconferencing,and mobile computing to fos-ter communications

� Collaboration tools for brainstorming and sharing knowl-edge,such as document management or electroniclibraries

� Knowledge search engines and visualization tools

� Process-driven applications,which integrate analysis intoday-to-day business processes

� Training,such as computer based training,simulationsand other learning experiences


W believe that there are four different types of technologiesneeded to fully support a data-to-knowledge capability.Theyare described below and illustrated on the following page.

1. Transaction Data Systems.Obviously,transaction systemsare a major source of data.They are the starting point ofthe data-to-knowledge transformation.The quality andlevel of integration of these systems is a critical aspect ofthe technology context.

2. Data Integration and Storage.Because transaction systemsare focused upon transactions as opposed to decisionmaking,there is generally a need to summarize,cleanseand integrate data from multiple sources.Together,thesetechnologies transform data into information that isready for decision making.Some of the technologies typi-cally used for this purpose include:

� Data warehouses and data marts

� Data cleansing tools

� Data extraction tools

3. Information Tools.There is a wealth of analytic tools avail-able which help users analyze and use information in thedecision making process.These tools are designed to sup-port differing types of analyses and require different skill


Working Paper

Appendix B:Transformation Technologies

Working Paper

These six Case Studies, highlighting the experi-

ences of individual companies in turning trans-

action data into business results, are an accom-

paniment to the larger working paper, “Data to

Knowledge to Results: Building an Analytic

Capability.”

June 2000 · ©2001Accenture · Institute for Strategic Change · www.accenture.com/isc

Data to Knowledge toResults:Case Studies in Building an AnalyticCapability

Thomas H. Davenport

Jeanne G. Harris

David W. De Long

Alvin L. Jacobson

Structuring Analytic Resources. Organizational boundariescan be barriers to the high levels of collaboration often need-ed to produce high quality analytic outputs. At Earthgrains,both the sales and customer service departments in theRefrigerated Dough Division have their own dedicated analystswho work closely with management, using SAP data to sup-port decision-making.

Orientation to Change. The Refrigerated Dough Division had arelatively new senior management team, with no emotionalinvestment in the old ways of doing things. This helped createhigh expectations concerning behavioral changes.

Data-Oriented Culture. The division’s president was known byhis colleagues as being data hungry. When he observed theinformation available from SAP, he noted enthusiastically, “It’slike getting your head blown off with data.” He had pushedhard to develop norms that would encourage employees tomake more data-driven decisions. In recent years, the qualityof management reviews had improved 100 percent, accordingto one senior manager, because executives were now muchmore reliant on numbers in explaining their performance andinvestments. This behavior had trickled down throughout theorganization, so that sales people were pushed to becomeusers of data. Management assumed that if the sales forceworked with the numbers themselves, they would be moreconfident about the ideas they were presenting to customers.

Skills and Knowledge Context

A variety of skills and knowledge are needed to leveragetransaction data for decision-making. At Earthgrains, both thecustomer service and sales departments have teams of ana-lysts whose combined skills include:� Detailed knowledge of the unit’s underlying business

processes.� Strong knowledge of the grocery industry.� Extensive skills for interpreting the meaning of the SAP

data, which requires understanding definitions of key ele-ments, how they relate, and their limitations for analysis.

Strategic Context

The company’s U.S. Refrigerated Dough Division is the onlymanufacturer of private-label refrigerated dough products inthe U.S. The company makes canned dough products that aresold in the grocery refrigerated section, including biscuits, anassortment of rolls, cookie dough, breadsticks, pie crusts, andpizza crust. Dough products are marketed nationwide undermore than 100 store brands.

The Refrigerated Dough Division competes primarily with Pills-bury, which, as the only branded manufacturer of refrigerateddough, dominates the category. Earthgrains’ strategy is tocopy Pillsbury’s successful products with private label offer-ings and to capture significant volume in the particular prod-uct. Operating margins are more than 10%, as compared withapproximately 5% in the Bakery Division.

Senior managers in the U.S. Refrigerated Dough Products unitchose to pursue a strategy of operational excellence, as opposedto taking a customer or product focus. However, although themanagement team recognized what types of decisions had to bemade in order to support strategy, the division lacked the opera-tional data it needed to make these decisions and measure theireffectiveness. Historically, the division had no integration betweenits order-to-cash, picking, delivery, and accounts receivableprocesses and systems. There was also no visibility into finishedgoods inventory if a customer had a question. And the companycouldn’t price an order until it was shipped, which significantlydelayed the invoicing process. In addition, product forecastingefforts were hampered, and the company had difficulty determin-ing manufacturing capacity. On the sales side, management alsohad limited visibility of the most and least profitable customersand products. In general, management lacked the detailed under-standing of what its business needed to improve operations.

Organizational and Cultural Context

Earthgrains exhibited some organizational and cultural factorsthat helped support the use of transaction data.


Working Paper

The Earthgrains CompanyEarthgrains is a $1.9 billion bakery products company that was formerly a subsidiary of Anheuser-Busch. Since being spun

off from the beer giant in March 1996, Earthgrains has been a publicly-traded company whose stock has appreciated more

than 200%. Its core businesses are organized in two divisions—Bakery Products and Refrigerated Dough. This case draws on

Earthgrains’ experiences in trying to leverage data from its SAP investment.

ferent types of users means everything in getting it used. Inaddition, special decision support software may also be need-ed for particular applications. Earthgrains loads data files intoManugistics, a supply chain planning and forecasting softwarepackage, which they use for analysis in a variety of ways.

Technology Support Budget. Any organization trying to lever-age transaction data for decision-making needs ongoing tech-nical support as the number of applications and the volume ofdata continues to grow. Successful use of data is likely to cre-ate demand for other applications. Earthgrains’ CIO said thatmanagement had underestimated the resources needed tosupport the ongoing rollout of new SAP modules. He explains,“Managing the technical platform is a constant juggling act.We have to keep adding application servers and disc space.”

Data Context

Relevant data issues will vary significantly across organiza-tions, but Earthgrains illustrates several factors that canimpede the use of data for decision-making.

Controlling Data Integrity. The ability to develop and main-tain “clean” data is always important, but acceptable errorrates will vary across functions and industries. Integrity wasless of a problem in Earthgrains’ Refrigerated Dough Divisionwhere transaction data captured in the SAP system was creat-ed internally based on sales orders and shipping invoices.While management could control these processes, the compa-ny’s Bakery Division was trying to use scanner data from gro-cery stores to transform its distribution processes. Data frommany retailers, however, had proved to be of variable unac-ceptable quality, limiting its usefulness. Clearly, it is easier tocontrol the quality of transaction data created within an orga-nization than across entities.

Synthesizing Data from Other Sources. SAP transaction datamust almost always be integrated with data from othersources, such as third party vendors, to make it useful formanagement decision-making. Earthgrains used census datacombined with scanner data to analyze trends in differentproduct categories, and to try to interpret what was happen-ing in the marketplace.

Completeness of Data. To be useful, transaction data mustinclude the data elements of fields that can be usefully com-pared to provide insights for decision-making. For example, ifthe system captures data on product sales, but the specificitem cannot be linked back to the specific promotion under

� Thorough working knowledge of several analytic and datapresentation software packages.

� Strong interpersonal skills needed to train and support endusers, particularly the sales people who were likely tobecome frustrated when they started working with the data.

Technology and Data Context

Earthgrains’ lack of operational data to support decision-mak-ing changed with the implementation of SAP’s R/3. Most ofthe ERP system’s modules had been implemented by early1999. The Refrigerated Dough Division now has unprecedentedvisibility into its operations and customer base, which has dra-matically changed its operations. Several modules had alsobeen installed in the Bakery Division, but so far, deploymentthere had been more limited. In addition to the hardware andsoftware technologies needed to create, capture and store thetransaction data used to support decision-making, severalother technology elements are present:

Data Communications. When transaction data crosses organi-zational boundaries, the communications technology neededto transfer the data between entities becomes an importantfactor. Earthgrains had invested considerable resources todevelop its electronic data interchange (EDI) capabilities.

Data Access Tools. Earthgrains has given its 28-member salesforce laptops to access highly detailed sales data now avail-able through the system. This access tool enables access toSAP data, but also constrains the types of questions analystsand managers can ask. For example, the vice president of cus-tomer services says, “SAP is good if you want to look at orderX by customer by day, but it’s very hard to use if you want tosee five customers over three days.”

Data access tools are probably the most visible example of soft-ware or hardware capabilities that can impede the use of trans-action data. But other technical capabilities must also be presentto support the data’s use. For example, initially, Earthgrains’ salespeople who work remotely couldn’t print any SAP outputsbecause they were not part of the company’s computer networkat headquarters. If not anticipated and resolved, the use oftransaction data can be inhibited by technical barriers.

