The Impact of Data Integration on the Costs and Benefits of … · 2016-01-29 · Data Integration...

Data Integration

The Impact of DataIntegration on theCosts and Benefits ofInformation Systems

By: Dale L. GoodhueInformation and Decision SciencesCarlson School of ManagementUniversity of Minnesota271 19th Avenue SouthMinneapolis, Minnesota 55455

U.S.A.

Michael D. WyboFaculty of ManagementMcGill University1001 Sherbrooke Street WestMontrdal, QudbecCanada H3A 1G5

Laurie J. KirschInformation and Design SciencesCarlson School of ManagementUniversity of Minnesota271 19th Avenue SouthMinneapolis, Minnesota 55455

U.S.A.

Abstract

For many organizations, the ability to make coor-dinated, organization-wide responses to today’sbusiness problems is thwarted by the lack of dataintegration or commonly defined data elementsand codes across different information systems.Though many researchers and practitioners haveimplicitly assumed that data integration alwaysresults in net benefits to an organization, this ar-ticle questions that view. Based on theories oforganizational information processing, a modelof the impact of data integration is developed thatincludes gains in organization-wide coordinationand organization-wide decision making, as wellas losses in local autonomy and flexibility, andchanges in system design and implementationcosts. The importance of each of these impactsis defended by theoretical arguments and il-

lustrated by case examples. This model suggests

that the benefits of data integration will outweighcosts only under certain situations, and probablynot for all the data the organization uses.Therefore, MIS researchers and practitionersshould consider the need for better concep-tualization and methods for implementing "par-tial integration" in organizations.

Keywords: Organization-wide information sys-tems, data integration, interdepen-dence, flexibility, implementationcosts

ACM Categories: H.2, H.2.1, K.6

IntroductionA common language for communicating aboutbusiness events is a prerequisite for coordinatingdiverse and far-flung units of organizations. In therealm of computer information systems, one formthis common language can take is data integra-tion, or data elements with standard definitionsand codes. Researchers studying the impact ofinformation technology on organizations oftenassume that the common language provided bydata integration exists, or that it will be developedbecause of the benefits of increased communica-tion within (or across) organizations (e.g., Huber,1990; Malone, et al., 1987).

However, there is evidence that this commonlanguage of logically compatible data does notexist in a great many large organizations today(Goodhue, et al., 1988). Within a single companythere are often different identifiers for keybusiness entities, such as customer or product,different schemes for aggregating key indicators,such as sales or expenses, or different ways ofcalculating key concepts, such as profit or returnon investments. These inconsistencies causemajor problems when firms ask questions thatspan multiple systems or multiple subunits,thwarting their ability to make coordinated,organization-wide responses to today’s businessproblems.

Researchers and practitioners have respondedto this problem by developing ways to increasedata integration. For example, network and rela-tional data structures that could accommodatethe varied needs of many users in a single in-tegrated data structure have been described

MIS Quarterly/September 1992 293

Data Integration

(Bonczek, et al., 1978; Date, 1981). The entityrelationship model has been proposed as ameans of conceptually modeling all of a corpora-tion’s data, providing the basis for integratedcomputer systems (Chen, 1976; McCarthy,1982). Other researchers have recognized thatorganizations do not start from a clean slate, andexisting systems typically are not integrated.Thus, theoretical approaches for integrating ex-isting, disparate database schemas have beendeveloped (Batini, et al., 1986). Still others havefocused on developing practical methods for im-mediate use in actual organizations, including in-formation engineering methodologies that wouldprovide an information architecture by whichorganizations could transform their systems fromnon-integrated to integrated forms (Finkelstein,1990; IBM, 1981; Martin, 1982; 1986). In general,the presumption has been that greater data in-tegration is universally desirable and leads togreater benefits for organizations.

In spite of the conceptual appeal of thesemethods for achieving data integration, manyorganizations that have attempted them havefailed or experienced major difficulties (Goodhue,et al., 1988; 1992; Hoffer, et al., 1989; Ledererand Sethi 1988; 1991). This raises an importantquestion: If data integration is universallydesirable, why is it so difficult to implement inpractice? Existing research has offered twogeneral explanations. The first considers im-plementation pitfalls of data integration, such asthe difficulty of obtaining top management sup-port or the need for managing expectations(Goodhue, et al., 1988; Hoffer, et al., 1989;Lederer and Sethi 1988; 1991). A second ex-planation is the possibility of shortcomings in themethodologies used (Goodhue, et al., 1992).