Data Analysis and Presentation Tools. Earthgrains’ businessanalysts use Microsoft’s Access and Excel to analyze the dataand put it in a format that managers and sales people willunderstand. Experience has taught the customer service vicepresident that the way data is formatted and presented to dif-


Working Paper

Analytic Modeling. Although Earthgrains in general had madelimited efforts in this area, they had begun forecasting prod-uct demand to better plan their manufacturing capacity.

One-Time Process Analysis. The integration of SAP data alsosupported one-time process analyses. For example, Earthgrainsevaluated the cash disbursements going through its accountspayable process in its Bakery Products Division, thereby pro-viding support for a management decision to centralize allaccounts payable activities from its 44 bakeries.

Decision-Making Processes

Different analytic capabilities support different types of deci-sions and decision-making processes. Access to transactiondata, and the ability to interpret and analyze it, can change:(1) the types of decisions being made; (2) the confidencemanagement has in making certain ongoing decisions; and (3)even the location of some decisions within a business process.

New Types of Decisions. At Earthgrains, SAP data made it pos-sible to identify which customers and products were most andleast profitable. It is worth noting that while the data analysisrevealed unprofitable customers and products, the more impor-tant management decision was what to do with those resultsand how to do it. Managers may gain new insights by specu-lating about what decisions are not currently being made thatwould add considerable value, if they became practical.

Increasing Confidence in Decisions. Using transaction dataeffectively sometimes means making decisions that were notpossible before, but other times means changing the confidencelevel in decisions already being made. SAP data at Earthgrainsprovided new levels of confidence and support in making deci-sions about where to invest in promotions, where to invest inmanufacturing capacity given expected product demand, andwhere sales managers should be focusing their attention.

Changing Decision-Making Processes. Finally, the availabilityof transaction data also sometimes enables major changes indecision-making processes and, notably, where decisions arebeing made. At Earthgrains, two processes were changed in dis-tinctly different ways. The accounts payable process was cen-tralized and standardized so that all invoices were paid after 28days. In this case, payables decisions were not only shifted awayfrom local bakeries, but they were also embedded in a series ofdecision rules that virtually automated the process.

which the retailer purchased it, then the analytic insights fromthe transaction will be limited. Management must alwaysthink through in detail the types of data elements needed toaddress key decisions.

Complete data also means having adequate history to doanalysis. Earthgrains needed a one-to-two year history of aretailer’s inventory movement and stock outs before theycould help the retailer make inventory management decisions.In many cases, just having transaction data isn’t enough. Itmust exist for the right time period to be useful.

Timely Data Extracts. Transaction data must also be available ina timely fashion to support decision-making. At Earthgrains,sales data is pulled daily from SAP sales and distribution mod-ules. This allows the vice president of sales to identify problemsearly and take action before the end of the sales period. To beuseful, data extracts must be available for analysis on a frequen-cy that matches a useful monitoring and decision-making cycle.

Transformation Process

Analytic Processes

All of the contextual elements described above – strategy,organization, culture, skills, knowledge, technology, and data –combine to shape an organization’s capabilities for dataanalysis. At Earthgrains, these combined inputs provided themotivation and ability to create five types of analytic outputs.

Standard Reports. The vice president of sales looked at a dailyreport that showed what products had been sold the previousday complete with their volumes and gross margin, as well asthe total sales year-to-date.

Simple Analytic Outputs. Shortly after the sales and distribu-tion module of SAP was installed, Earthgrains’ managementbegan doing basic customer and product profitability analysis.This analytic approach is simple, although it does require theorganization to have previously adopted a disciplinedapproach to activity-based costing in order to be successful.

Complex Analytic Outputs. Over time, as management’sunderstanding of the system developed, they began posingmore complex queries to the analysts, e.g., How much do salesactually increase as a result of different types of promotions?


Working Paper

force in the Refrigerated Dough Division was told to stop ser-vicing them. At the same time, another initiative resulted inthe elimination of 20 percent of the division’s product linewhich analysis had shown to be unprofitable. In the first year,the division’s operating profit jumped over 50 percent.

New Processes. Earthgrains recognized that transaction datacould create opportunities for redesigning fundamental busi-ness processes that create entirely new sets of decisions. Forexample, Earthgrains is using data from grocery retailers’inventory control systems, combined with historical data onstock outs, to set up a vendor-managed inventory process thatsignificantly changes its relationships with retail grocers.Proactive use of transaction data may enable the redesign ofcore processes or it may simply improve decision-makingwithin the existing process.

Finally, SAP data made it possible to redesign the accountspayable process in the Bakery Products Division. The decisionto centralize and standardize this process increased the com-pany’s working capital by more than $40 million almostimmediately, and provided resources that could be used tosupport the division’s strategy to acquire additional bakeries.

Conclusion

Earthgrains’ experience of turning transaction data intoknowledge and results can provide other firms with valuableinsights into how to develop such a capability themselves.Here are some of the issues raised by Earthgrains that compa-nies should consider: � Management needs to assess the complexity and ambigui-

ty of problems requiring analytic support in order to decidehow to organizationally structure their analytic resources.More complex issues, requiring sophisticated modeling anddata analysis, are better served when analysts and decisionmakers are closely linked organizationally, as they are inEarthgrains, because of the intense communication andcollaboration required.

� Reward systems aligned with management’s objectives forcreating and acting on insights from the data are a criticalfactor for leveraging data-based decisions.

� Also critical is the development of technical capabilitiesfor the inter-organizational transfer of transaction data.

� Managers looking to increase the confidence in a decisionbeing made through the use of transaction data shouldask: what decisions still involve a considerable amount ofuncertainty, which if reduced would add significant value?

In the Refrigerated Dough Division, on the other hand, anattempt to implement a vendor-managed inventory processmeant taking decisions about ordering away from buyers inthe retail chains and embedding them in an inventory moni-toring system managed by Earthgrains. In both cases, chang-ing the location of decision-making was intended to improvethe effectiveness of the overall business process.

Outcomes

Converting transaction data into knowledge is only effective ifit produces business outcomes that improve financial perfor-mance. This usually happens as a result of new behaviors, newprograms, or redesigned processes.

Changed Behaviors. Historically, disputed invoices have been amajor problem for food manufacturers, since retailers oftendisagree with the prices they are charged on an invoice. Retail-ers always pay what they contend as being the price agreedupon, which is invariably lower than the price on the invoice.Because the cost of resolving disagreements is so high, manu-facturers have traditionally lived with the deductions. This hasresulted in losses of millions of dollars each year.

Having easy access to invoice data and a clear transactionhistory has enabled Earthgrains to improve processes support-ing price synchronization before the invoice is printed. Theability to minimize disputed invoices has reduced invoicedeductions by more than $4 million annually.

In order to encourage new behaviors, Earthgrains’ managersaligned incentive and compensation structures with the proposedbehaviors. Historically, the sales force in Earthgrains’ RefrigeratedDough unit had been rewarded only for the quantity of productssold. But after SAP was implemented the compensation systemwas changed to reward sales people 50 percent on sales volumeand 50 percent on gross profit. This significantly changed thebehavior of the sales force, who became much more interested inunderstanding the current and potential profitability of their cus-tomers. Reward systems that aren’t aligned with management’sobjectives for creating and acting on insights from the data are aserious impediment to leveraging this resource.

New Programs. With the ability to analyze customer prof-itability came a new initiative to change the product mix pur-chased by those customers identified as being unprofitable.About 180 of these low margin retailers were unwilling tochange their purchasing patterns. After 90 days, the sales


Working Paper

just a few customer visits. Thus, Harrah’s was able to pursue amicro-segmentation marketing strategy that dictated how muchthe company should spend for promotions on each customer toencourage either more frequent visits or longer stays at thecompany’s properties.

Developing Analytic Resources. Under the sponsorship of GaryLoveman, who had recently joined Harrah’s as COO, the companyalso built up a core team of highly experienced database mar-keters and analysts at corporate headquarters. In the process, theCOO replaced the entire corporate marketing team, which hedeemed not sophisticated enough to implement the new rela-tionship marketing strategy. Placing less value on casino industryexperience and more on quantitative skills and broader businessintelligence, Loveman sent a clear signal to the organization thatHarrah’s hiring criteria were changing. “Analytic groups need verysmart people if they are to attract other smart people,” he said.

Demonstrating Profitable Results. By mid-1999, the marketingteam had run several successful pilots, including one using theirnew segmentation models to target customers of the company’sTunica, Mississippi casino. That direct mail campaign not onlyproduced important new knowledge about Harrah’s customersand their responses to specific promotional offers, but profitsjumped from $29 per customer to $62 per customer for thosewho received the targeted promotions.

Barriers to Implementing Analytic Capabilities

Despite Harrah’s early success in the process of building data-to-knowledge capabilities, management still had to work throughseveral critical issues that threatened the long term success ofthe initiative.