This article considers a third possible explana-tion. Counter to the usual presumption, data in-tegration efforts may fail because in certainorganizational contexts they do not provide suf-ficient benefits to offset their costs. This conclu-sion emerged gradually as we pondered theevidence of over 35 case studies of data manage-ment efforts conducted at the Massachusetts In-stitute of Technology (MIT) and the University Minnesota over the past eight years. 1 The

The case examples come from large ($500 million and above)U.S. or Canadian firms. All company names are disguised.For more details see Goodhue, et al. (1988) and Goodhue,et al. (19.~2).

cumulative experience represented by thosecases strongly suggests that the net benefits ofincreased data integration will not always bepositive but will depend upon three mainorganizational factors: (1) the interdependenceof subunits, (2) the need for locally unique or flex-ible action by subunits, and (3) the difficulty designing and implementing systems with in-tegrated data.

First, theories of organizational information pro-cessing (Daft and Lengel, 1986; Galbraith, 1973;Tushman and Nadler, 1978) are discussed toshow that the same three factors that emergedfrom the case-study evidence can be derivedfrom an existing theory base. Next, each of thethree factors is explored further. They are defend-ed by additional theoretical argument and il-lustrated with selected anecdotal evidence drawnfrom the case studies. This leads to several prop-ositions for empirically testing the main tenets ofthe new model in future empirical work. Thediscussion begins with an overview of the dataintegration problem in the context of largeorganizations.

What Is Data Integration?Data integration generally means the standard-ization of data definitions and structures throughthe use of a common conceptual schema2

across a collection of data sources (Heimbignerand McLeod, 1985; Litwin, et al., 1990). Data in-tegration ensures that data have the same mean-ing and use across time and across users,making the data in different systems or databasesconsistent or logically compatible (Martin, 1986).

The operational definition used here focuses onlogical compatibility at the level of the individualdata element. We define data integration as theuse of common field definitions and codes acrossdifferent parts of the organization.3 This defini-

A conceptual schema specifies field and record definitions,structures, and rules for updating data values.

An alternate approach to hammering out agreement on defini-tions and codes across all systems would be to allow eachsubunit to design its systems autonomously, and then meldthese disparate databases into a federated database system(Heimbigner and McLeod, 1985; Sheth and Larson, 1990)where presumably at least some data could be translated ormapped to permit limited sharing. This leaves the sharabilityof data somewhat up to chance, because without central plan-ning and discipline the actions of any single subunit couldnegate the sharability of any data element.

294 MIS Quarterly~September 1992

Data Integration

tion lets us easily conceptualize the degree ofdata integration in a firm. Beyond none at all, theleast amount of data integration is a commondefinition and set of codes for a single field acrosstwo databases or information systems. Data in-tegration can be increased along one or both oftwo dimensions: (1) the number of fields withcommon definitions and codes, or (2)thenumber of systems or databases that adhere tothese standards.

Table 1 shows three examples of integrated andnon-integrated data for two divisions of a singlecompany. In the first example, both divisions inthe integrated environment have the samePART_ NUMBER code in all systems for everypart used in their products. In the non-integratedenvironment, the two divisions have not coor-dinated the assignment of PART_NUMBERcodes. Though it is possible to translate from onedivision’s part numbers to another’s, if there aremany part numbers it is quite difficult to deter-mine the mapping (and keep it up to date!).

In the second example, both divisions in the in-tegrated environment use a single code to iden-tify customers. In the non-integrated environ-ment, Division B has a two-level structure forcustomers, so separate locations that order andpay for their own purchases can be givenseparate customer identifiers, even if they arereally part of the same large company. With thisarrangement it would not be possible to translatedata from Division A into the form used by Divi-sion B. Finally, the third example shows that dif-ferent divisions can have different ways of

handling adjustment (e.g., returns or discounts)in their sales figures. In such situations, transla-tion of the data to a common form is often possi-ble, but what might be a mere inconveniencewhen a single data element is involved becomesa serious problem when there are tens or hun-dreds of such data elements that need to beshared. The difficulties are magnified when theindividuals making the translation are not awareof the differences in the definitions or cannotdetermine exactly what has been included or ex-cluded in calculating a particular field.

A concrete example will assist us in understand-ing what data integration or the lack of it mightmean in an organization.

The Van Buren Bank has felt the effects ofderegulation, which has made the once stablebanking industry highly competitive. With thedecreased spread between borrowing and lend-ing rates, profits on loans have dwindled, makingprofit on services to customers critical. In the cor-porate banking group, account managers who inthe past could concentrate only on loan volumesmust now focus on customer and product prof-itability. This means they must make decisions dif-ferently and need different kinds of information.For example, if a long-time customer threatens totake his or her business elsewhere unless he orshe is given an unusually low interest loan, the ac-count manager must decide whether this is anotherwise profitable customer in terms of his or hertotal business with the bank.

In order to determine how profitable a customeris, the account manager must gather informationabout the various products and services the

Table 1. Examples and Counter Examples of Data Integration

Integrated Environment Non-Integrated Environment

Division A Division B Division A Division B

1. PART_NUMBER 115899 115899 115899 337189Codes for a 3/4" SCREW

CUSTOMER _ ID 42765 42765 42765 42675,Codes for IBM 49345,

51349, etc.

3. Definition for SALES Not adjusted Not adjusted Not adjusted Adjusted forfor returns for returns for returns returns

MIS Quarterly~September 1992 295

Data Integration

customer buys and the profitability of each. Con-ceptually this could be done by communicatingwith other account managers around the worldwho also do business with this customer (or com-municating with their electronically readablerecords) and consolidating the information.

However, it may or may not be true that thecustomer is identified in the same way in eachaccount manager’s records; or that the variousproducts and services the bank sells are iden-tified the same way or grouped into the samecategories; or that the size of each account isrecorded in the same currency; or that discounts,refunds, and so forth are handled in the sameway. The account manager must translate all thisinformation from many sources into a commonform and determine how profitable this customeris.

Unfortunately at the Van Buren Bank, all customeridentifiers were typically assigned by the branchesand were not standard across the bank. Therefore,there was no way to identify all the business of agiven customer short of phoning up every branchand asking. It was clear that the Van Buren Bankrequired much more data integration than was cur-rently built into its information systems.

Few large international banks would be willingto require every account manager (regardless ofproduct type, customer type, or country) to recordall his or her information in exactly the same for-mat. On the other hand, neither would suchbanks be likely to permit each local accountmanager to record all information to be mosthelpful to him or her individually, with no effortat standardization. Most banks would likelychoose a middle road. This "partial integration"might be achieved by standardizing on certaindata elements (perhaps key identifiers, such ascustomer and product) and allowing flexibility oncertain other data elements. Such a "partial in-tegration" solution has the advantage that themost frequently used and troublesome data couldbe standardized, while leaving some flexibility tolocal account managers in defining other dataless critical for cross-bank questions. But deter-mining how many and which data elements tostandardize then becomes an important and dif-ficult data management question. One way to ad-dress this question is to consider the impact ofincreasing levels of data integration on the costsand benefits of information systems.

A Model for the Impact ofData IntegrationThe model to be developed here depends upona very rational view of organizations andorganizational design. It assumes that benefitsand costs will be summed at the level of theorganization and that everyone in the organiza-tion shares the goal of maximizing overall

¯ benefits minus costs. In a later section of this ar-ticle several factors are considered that do notfit within this very rational view, including powerand politics, and the interdependence betweendata integration, strategy, and structure. In spiteof these caveats, the rational model provides apowerful conceptual framework for thinking aboutthe issue of data integration.

Costs and benefits of IS

The idea that a careful conceptualization of costsand benefits can lead to useful insights aboutdesign choices for information systems (IS) is notnew. As early as 1959, Marschak conceptualizedthe differences in information system designs asdifferent ways of selecting and partitioning theavailable data from the environment. Informationsystems can be seen as having costs (design andimplementation costs to the provider of thesystem) and benefits (improved decision makingand more informed actions for the user of the in-formation), which Can vary quite independently.Organizations face an array of possible informa-tion systems, each with certain implementationcosts and different expected decision-makingbenefits. Rational organizations should choosea design that maximizes the benefits minus thecosts (Emery, 1982; Marschak, 1959; 1968; 1971;Mendelson and Saharia, 1986).

To look at data integration in this light, a clearconceptualization is needed of how integrationaffects both the benefits and the costs to usersand implementers. Organizational informationprocessing theory (Daft and Lengel, 1986;Galbraith, 1973; Tushman and Nadler, 1978)gives us a basis for such a conceptualization.

OrganizationalinformationprocessingOrganizational information processing theory isconcerned with the design of organizations, in


D~t~ Integration

particular the design of structures or mechanismsto deal with information processing requirements.Uncertainty, a central concept in the theory,drives the need for information processing.Galbraith (1973) proposes that when uncertain-ty in an organization is low, three differentmechanisms can be used to resolve it: rules andprocedures, hierarchies, and goals. However,when an organization faces greater uncertaintythan can be handled by these three mechanisms,it must either reduce the need for informationprocessing (through slack or self-containedtasks), or it must increase information process-ing capacity (through computerized informationsystems4 or lateral relations).