Redesigning Local Marketing Roles More sophisticated analyt-ic capabilities being applied by the corporate marketing teammeant the roles for local marketers had to change. Marketingmanagers at the firm’s properties generally lacked an under-standing of how to use the data to implement a new relation-ship marketing strategy. Transforming existing roles meantconfronting considerable resistance among the local market-

When the explosion of newly legalized gaming jurisdictions inthe mid-1990s ground to a halt, Harrah’s management realizedthat growth could no longer come from the construction of newcasinos. Rather, growth would need to come from existing casi-nos and from an increase of cross-market visitations betweenthe company’s 18 properties. Harrah’s new strategy was to drivegrowth through customer loyalty and data-driven marketingoperations. The implementation of this strategy required the useof Harrah’s extensive transaction data amassed on the gamingbehaviors and resort preferences of existing customers. Harrah’sexperience provides important insights about the challengesfaced by organizations seeking to enhance their competitivecapabilities by using transaction data to support decision-making.

Building an Infrastructure

By mid-1999, Harrah’s had made considerable progress in trans-forming its marketing function to take advantage of the exten-sive transaction data collected on its customers. In addition to aclear strategic purpose for using the data, the company had sev-eral other building blocks in place:

Transaction Data Infrastructure and Front End Analytic Tools.An Informix-based online database had been implemented in1997 to capture daily transactions from the operational systemsin all of Harrah’s hotels and casinos. A year later, the informationtechnology department completed a data mart, known as Mar-keting Work Bench, which gave analysts and managers queryingand analytic capabilities, using software tools from BusinessObjects, Cognos, and SAS.

Generating Customer Data. The main source of customer datawas Harrah’s Total Gold program, a frequent flyer-like systemthat captured information on individual customer’s playingbehaviors, and rewarded repeat customers with complementaryaccommodations, meals, and show tickets. More than 60% of allslot play is captured on the Total Gold loyalty card, which tracksthe type of game an individual customer plays, the amount oftheir wagers, and the length of time they play. Because the casi-no’s take on each game is preset, Harrah’s can predict thepotential profitability of each customer for the company after


Working Paper

Harrah’s EntertainmentHarrah’s Entertainment is the world’s second largest gaming company with more than $2 billion in sales. About 80 percent

of the company’s revenues come from gambling at its casinos spread across 11 states.

tion, in large part because of their initiative’s strong sponsor-ship. “Now we have the talent, the strategy is defined, and thecapabilities are all in place,” said one team member. “But youcan have the best segmentation schemes and the brightestanalysts in the world, and it won’t get done or, more impor-tantly, ingrained in the organization unless you have the sup-port of senior management.”

ing managers who felt left out of the newly centralized mar-ket segmentation process. It also required developing moresophisticated understanding throughout Harrah’s of what itmeant to be a data-driven, customer-focused organization. “Alot of people don’t know how to tap the power of the data,”said one manager. “We find people trying to recreate the samereports available from the old casino system, instead of figur-ing out new questions to ask about the business.” Marketingmanagers thus were challenged to become more analyticaland creative, yet less technical. Ultimately, Harrah’s seniormanagement had to decide who could be retrained and whoshould be replaced.

Creating New Knowledge Transfer Processes. Over time thecorporate marketing team’s analysis developed important newinsights about customer behavior that could impact marketingdecisions. But most of this new knowledge was only in theheads of the corporate analysts, and the COO knew the com-pany had to create processes to transfer it to the operatingunits. Harrah’s long term success would depend on creating aninfrastructure to rapidly share new customer knowledge thatcould impact local marketing decisions. Harrah’s has takensteps towards this goal by initiating a series of marketingworkshops facilitated by the corporate marketing team, cus-tomizing analytic tools for each property, and encouragingconstant communication between the corporate marketingteam and the operating units.

Creating Standard Data Definitions. One of Harrah’s primarystrategic objectives was to increase revenues by increasing thenumber of cross-property visits. In order to measure theimpact of marketing programs seeking to do this, the companyneeded standard data definitions across its operating units.But analysts found these standard definitions were missing.For example, the definition of when a customer’s “trip” beganand ended varied from casino to casino, making it difficult tocompare the “average play per trip” across properties. In adecentralized operating environment like Harrah’s, seniormanagement still faced a special challenge in overcoming cul-tural barriers to standard data definitions. Without that stan-dardization the long term payoff from the new strategy wouldbe reduced.

Conclusion

Even though they faced significant challenges, the marketingteam pushed forward, confident in their ability to shift thefirm from a product-focused to a customer-focused organiza-


Working Paper

processed orders became part of an order history databasewhere address information was stored in a single field ratherthan in separate fields, and the individual customer’s namewas not retained. Moreover, once a product was shipped, theaddress in a particular record was never updated. Withoutupdated individual customer names, EMD could not pursue atrue one-to-one relationship marketing strategy.

EMD also lacked control over the marketing contact database,since it was owned by the sector’s sales organization. Accordingto Laura Hammond, manager of outbound marketing programs,“The field engineers [sales people] own the data, but they havenever signed up for keeping it clean,” she said. “As a result, theydon’t take time to implement data quality processes. There’s noalignment between responsibility for the data and the need touse it, even though relationship marketing depends on how goodyour data is.”

To identify their customers, EMD first extracted all customer sitesfrom the order processing data base that showed a purchase ofproducts from Test & Measurement divisions within the past fiveyears. These name and address extracts were sent to an outsidevendor, where they were cleaned and matched with records fromthe marketing contact database, which contained 700,000 namesand records of 15 million interactions with potential customers(e.g. requests for a catalogue, response to a promotion, etc.). Cus-tomer sites found on both files were loaded into a cross referencetable, which then gave marketing managers access to both sitelevel order history and contact level demographic and responseinformation.

Developing Effective Working Relationships

By late 1996, EMD marketing had begun working with JeanneMunoz, an analyst from the California-based American Mar-keting Organization, to rank and score all the customers iden-tified in the order history data. Since Munoz was based inNew Mexico, while the EMD marketing team was based northof Denver, initial team meetings were frequently held at theDenver airport. There, they sought to learn each other’s per-

The Electronic Measurements Division

In the mid-1990s, Hewlett Packard’s Electronic MeasurementsDivision (EMD), a producer of oscilloscopes, digital test equip-ment, and power supplies, was faced with increased competi-tion from smaller companies and shrinking profit margins in amarket that was no longer growing. Yet division managementset a goal of increasing sales and boosting its market sharedramatically by 2003. To pursue this strategy, EMD’s market-ing group had to find ways to increase revenues, while alsoreducing the division’s marketing costs. EMD could not rely onthe sales force for Hewlett-Packard’s (HP’s) Test & Measure-ment Organization (TMO)1, since they tended to favor sellingthe higher priced products from other divisions within thesector. EMD marketing had to find other channels such asadvertising, catalogue sales, and direct mail to promote thedivision’s 50-plus low priced products. This case is about howEMD effectively utilized these channels by developing thecapabilities of using order history data to segment and targetits potential customers in a global marketplace.

Building a Database for Marketing Analysis

In early 1996, EMD’s sales efforts were hampered by not knowingwhere it could most effectively invest its marketing resources toboost sales for the division. The initial challenge faced by EMD’smarketing team (based in the division’s Loveland, Colorado, facili-ty) was identifying specifically who its existing customers were.To do that, the marketers needed a customer file they could ana-lyze. However, this was challenging because EMD lacked controlover the order entry and the marketing contact databases.

HP’s order processing database was being managed centrally bythe TMO’s Americas’ Marketing Organization in Santa Clara, Cali-fornia. Moreover, the order processing system had not beendesigned with marketing in mind. Since there was no unique cus-tomer number attached to sales orders, it was all but impossiblefor HP to develop a unified view of how its customers werebeing served by the firm’s many divisions. In addition,


Working Paper

Hewlett-PackardFounded in 1939, Hewlett-Packard is one of the world’s largest computer companies. It is the foremost producer of test and

measurement instruments, but also offers networking products, medical electronic equipment, instruments and systems for

chemical analysis, handheld calculators, and electronic components. Headquartered in Palo Alto, California, Hewlett-Packard

has 42.9 billion in revenues (1997) and employs more than 120,000 people.

to 100,000 names. The names were determined by the mostrecent catalogue subscriptions and product inquiries.

In November 1996, EMD began using its three higher potentialclusters of existing customers when developing its cataloguemailing list. Unfortunately, it was impossible to track the bot-tom line impact of specific catalogue mailings because nounique customer numbers were attached to the sales ordersthat catalogue customers placed through HP’s Colorado-basedcall center. But senior management accepted other, more indi-rect evidence that using order history data to profile cus-tomers added value to the marketing efforts. For example,EMD’s revenues always trended up during the three monthsafter a catalogue was mailed. In one test, the division gained$1.5 million in revenues in the three months after mailing150,000 catalogues outside of its normal marketing cycle.And, when a catalogue mailing in Europe was unexpectedlydelayed three months, EMD’s revenues there dropped signifi-cantly in those months as compared to the previous year.Although such evidence helped garner senior managementsupport within the division, it was generally recognized thatdocumenting bottom line business impacts of specific cata-logue mailings would be necessary to gain broader politicalsupport for this type of data-to-knowledge initiative.