While recognizing that uncertainty can be dealtwith by any of Galbraith’s options, this paperfocuses primarily on his option of computerizedinformation systems. Such systems differ alongseveral dimensions, two of which parallel nicelythe concept of data integration discussed in thispaper: the degree of formalization of the data andthe scope of the database (Galbraith, 1973). Dataintegration (common field definitions and codes)is an example of a highly formalized languagefor describing the events occurring in anorganization’s domain. The scope of data integra-tion is the extent to which that formal languageis used across multiple functions or subunits ofthe organization. Thus, the degree of data in-tegration of a computerized information systemcan be seen as an important design characteristicin dealing with organizational uncertainty. Tounderstand when data integration is especiallydesirable, we must delve deeper into the sourcesof uncertainty facing the organization.

Sources of uncertaintyFor Galbraith, the uncertainty that motivateschoices among information processing mecha-

Galbraith uses the term vertical information systems ratherthan computerized information systems, and his examplestypically refer to systems providing information vertically upthe hierarchy to managers responsible for all units involved.However, he makes it clear that the essential aspect of thisalternative is the provision of information to those who needit: "It should be pointed out that taking information from itspoints of origin and collecting it into a global data base to bepresented at a level high in the hierarchy is only one methodof operationalizing this dimension .... However, the policydimension is to increase the scope of the data base availableto the decision mechanism independent of what it is or wherein the hierarchy it is located" (Galbraith, 1973, p. 33).

nisms is defined in general terms for the wholeorganization, without any detailed description ofwhere it comes from. Tushman and Nadler (1978)pushed the analysis down to the level of thesubunit and distinguished between three differentsources of uncertainty: complex or non-routinesubunit tasks, unstable subunit task environ-ments, and interdependence between subunits.For example, a manufacturing unit employing atemperamental, thin film manufacturing tech-nology might face uncertainty in the operation ofthis highly variable process (complex tasks),uncertainty from the environment where custo-mer demands were constantly changing (un-stable environment), and uncertainty in dealingwith a procurement office that did not alwaysunderstand the urgency of its requests (in-terdependence between subunits). Figure 1 com-bines the ideas of Tushman and Nadler (1978)with those of Galbraith (1973), showing the threesources of uncertainty.

Tushman and Nadler argue that subunits in asingle organization may face quite different taskcharacteristics and task environments andtherefore might need quite different informationprocessing mechanisms. For example, the thinfilm technology manufacturing unit might needquite different information processing capabilitiesfrom a more traditional assembly line subunit fac-ing a stable environment, even though both arein the same firm. In addition, the most appropriateinformation processing mechanism for a givensubunit could change over time if task charac-teristics or task environments change. Thus, in-dividual subunits might desire the autonomy andflexibility to design and change their informationsystems independently.

If we look at data integration in this light, we cansee that its impact might differ depending onwhich source of uncertainty dominated. Considerthe three subunits mentioned above: the thin filmmanufacturing unit, the assembly line unit, andthe procurement office. Where uncertainty comesmainly from interdependence between subunits(for example, when the two manufacturing sub-units must deal with the procurement unit to pur-chase material), data integration would be highlydesirable because it provides a standardized, for-malized language shared by all subunits andfacilitates communication between the interde-pendent subunits. On the other hand, when un-certainty comes mainly from task complexity or


Data Integration

Interdependenceamong subunits

Complex ornon-routinesubunit tasks

Unstable subunittask environment

Uncertainty

Reduce need forinformation processing

¯ Slack resources¯ Self-contained tasks

~Increase informationProcessing capability

¯ Computerized informationsystems

¯ Lateral relations

Figure 1. A Model of Organizational Information Processing(Combining Galbraith, 1973, and Tushman and Nadler, 1978)

environmental instability unique to individualsubunits (for example, the temperamental proc-ess and changing demand faced by the thin filmmanufacturing unit but not faced by the otherunits), mandatory data integration might reducethe flexibility of an individual subunit to redesignits information systems to address its uniqueneeds. Here a high level of data integration couldbe undesirable.

Uncertainty and equivocalityRecent additions to organizational informationprocessing theory take the concept of uncertaintyas used by Galbraith (1973) and Tushman andNadler (1978) and break it into information re-quirements of two types: uncertainty andequivocality (Daft and Lengel, 1986). This distinc-tion can help us understand when computer-ized information systems (whether with dataintegration or not) may be of little assistance inmeeting a firm’s most pressing informationneeds. Uncertainty by this new definition is theabsence of specific, needed information. For ex-ample, a manager might want to know whethersales of widgets dropped in the southern regionlast month. The question is clear, and given dataon specific variables, the uncertainty would beremoved.