Once the customer profiles became the basis of EMD’s out-bound marketing strategy, Hammond’s team could begin track-ing the migration of specific customers through different clus-ters and evaluate the impact of various marketing initiatives. Inaddition, marketing programs were created to move customersinto higher value clusters. A cross-sell mailing in late 1997, forexample, targeted New Stars and Middle of the Road Folks witha special promotion on oscilloscopes in an attempt to movethem to Cash Cow status. And a special reactivation mailing ofthe Basic Instruments catalogue sent to customers in defectorclusters enabled Hammond’s team both to cut 10,000 namesfrom its overall mailing list and identify a small percentage oflapsed customers worth pursing more actively.

Learning Over Time

Analyzing responses to catalogue mailings and other promo-tions has helped develop EMD’s understanding of which cus-tomers to invest in. “Our thinking about clusters has matured,”said Hammond. “Looking for migration patterns between clus-ters has given us a much better understanding of customeracquisition and retention rates, and how the value of a cus-tomer almost doubles if we can get them to purchase in the

spectives on the problems at hand. Munoz explains that in thebeginning, she “didn’t know their business, and wasn’t surewhat they wanted.” John Shields, a program manager respon-sible for catalogue marketing noted “it could take an hour toget a simple question answered.”

Shields notes, however, that “Over time, we developed a com-mon language, and our knowledge evened out. It was verycollaborative.” The marketing team learned how to ask ques-tions so that the analyst could interpret them in the contextof the capabilities of the new data file. “To do that,” saidHammond, “we had to understand the analyst’s language, aswell as her mental models of the different data files and howto manipulate them together.”

Profiling Customers

Working with Munoz, the marketing team identified five mar-ket clusters, based on variables such as recency of purchases,number of products purchased, dollar volume, and date of firstorder. The higher potential clusters included:� Cash Cows: customers with a long history of purchasing

EMD products. They had purchased frequently over time,had made a purchase within the last 18 months, and werethe biggest source of revenue.

� New Stars: new EMD customers whose dollar volumesranked at the same level as Cash Cows.

� Middle of the Road Folks: customers whose volumes didnot put them in the more profitable clusters, but whoseactivity was enough so they didn’t fall into any of the“defector” clusters, that is, customers who had not pur-chased for a number of years.

This market segmentation immediately helped John Shields inmaking decisions about where to mail EMD’s semi-annual BasicInstruments catalogue. Historically, the 170,000 names qualify-ing for the division’s mailing list came from catalogue sub-scribers, recent customers, product inquiries, and peopleresponding to promotions. But the marketing department hadno idea which of these names were potentially the most prof-itable. The subscriber list was suspect, for example, becauseShields felt the sales force tended to sign up their customers forall HP publications, like the catalogue, regardless of the cus-tomer’s interest. The inability to differentiate between high andlow potential customers created an unrealistically large pool forany marketing promotion. But because each catalogue costs$2.00 to mail, budget constraints always limited mailing volume


Working Paper

mond’s team wanted to begin accessing the data directly.Although the marketers had developed a highly collaborativerelationship with analyst Jeanne Munoz, she was the only oneaccessing and analyzing the data in EMD’s evolving marketingsystem. Increasingly, this created a bottleneck for marketerswho needed analytic outputs to support their efforts. As aresult, marketing managers were not learning enough aboutthe data available and how it could be applied to design andimplement more cost-effective marketing programs.

Thus, in late 1998, a pilot program was launched that imple-mented a web-based analytic data mining tool from E.piphanyfor marketing managers.2 E.piphany’s software tools allowedEMD’s marketers to use order history data to make decisionsin hours or days as compared to the weeks of interacting withMunoz that would have been required previously. E.piphany’ssoftware, for example, could support decisions such as:”Which products have complementary relationships, given ananalysis of customer purchasing behaviors?” or “Which cus-tomers should be offered cross-sell products?”

Providing EMD’s marketers with direct access to customerdata had several important impacts. First, it changed howmanagers work with the data. Providing direct access allowsthem to “rehearse” different questions, exploring the databefore calling the analyst to discuss things. “Before we hadE.piphany,” said John Shields, “I would call Jeanne and say,‘The catalogue is on the press now. I need a mailing list.’ Butnow, before the catalogue goes to press, I look at the trendingdata and see what’s going on with the various clusters andmarket segments. I try to determine where the best marketingopportunities are.”

Providing direct access through E.piphany’s software has alsochanged the analyst’s activities. “It takes me out of thestraight forward questions that have simple answers, but thattake time to do. That gives me more time to deal with realanalysis, using statistical programs like SPSS.”

Changing how decision-makers and analysts work with the dataalso inevitably changes how they interact. Munoz recalled a timerecently when she arrived in Loveland for a meeting about a newcustomer profiling report. “I put the report out on the table,” shesaid, “and told them ‘I’m here to answer your questions.’ I real-ized that was a big change because I no longer had to presentthe findings to the team, or teach them how to read the report.Instead, we immediately started having a discussion about whatit meant for our business.”

second year.” Hammond’s team also came to believe that Mid-dle of the Road Folks was the cluster that held the greatestpotential for increased revenues, as opposed to the segmentsthat they normally targeted. “That’s a real culture shift for us,”she said, “because traditional HP training says follow themoney in your installed base and go after big accounts.”

Because of changes in the customer base, and to ensure thatcustomer profiles were up-to-date, the matching process link-ing order history and marketing contact data was rerun everysix months. Customers were again rescored and assigned toclusters. One surprise was that many of the individual cus-tomer sites that represented a specific plant or office locationwere disappearing from EMD’s data file over time. “Finally, wefigured out that things were going on in the source [orderprocessing] system that we didn’t know about,” explained theanalyst. Munoz determined that the order history databasehad been programmed to find duplicate records for customersites, and delete them, even though they contained data criti-cal for EMD’s evolving customer profiles. “We learned that it iscritical to know the details of all the systems upstream fromyou,” said Hammond, reflecting on the costly surprise ofassuming that their core transaction systems were static.

Data Quality Project

As EMD came to rely on order processing and marketing con-tact data to implement its relationship marketing programs,new questions arose about the quality and accuracy of thedata being used. For example, the order processing system hadnot been designed to produce data that would serve as asource for direct mail lists. Thus, a data quality project wasundertaken to evaluate the data being used and to identifyany processes related to data quality that EMD could impact.

The three-month study showed that data in the marketingcontact database was about 90% accurate, which was higherthan expected. But it also revealed that about 20 percent ofthe database changed each year when people changed posi-tions or left their company. As a result, Hammond’s team real-ized they had to give greater weight to the “recency of pur-chase” criteria when selecting the best, most reachable names.

Broadening Access to the Data

By early 1998, two years after EMD had begun using clusteranalysis to invest its marketing budget more effectively, Ham-


Working Paper

Working Paper

John Shields, who manages the catalogue marketing program,had a related observation. Reflecting on the changing role ofmarketers, he said: “To be successful, you have to learn thelingo and be willing to talk about what the data is saying.When we first started, Jeanne would try to explain somethingto me, and I’d be confused and say, ‘You’re blowing my mind,Jeanne.’ But during a recent meeting I said to her, ‘I neverthought I’d be talking with you about all this data stuff.’ Now,instead of two hour conversations with Jeanne, we do it in 15minutes, and we get a better decision. It’s much more effi-cient for me because I understand the database.”

Finally, the evolving relationship between EMD’s marketingteam and its database analyst has produced more effectivedecision-makers got direct access to the data, Munoz was sobusy handling routine queries that she didn’t have time forrepeated discussions about a particular mailing list she wassupposed to pull. But more frequent interactions now enablemarketers to continually refine the criteria used to developmailing lists, which result in more productive lists. And Shieldssays he is able to make better decisions when deciding whereto mail the Basic Instruments catalogue because Munoz canprovide him with richer insights about customers in differentclusters. “This is where it all pays off,” said Hammond. “We’vegotten beyond just understanding each other’s language andthe context of the problem to a true partnership where theanalyst is actually engaged in the business conversation.”

Evolving Data-to-Results Capabilities

Reflecting back on her experiences in using transaction data topursue a relationship marketing strategy, Laura Hammond said:“It’s not a linear process going from data to analysis to deci-sion-making. It evolves very organically. We have great datanow, but it has taken us a long time to get there. And we needto assume we’re always going to have to work on the database.Over time, you think you know more about your customers andthe market, but there’s also a lot more ambiguity created by theinsights you get. The company’s culture has made a big differ-ence in this because we’re basically a really curious organiza-tion. They created a risk-free environment for us where it wasokay to learn and fail, as long as we applied the learning.”

One of Hammond’s colleagues added a final point in explain-ing the success of EMD’s relationship marketing initiativewhen she said, “It really helps to have someone who is a data


advocate, and that describes Laura. She’s passionate aboutusing the data.”

Conclusion

EMD’s use of data is seen as a model for other Hewlett-Packard businesses to follow. Six programs are currently usingthe E.piphany tool set, and initiatives are underway to teachusers of the tool set about the data. EMD’s experience inusing transaction data to generate business value may be lim-ited in scope, but valuable in its implications for other compa-nies seeking to transform data into knowledge and results. Thefollowing are some of the key insights extracted from ourobservations at EMD that may be leveraged by any companyseeking to transform data into knowledge and results:� There is a strong need for coordination and alignment

between those who feed and maintain the database andthose who use it.

� Expect changes in the core transaction systems and anticipate their impacts on data quality.

� Analysts and decision-makers will work together moreeffectively if they have time to get to know each other.

� Data-to-knowledge capabilities will have much more credibility if their benefits can be measured.

� Effectively using data for decision-making will help solveproblems, but it will also create many new questions andopportunities.

� Data-to-knowledge capabilities must be developed withinthe existing cultural and structural context, unless topmanagement is willing to drive the transformationalchange to create new patterns of coordination and communication.

Notes1 At the time this case was developed, EMD was still one of 22 divisions in

the Test & Measurement Organization of Hewlett-Packard. It had recently

been announced, however, that TMO, along with the chemical analysis

and medical businesses and a computing and imaging company, would

form the independent company Agilent Technologies as of November 1,

1999.

2 See www.epiphany.com for more information.business outcomes. For

example, before

he believed that Kraft could achieve a competitive advantagethrough fact-based marketing innovation. Finally, he sought toremake Kraft’s trade marketing execution capabilities byempowering the sales force with superior information, analyticcapabilities, and results-oriented recommendations.

At the end of a lengthy evaluation process, a particular infor-mation supplier was chosen. The supplier’s top managementhas enthusiastically embraced Kraft’s vision, describing it as amodel for teaming with their customers. But getting to thispoint took a great deal of work on the part of both organiza-tions. Ten senior Kraft people spent six months hammering outthe first deal. It took a long time for the two organizations todevelop a shared understanding and vision for workingtogether. One key step to achieving the seamless integrationof the companies was the creation of an onsite “mirror” of theinformation supplier dedicated to servicing Kraft.

Kraft is now the supplier’s biggest “ship to” customer in theUS. The relationship between the companies appears to bequite strong and focused on results. Both companies mutuallydeveloped and live by an annual and five year plan. There isalso a clear process for communications, review, and a mutu-ally agreed upon reward structure. The plan has interim mile-stones, metrics, and objectives for the alliance and for eachcompany. The alliance has the attention of the CEO’s of bothcompanies, who regularly monitor progress against agreedupon metrics.

Key Applications of Data

Brand management is a major source of competitive advan-tage for Kraft. Kraft believes that its ability to analyze andexploit category data is central to marketplace success. Kraftuses data in several different applications.

Production planning is a continuing challenge for Kraft, sincetheir products are sent to warehouses and typically have along shelf life. QUEST is a goal setting application that inte-

Data is like oxygen at Kraft. Our executives don’t questionthe need for it, because we use it daily.

— Eric W. Leininger, Kraft Foods Vice President, Marketing Information Services

Kraft measures its success by its progress towards its goal of“undisputed industry leadership.” Bob Eckert, Kraft’s CEO, hasasked his senior management team to “work with employees todeliver the numbers; live the values of focus, innovation andpassion; and lead through teamwork.” To achieve this goal,Kraft is pursuing several innovative strategies. One of thesestrategies is the transformation of its vast stores of scan andother transaction data into actionable business knowledge.

Kraft has several initiatives aimed at this transformation. Theseinitiatives are aimed at better service and better profitability forKraft as well as for key information providers and customers.The win-win value proposition is at the heart of Kraft’s successwith these tools and the analytical and decision-makingprocesses which enable them. Our interviews focused upon afew examples of partnering to use transaction data:� Marketing Effectiveness (partnering with a key information

supplier)� Category Captaincy and the 3 step Category Builder

(partnering with customers)

Marketing Effectiveness

A key initiative in Kraft’s data-to-knowledge process is itspartnership with a leading market information supplier. Afteryears of frustration with various information vendors, Kraftdecided to take a different approach – to partner with onevendor and commit to making it a win-win relationship. Kraftasked its vendors to propose to provide “everything for every-one and make it available right away…”

CEO Bob Eckert had several objectives for this partnership. First,he envisioned an “information utility” where everyone in Krafthad access to current, decision-oriented information. Secondly,


Working Paper

Kraft FoodsKraft Foods is the largest food company in North America. This $17.3 billion dollar subsidiary of Philip Morris produces one

out of ten products purchased from U.S. supermarkets. Kraft’s well known brands include Kool-Aid, Oscar Mayer, Maxwell

House, Post cereals, and, of course, Kraft cheese products. Kraft’s 37,000+ employees work at more than 50 manufacturing

facilities and 230 distribution centers throughout the US with 100-plus facilities in 35 other countries.

As a result of the growing trend to name a category captain,the role of the sales force has changed dramatically. The salesforce, already subject to long hours and high turnover, under-stood the importance of becoming category captains. Yet theywere concerned about the additional work it would entail. Man-agement recognized the need to build the analytic capabilitiesof the sales team. The sales force required a greater under-standing of issues such as assortment, space planning, and pro-motion strategies. Training the sales force to shift to a moreconsultative sales role was also vitally important to success.

3-Step Category Builder

The 3-Step category builder is a critical component of Kraft’scategory management strategy. It enables Kraft sales repre-sentatives to effectively partner with customers (e.g.: an indi-vidual grocery location, such as Kroger or Safeway). The cate-gory management process enables the sales representative toquickly collect the necessary data and present sound recom-mendations to the store manager.

Nick Feimer, a Kraft executive responsible for deploying thisprogram to the sales force, states “the reason why we havedone so well is that we focus on what is truly critical.” Kraftbelieves that they have identified the most important dataelements which drive sales. The 80-20 rule is a crucial guidingprinciple for this process, since Kraft and its customers areswimming in data and struggling with the question of “howdo I unleash the power of information?” Mark Froseth, a KraftVice President in Sales and Customer Service, perhaps said itbest, “the amount of information continues to multiply so thechallenge is to choose the right tools to turn the data intouseable knowledge.” Kraft sales representatives collect lessthan 20% of the data used by others for performing similaranalyses, with superior results. Stores generally realize a mar-ginal sales growth of 3-4% vs. a control group.

Kraft sales representatives are assisted in their work by ananalytic group deployed out into regional sales offices. Ana-lysts are able to perform the sophisticated analyses quicklyand efficiently, in a fraction of the time required by competi-tors (25 hours vs. 500 hours). The conclusions are delivered tothe sales representative in a “standardized business case tem-plate.” While the analysis itself is quite complicated, theresults are put into a Powerpoint presentation. This presenta-tion provides a clear analysis and a set of compelling businessrecommendations. Supported facts are easily accessible aswell. As new insights into category management are devel-

grates internal inventory and forecasts together with data tocreate a production planning forecast. This system creates a“true shipments” forecast, which is more reliable than invoic-ing systems or shipment data alone.

Promotion/pricing effectiveness models link “causal” data tointernal transaction data, such as divisional P&L, inventory fore-casting, and weather data. The objective is to mine new insightswhich can influence buyer behavior. For example, through thissystem they learned that Mayonnaise sales are affected by theweather, but BBQ sales are more affected by seasonality andpromotions. The pricing models also are extensively used bybrand managers to make tactical promotion decisions. Kraft isseeking to standardize their diverse product managers aroundsimilar tools and applications. The planner workbench isdesigned to reduce the 40-60% of time spent by planners ondata extraction and manipulation. It is also intended to providea consistent and robust analytic capability (based upon SQLServer, Visual Basic and SAS) which can replace a variety of“homegrown” applications ranging from Fortran code to 123macros which perform regression.

Category management is viewed by both organizations as amajor achievement. Through an in depth analysis of categoryinformation, Kraft sales representatives can work effectivelywith individual grocers to identify problems and opportunities,and to make sound recommendations supported by information.

Category Captaincy

Jim Kinney, CIO of Kraft, has experienced a “revolution in Con-sumer Packaged Goods over the last 4-5 years.” Grocery storesare frequently “outsourcing” each product category to a cate-gory captain. Store managers choose each category captainamong the major CPG companies supplying the category. Cat-egory captains perform category analysis and advise storemanagers regarding product placement, inventory, and vendormix for the entire category, not just their own products. Whilecategory captains are generally large companies with a Top 3market share, they may not be the market leader. Rather, gro-cers choose the vendor who demonstrates the best capabilityto optimize the bottom line value of the category. Kraft andits competitors place a very high value on earning categorycaptaincies. Establishing a productive partnering relationship,together with the ability to demonstrate superior analyticcapabilities, is the key to achieving category captaincy.


Working Paper

zation feels with using numerical data to make decisions isfiltering down to the field sales force with the growing use ofcategory management data and reports.

Until about 4 years ago, Kraft was a highly decentralizedorganization. As a result, data and organizations are oftenquite dissimilar, making comparisons difficult. It is often nec-essary to work through several individuals in each productline, persuading them to share their information with others.For example, Finance has been unwilling to share some datawith the business units “because they won’t understand it.”

Skills and Knowledge Context

Many Kraft executives viewed business capabilities as somethingwhich must evolve over time. People are able to truly understandthings only after going through a process. They must have time tograsp the fundamentals before they can progress to the next level.Kraft has managed this by rolling out new reports and processes ina staged manner. They believe it is better to start with only themost essential, basic information and add complexity over time.This staged roll-out also facilitates a fire-aim-fire methodology, aprocess allowing quick deployment with refinement of tools andmethodologies over time.

Often, initiatives will first provide data access to a group ofusers. Sometimes, these interim solutions are “throw aways”which are replaced as users become familiar with the data. Asthey gain insight into the data and how it can be applied,there is generally an effort to cleanse data and make it moreuseful. This next stage yields additional insights, and begins toallow users to more clearly understand the strategic levers fortheir business. These efforts lead to the development of acommon understanding of how Kraft creates results and addsvalue to the business. In turn, more variables are identified(often external to Kraft), which need to be incorporated. This“external data,” once integrated into the business model, pro-vides additional insights and leads to the evolution of theorganization’s “mental model” of the business. As the businessdrivers become broadly understood, it is possible to encapsu-late that knowledge into broader business processes andapplications. Often this will involve teaming with an externalvendor, such as a consultant or information provider, to createnew applications.

There are some distinct organizational roles at Kraft that helpthe data-to-knowledge process. The program manager isresponsible for deploying the new business capability. For exam-ple, in the category management process, program managers

oped, the business case template is revised so that it onlyfocuses on the information that Kraft’s customers need tobetter manage their categories and business.

To Kraft, the effort to partner more effectively with their cus-tomers is clearly worth it. In a recent Canonndale Associatesstudy of retailers, Kraft rose from 14th to the top 2 in havingthe best customer insight and the highest ability to exploitcategory information.1 They monitor and have achieved sig-nificant improvements in: I) volume by chain, II) new productsales, and III) new category captain wins.

Additional Points Using the Data-to-KnowledgeFramework 2

Strategic Context

Kraft executives pride themselves on superior execution oftheir marketing strategy. Initiatives to transform transactiondata into actionable knowledge and results are central toKraft’s competitiveness. These efforts are based upon manage-ment’s belief that win-win relationships with its customerswill be by increasing data-based decision-making.

Kraft’s sales force is moving away from a relationship-basedsales model. Retailers’ razor thin profits and the vast array ofproduct choices require a more analytic approach to bottom-line oriented decisions.

Culture / Organizational Context

From the top down, Kraft’s key metrics are product sales andyear-over-year volume growth. Almost everyone’s bonuses aretied more or less directly to sales. The company lives and diesby its market share and unit-volume growth.

Kraft has a strong analytic culture. CEO Bob Eckert and theleadership team are firm believers in data-based decision-making. Senior management takes a metric driven approach togauging the success of key initiatives such as the informationprovider alliance. CEOs of both companies personally approvedthe interim and long term metrics for the alliance team.

Similarly, there is a long history of using marketing data tomake decisions about advertising and promotion. The use ofinformation provider, scan-based, and other transaction datahas gained greater acceptance over the last four years. Itappears that some of the comfort that the top of the organi-


Working Paper

of historical sales data by product, customer, and date. APower-Builder front-end GUI is available for accessing thisinformation. The customer service systems store data in awarehouse built by Fast Tech on an Oracle platform. Thisdata warehouse stores sales information weekly by productand by customer store. Both data warehouses are movingtowards a corporate Intranet environment.

� Data Marts/OLAP/Decision Support: Supported toolsinclude eSSbase, SAS, spreadsheets (such as 123, Excel),and specialized marketing applications (both custom andpurchased).

While Kraft has a relatively sophisticated technical environ-ment, they still deal with issues of technological complexity.They are in the process of moving to more of an Intranet envi-ronment and are seeking to replace “homegrown” analyticapplications (such as the regression analysis model basedupon 123) with a more consistent analytic tool set.

Analytic and Decision-Making Processes

The category management process is very well defined andbroadly understood. A Kraft sales representative partners witha buyer at a particular retail outlet to gather basic informa-tion through a structured interview tool (an Excel template).The Kraft sales representative then takes this information backto his regional analyst. The analyst and the sales representa-tive couple this information with relevant panel, census, orother third party data. The analyst then performs a structuredseries of analyses using this data and store level transactiondata. The results of these analyses are put into a standardbusiness case template (a Powerpoint presentation) whichcontains a situation analysis, recommendations, and support-ing information. Using this information, the representativeuses the data to help customers sell to the end-consumer. Aclear understanding of what drives sales and profitability arean essential component of this process. This understandingallows Kraft to reduce analytic time (and therefore costs) aswell as produce a superior outcome.

Outcomes

The 3-Step Category Builder is a great example of the trans-formation of data into knowledge and results. According toKraft, it has helped customers achieve 2 to 5 percentage pointhigher growth in managed categories as compared to controlgroups. Although this growth rate may seem low when com-pared to other industries, in the flat growth retail grocery

worked with the sales force to ensure that the standard reportsand processes were useful and being adopted correctly.

A data analyst role exists in many different parts of the orga-nization. In category management, for example, it was decidedthat the role of the analyst was to leverage the sales staff andto utilize specialized skills effectively. There are two groups ofthese analysts, one at headquarters and the other deployedinto regional sales offices. The headquarters group (~35) didmore of the ad-hoc analysis and the development of the tem-plates used for standard reporting while the regional analysts(~180) partnered with individual sales representatives to per-form various analyses including the category managementanalysis. Kraft is challenged to find and retain analysts whichhave the requisite technical skills and understand the busi-ness. Competitors have begun to target Kraft analysts forrecruitment. Kraft has had to look to non-traditional sourcesto locate the necessary skills.

Data Context

Kraft has long used data for marketing purposes but over thelast four years has integrated the use of data into the salesfunction as well. Marketing historically relied on panel, demo-graphic, and transaction data to better understand end-con-sumers. With the advent of category management, transactiondata is increasingly being used in partnership with retailers tooptimize the retailers’ shelf space. The retailers’ scan data ispaired with panel, demographic, and other transaction data todetermine opportunities and to improve retailer financial returns.

Since most Kraft products are distributed to retail grocerychains’ warehouses, Kraft does not have data on where its prod-ucts are in the pipeline until they are sold. This is considered anissue for Kraft since visibility essentially disappears after theproduct has been shipped to the customer. Visibility does notreturn until the product is scanned at the check-out line. Kraftmust perform back-end analysis to ensure that product sold to acustomer on discount for a promotion in Des Moines doesn’t endup being sold for a substantial mark-up in New York.

Technology Context

Like many diversified organizations, Kraft employs a broadrange of analytic technologies to manage its business. Tech-nologies include: � Data warehouse: The most significant transaction systems

for the marketing organization are fed into a Redbrick-based data warehouse that is populated with three years


Working Paper

environment, this impact is phenomenal. The impact on Kraft’sbusiness can also be seen through its category managementor “category captaincy” win ratio. Kraft is not the only manu-facturer that would like to manage retailer categories. It isbeing asked with increasing frequency to be the category cap-tain even in categories where it is not number one in sales.Kraft’s data-driven transformation process has demonstrableresults for its customers.

Notes1 “Kraft Makes a Better Case,” Advertising Age, November 30, 1998.

2 The data-to-knowledge framework is presented in “Data to Knowledge to

Results: Building an Analytic Capability,” by Thomas H. Davenport, Jeanne

G. Harris, David De Long, and Al Jacobson.


Working Paper

2. Market Network: This initiative is aimed at taking theguess work and personal biases out of branch closingdecisions by using a modeling process that evaluates eachbranch’s current and long-term profitability. The modelingprocess uses a variety of data, including transaction datafrom the branch, business commerce trends in the area,department of transportation usage data, drive-aroundinformation, geo-mapping, and staffing configurations.Transaction data is aggregated at the branch level.

3. URL (Unprofitable Customer Segments): The mirroropposite of PRO, URL focuses on the bottom three cus-tomer deciles (in terms of non-profitability). The modelingand decision-support system seeks to identify salesstrategies for converting more of these households intolong-term profitable customers. URL is the most recent ofthe three initiatives launched by Wachovia.

Wachovia’s Supportive Culture and Direction

The ease with which Wachovia turned to data modeling toadvance the strategic direction on profitability was groundedin a highly supportive culture and prior experience with thesekinds of techniques. During the 1980s for example, whilemany peer banks’ credit ratings were going south and largeportions of competitors’ loan portfolios were being written off,Wachovia was envied for the quality of it holdings, strict lend-ing practices, and loan review policies. Part of this was theresult of good, solid policies and practices. But these policiesand practices were strongly supported, indeed in someinstances driven by, long established database managementinformation systems. Further, there was active encouragementfrom managers to use data in the initial and ongoing monitor-ing of risk management.

This and other early-on successes and experiences servedWachovia well when they decided to build a stronger cus-tomer-centric process utilizing data. The original focus was on

Strategic Setting

Throughout the 90’s, Wachovia generally played the “bigger isbetter” scale game, as did most banks at the time. The ruleswere pretty simple: profitability was assumed to be tantamountto acquiring more customers, more branches, more lines ofbusiness, more credit cards, etc. However, as industry consolida-tion occurred at an increasingly rapid rate, the stakes in thegame got higher and higher. Wachovia’s two nearest neighbors[First Union and NationsBank (later BankAmerica)] were partic-ularly aggressive. Wachovia instead adopted a more qualityniche role, although it still had a very large presence in thesoutheast. The code word at Wachovia became profitability. Thismeant quality customers, quality products and services, andquality growth potential. Businesses that did not fit this model(e.g., student loans) were either spun-off or sold.

Data-to-Knowledge to Results Initiatives

For Wachovia, this refocus on profitability naturally extendedto using data-driven models to enhance the strategic initia-tive. The first applications were on the retail side of the busi-ness. Three somewhat independent but related projects werelaunched:

1. PRO-Strategy: PRO is aimed at the bank’s top 25% mostprofitable consumer households who are current cus-tomers of Wachovia. PRO uses customer level transactiondata (exclusive of consumer credit products such as creditcards and mortgages) extracted from the bank’s main-frame applications to generate new product leads. Theseleads are electronically linked to personal financial advi-sors (assigned to the top tier accounts), bankers (regionaland branch level) and call service representatives. Theessence of the analytic process is a consumer scoringtechnique that calculates and then rank orders the proba-bility of customer purchase for each of ten major productgroupings.


Working Paper

Wachovia BankWachovia Corporation is a leading interstate bank holding company with dual headquarters in Atlanta and Winston-Salem,

North Carolina. In terms of the standard bank measuring rod, Wachovia ranked 16th overall in asset holdings ($64.1 billion

as of the beginning of 1999). Total market capitalization is $17.8 billion. Wachovia has more than 20,900 employees

servicing over 750 offices, principally located in the southeastern US. Wachovia offers a comprehensive set of products and

services serving the retail, corporate and institutional marketplaces.

inside field sales force. Some of the southern employeesreferred to this as “getting the dog to hunt.”

Cooperatives largely achieved by the development of a compre-hensive sales communication process that was managed by asenior vice president reporting directly to both the president ofthe retail banks (organized by state) and the executive vicepresident in charge of retail banking (i.e., the lines of business).The specific techniques used to enlist this support tended tovary from case to case, but the principal types were as follows:

� Implementation plan and roll-out: “We had a verydetailed implementation program. Very specific thingswere noted for each day. The market representative wasresponsible for the overall implementation. It was impor-tant that the implementation program convey the messagethat it was the market or field staff that had really gener-ated the ideas underlying the analytics. We (analytic, headoffice types) tried to stay in the background. We built avery strong communication program that really definedthe entire sales approach.”

� Presentations / Education: “We worked very hard to makethe case a compelling one. We had a presentation thatwent out into the field and said: ‘Here’s the problem,here’s the analysis, here’s the findings, here are theoptions, and here is what we are going to do’.”

� Training: “Everyone had to go through the same training(for PRO). It was this idea of ‘sweat equity’ that really cre-ated ownership. This was the idea of using PRO – to gothrough the training to gain ownership. This implementa-tion period took about 120 days.”

� Performance measurement: “We collect and maintainmonthly performance outcome numbers on our modelsand review how we are doing in terms of profit behavior.The results are used for senior and product managementreporting but for improving the model itself.“

� Weekly Conference calls: “We scheduled weekly confer-ence calls with the field sales staff to hear from themdirectly what was working and not working. These confer-ences were extremely valuable and in a couple ofinstances led to changes the field staff wanted to incorpo-rate in the delivery system.”

� Joint Data Gathering and Model-building: “We spenttwo days in a van going around gathering this data. Theycouldn’t argue that we did not understand their markets. Aprocess of bonding occurred precisely as a result of thismapping process. So they got to know us personally and

2.5 million retail households. As Lynn Brown, Executive VicePresident of retail banking noted,

“This (customer profitability models and data warehouse)is not something new that we have come to. Wachovia hasalways used information more aggressively than ourcompetitors. It is the way that we keep our customer at thecenterpiece of our relationships. Wachovia’s competitiveposition depends upon our ability to use information fasterand smarter than our competition.”

For Wachovia, using data and analytic models for decision-making was not a passing fad to be embraced and later dis-carded; rather, it was a way of doing business. In effect it waspart of their culture. Like other companies in our study,Wachovia experienced their share of data quality issues: How-ever, one of the differences with Wachovia was that no onequestioned the importance of having quality data. Perhapsmore importantly, no one questioned their own individualresponsibility in creating and maintaining quality data. Addi-tionally, because transforming data into knowledge and thenresults was viewed as part of Wachovia’s core strategic posi-tion, there was virtual unanimity in support of the entireeffort. Fred Koehl, head of the analytic group, noted this cor-porate imperative when he commented as follows:

“It was a blessed project from the outset. Managementsupported it.. Everyone was on the same playing field. Thegeneral attitude was ‘guys we have to get this done’.”

This kind of senior management support often spells the dif-ference between success and failure in data-to-knowledge ini-tiatives. The road from pure data to useful outcomes is farfrom an easy or straight-forward process. Besides the obviousand considerable issues involved in translating business objec-tives and understanding into analytical models, there are avariety of data quality, technical, and human resource issuesto contend with. Wachovia was indeed “blessed” to have thiskind of support, and it made all the difference in the world interms of the successes they achieved.

Sales Management Process

Wachovia was “blessed” in another very important way. Inaddition to support from senior management, there was aclear recognition at the outset of the PRO and Marketing Net-works that management’s ability to realize outcomes depend-ed in large part on the acceptability and cooperation from the


Working Paper

this established an important part of the communicationprocess.”

“We hear directly what leads have worked and which oneshaven’t. Comments have even led to changes in the modeland the delivery system.”

Outcomes

Direct measures of the outcome value of Wachovia’s data-to-knowledge initiatives were not available, except for one quali-tative note regarding the fact that the branch closing processusing Market Networks was far more efficient. Wachovia hasmoved to create a unit dedicated to monitoring performanceoutcomes of each of the projects, and these results are rou-tinely reported to senior management as well as used toimprove next iteration model results. Clearly, if management’sintentions to replicate this data/analytic effort in other areasof the bank is any indication, then there is every reason tobelieve that the early returns have been extremely favorable.


Working Paper

� Develop sponsorship� Change beliefs and behaviors� Create the technical and data infrastructure� Manage the implementation of multiple projects� Change roles and responsibilities

Let’s explore each of these in more detail:

Develop Sponsorship. The vice president believed that if hewas going to get executive support for an enterprise-wideCRM initiative he had to link it directly to financial impacts,and even earnings per share. As a result, his change manage-ment approach included a major program of educating seniormanagement about the importance of CRM to the firm’s over-all business strategy. The vice president identified 16 key deci-sion-makers and delivered a targeted message to each one,showing how CRM could help them with their issues.

“Once we had senior management’s attention and linkedCRM’s benefits to the company’s long range plans, the dollarsjust opened up,” said the Vice President. “We aligned ourvision of CRM into what other top managers were doing andit became a critical project.” One measure of the Vice Presi-dent’s success was that U S West executives began touting theemerging CRM strategy to Wall Street analysts as a source ofcompetitive advantage for the firm. Demonstrating businessimpact is usually a key step in gaining long term executivesupport for data-to-knowledge initiatives.

Change Beliefs and Behaviors. One of the biggest challengesfacing the vice president and his team was changing the cul-ture at U S West. The Small Business and Consumer marketingunits had to be changed from a monopoly-oriented environ-ment that was distinctly product-focused to a more customer-oriented attitude. This meant overcoming cultural resistancethat the vice president estimated was 9.5+ on a 10 scale.

Creating new beliefs around the use of data for decision-mak-ing was partly a matter of showing how this new approachwas linked directly to business results. It also meant educatingthe organization about the value of using analytic outputs.

When U S West lost its monopoly position in the regionaltelecommunications industry, senior management expectedthe firm would lose 35% of its customer base. Thus a majorchallenge facing the company was how to grow revenues inan increasingly competitive market. In late 1997, U S West’svice president of relationship and database marketinglaunched a major initiative to increase sales with existing cus-tomers, retaining the most profitable and potentially prof-itable ones. His goal was to transform the former monopolybusiness into a truly customer-centric entity by leveraging thefirm’s transaction and marketing data.

This case illustrates some critical tasks for data-to-knowledgeimplementation teams that must be addressed if a company isto maximize the value of its data infrastructure.

Company Context

As head of database marketing for U S West’s retail business,the vice president and his team focused mostly on the firm’sConsumer and Small Business segments. From the outset, theirgoal was to create a “closed loop marketing system” whichmeant linking inbound and outbound channels so that no dataabout customers would fall through the cracks. For example,when a customer who had recently been sent a mailing aboutU S West’s Internet services called in with an unrelated billingquestion, the customer service representative would knowabout the Internet mailing and could try to close the sale. Thisapproach would enable U S West to capture, analyze, andapply customer data to target its most potentially profitablecustomers more effectively with the right products in the rightchannel.

But to create this closed loop marketing capability, commonlyreferred to as customer relationship management (CRM), thedatabase marketing vice president had to pursue a complexorganizational and cultural change initiative. Five activitiesstand out in the U S West experience as critical to developingbroad organizational capabilities to better use transactiondata for decision-making. They are as follows:


Working Paper

U S WestU S WEST is one of the largest telecommunications firms in the world. With revenues of $12.4 billion (1998) and nearly

55,000 employees, U S West serves both retail (residential, small business, large business and government) customers as well

as wholesale “carrier” customers.

makers’ levels of trust in the evolving data infrastructure.Developing this infrastructure is an ongoing process that con-tinues to evolve as an increasingly sophisticated group of endusers demand broader access to high quality data.

Manage the Implementation of Multiple Projects. Buildingbroad data-to-knowledge capabilities at U S West has meantmanaging multiple interdependent projects needed to createthe necessary organizational changes. “But it’s not a linear,staged process,” said the database marketing vice president.“It’s more like managing a 15-level chess game because thereare 15 CRM-related projects happening concurrently, and theirinterdependencies must be managed.”

To add to this complexity, most projects required cross-func-tional integration of multiple types of expertise. One U S Westproject, for example, has focused on implementing “intelligentscripting” capabilities in the company’s call centers. Thiswould enable customer service representatives to use the lat-est information on a specific customer to determine the bestproduct to try to sell them. The intelligent scripting projectrequired expertise in call center management, leading edgescripting software, model building, training and change man-agement processes, as well as knowledge of the retail market.

Managing these cross-functional projects was tricky and timeconsuming, not only because of the different experts involved,but because sponsors had to take into account the company’sability to absorb the new capabilities being provided. Thedatabase marketing vice president worked hard to see thatdecision-makers would view the new data-to-knowledgecapabilities as an enablement, rather than as a threat.

At U S West this meant that after piloting each project, theteam went through a significant period of educating the orga-nization about the benefits of the new capabilities to get thenecessary buy-in. And, where necessary, developers wouldadjust the system’s capabilities to provide something theorganization would accept. Sometimes, while creating a pilotfor a project, the CRM team would simultaneously beginbuilding the infrastructure required for full implementation,even though the program had not been approved for a full rollout. This was done in recognition of the lead time necessaryto create the larger infrastructure. And the CRM team wantedto minimize decision-makers’ frustration with being deniedaccess to the new capabilities once their enthusiasm waspumped up by seeing the results of a pilot.

“We’re trying to move business units from a monopoly, cost-plus paradigm to where they value targeting people based onmodel outputs,” said the database marketing vice president.“But that requires a new recognition that analytic modeling isnow central to our core competencies.”

Getting this message across was not easy, however. Often thedata-to-knowledge sponsor thought he was succeeding onlyto find out that the old behaviors still prevailed. “At times theculture can seduce you into thinking you’re making progressselling CRM,” said one team member, “only to find out laterthat nothing has changed. Trying to change this culture is likehitting up against a Jell-O wall.”

One critical action taken by the vice president was to use hisinfluence to replace three IT directors and one marketingdirector, all of whom had refused to change in support of thenew strategic vision. These moves sent an important signal tothe rest of the company that new behaviors around data usewere now expected. Culture change is one of the stickiestissues in changing a firm’s orientation to data use. It takestime, and invariably involves a combination of carrot and stickinitiatives to encourage the new behaviors needed.

Create Technical and Data Infrastructure. Throughout thedifferent phases of implementing new data-to-knowledgecapabilities, the vice president’s team continued their effortsto improve the technology and data infrastructure needed tosupport new ways of working. As a corporate group they triedto enforce the sharing of common technology platforms asmuch as possible to reduce costs, enhance organizationalknowledge around particular technologies, and increaseopportunities for cross-unit interaction.

At a tactical level, the database marketing group also workedto implement integrated software tools that would providefaster and more sophisticated analytic and data productioncapabilities. For example, U S West has linked daily updatesfrom the firm’s service order system to analytic models andmore automated Valex Campaign Management software tosupport event-triggered marketing campaigns.

At the same time, programs to enhance data quality demand-ed ongoing resources and attention. Data infrastructure prob-lems sometimes created unexpected set backs at the tacticallevel as product managers tried to change their approach tomarketing only to find the data they were using did not meetexpected quality standards. This not only placed new resourcedemands on the technical staff, but it also reduced decision-


Working Paper

Finally, as new data-to-knowledge capabilities took hold,sponsors found their resources stretched increasingly thin, asend users began demanding more and more changes andimprovements in the systems involved. The database market-ing team at U S West found this conflict an ever-present chal-lenge. “There is constant tension between managers withmore tactical requests who want us to help them stop thebleeding,” said one team member, “and our goal of providingthe organization with world class CRM capabilities.” Balancingpriorities and managing conflicting demands with limitedresources is an inherent part of evolving data-to-knowledgecapabilities.

Change Roles and Responsibilities. Using multiple projects tocreate the technology and data infrastructure needed to sup-port new analytic capabilities is only part of the story.Redefining the roles and responsibilities necessary to makeuse of the data in analysis and decision-making was alsoimportant at U S West. For example, implementing the intelli-gent scripting capabilities in the company’s call centers wouldchange the responsibilities of customer service representa-tives, turning them into more proactive salespeople. Thisrequired new skills and training.

In another part of the company, one director changed theentire structure of her marketing team. Instead of asking indi-viduals to perform a broad range of database marketing tasks,team members became more specialized into roles such ascampaign manager, list manager, data analyst, modeler, andproduct manager. Using transaction data effectively toenhance or transform key business processes will almostalways require significant changes in key roles. Data-to-knowledge sponsors must anticipate this and plan to imple-ment these changes.


Working Paper

Working Paper

About the Institute for Strategic Change

The Accenture Institute for Strategic Change is a "think andact" tank dedicated to keeping Accenture among the worldleaders in breakthrough management thinking. In today'sultra-competitive, knowledge-based economy, such thinking iswhat enables businesses to establish and redefine markets.

Based in Cambridge, Massachusetts, the ISC is made up ofexperienced management researchers working in concert withbusiness educators and executives.

The ISC conveys its ideas through Research Notes, WorkingPapers, articles, books, conferences, and hosted client visits.Each of these products shares a common goal: to provideactionable insights and ideas for addressing strategic businessissues.

The Institute is currently working on these seven researchprojects:� Global mCommerce� The Art of Work� mMe� Next Generation Leaders� Supply Chain Effectiveness� CRM/Marketing Strategy� Capturing Value in Financial Services

For more information, please contact the Institute at (617)454-4180 or send an email to [email protected].

About the Authors

Thomas H. Davenport is the Director of the Accenture Institutefor Strategic Change. He is a widely published author and aspeaker on the topics of information and knowledge manage-ment, reengineering and enterprise systems. His latest book,published by Harvard Business School Press, is entitled MissionCritical: Realizing the Promise of Enterprise Systems. Tom maybe reached at (617)454-8201 or [email protected].

Jeanne Harris is a Senior Research Fellow at the AccentureInstitute for Strategic Change. She has extensive experience inhelping clients realize business value from transaction data.Jeanne has been involved in numerous knowledge manage-ment, data warehousing, decision support, data mining andcustomer intimacy initiatives. Jeanne can be reached [email protected].

David W. De Long is a Research Fellow at the Accenture Insti-tute for Strategic Change. A former researcher at both Har-vard Business School and M.I.T's Sloan School of Management,his work focuses on helping firms manage the organizationalchanges necessary to derive real business value from electron-ic information. Other current research interests include knowl-edge management and the changing role of management inecommerce. He may be reached at [email protected].

Alvin L. Jacobson is a founder and a partner of Hartwell Asso-ciates, a management consulting practice focused on financialservices and the securities processing businesses. His mostrecent research examines the role of the account executive inmanaging customer knowledge and customer relationships.


Data to Knowledge to Results: Building an Analytic Capability...Results: Building an Analytic...

Documents

Transcript of Data to Knowledge to Results: Building an Analytic Capability...Results: Building an Analytic...