Equivocality means there are multiple, conflictinginterpretations of a situation. When equivocalityis present, the precise information needed toresolve the situation is not clear. For example,the same manager might want to know why salesdropped in the southern region last month. Forsuch a question, a number of types of informa-tion might be appropriate, including the attitudeand behavior of the sales force, the actions ofcompetitors, the economic situation, trends incustomer satisfaction, the weather, etc. Com-peting perspectives might be as useful to themanager as factual data.

In general, uncertainty can be reduced by a suf-ficient amount of information, while equivocalitycan be reduced by sufficiently rich information(Daft and Lengel, 1986). Certain types of infor-mation processing mechanisms (such as com-puterized information systems and, by impli-cation, systems with integrated data) providelarge amounts of information and can thus helpreduce uncertainty. However they are not as richa source as other information processing mech-anisms (such as face-to-face meetings) and thusare not as effective in reducing equivocality.5

Note that in general it is not necessary to cl~oose betweencomputerized information systems and rich informalsystems--both exist side by side and, ideally, support eachother.


Data Integration

Therefore, the appropriateness of informationsystems with integrated data might depend onthe degree to which unmet information needs arecharacterized by uncertainty as opposed toequivocality. Using the framework provided byTushman and Nadler, Daft and Lengel carry theiranalysis down a level to look at each of the threesources of information requirements: interactionsbetween subunits, complex subunit tasks, andunstable subunit task environments.

First, consider information requirements thatresult from interaction between subunits. Inter-dependence between subunits leads to a needfor greater amounts of information, and differen-tiation between subunits leads to a need for richerinformation (Daft and Lengel, 1986). Applyingthese ideas to data integration, if interdepen-dence is high (for example between the twomanufacturing units and the procurement office)the formalized language of integrated data willallow large amounts of information to be easilyshared. But if subunits are highly differentiated(for example the thin film technology unit and theassembly line unit), different perspectives,values, or frames of reference may lead to highequivocality and make it difficult to implementdata integration because it would be more difficultto agree on common definitions for data itemsand the events they describe. This suggests thatdata integration would probably be most usefulwhere subunits were very interdependent andmost easily implemented where subunits werenot highly differentiated.

Now consider the other two sources of uncertain-ty: complexity of the subunit’s tasks and stabili-ty of the subunit’s task environment. In generala lack of analyzable cause-and-effect relationsleads to a need for more "rich" information proc-essing mechanisms (such as face-to-facemeetings) to reduce equivocality, while varietyand constant change lead to a need for greateramounts of information processing (usingmechanisms such as computerized informationsystems) to reduce uncertainty (Daft and Lengel,1986). Where organizations face great equivocali-ty as opposed to uncertainty, data integration andthe formalized language it implies (as well ascomputerized information systems in general)may not be as useful in addressing informationprocessing needs.

Using this logic we might expect that informationsystems with integrated data would be quite

useful in, for example, organizations in the snackfood industry in the U.S., where new productsand changing consumer tastes make for varietyand constant change but the basic ground rulesare well-understood. On the other hand, systemswith integrated data might be less useful (andface-to-face meetings more useful) in organiza-tions trying to gain a foothold in the new marketsof Eastern Europe, where cause and effect rela-tionships are ambiguous and critical success fac-tors are unclear and constantly shiffing.

Balancing benefits againstimplementation costsFinally, both Galbraith (1973) and Tushman andNadler (1978) emphasize the importance balancing greater effectiveness of more complexinformation processing mechanisms against theirgreater costs: "[T]he basic design problem is tobalance the costs of information processingcapacity against the needs of the subunit’swork--too much capacity will be redundant andcostly; too little capacity will not get the job done"(Tushman and Nadler, 1978, p. 619).

While organizational information processingtheory does not give much detailed help in con-ceptualizing the impact of data integration on thecosts of design and implementation of systems,we can resort to arguments on design difficultyin general (Simon, 1981) and the design and im-plementation of information systems in particular(Banker and Kemerer, 1989; Brooks 1975; Mar-tin, 1982). These suggest that data integrationcould have a positive impact on reducing costsby reducing redundant design efforts. However,because multiple subunits would be involved,data integration could also have a negative im-pact on costs by increasing the size and complex-ity of the design problemor increasing the dif-ficulty in getting agreement from all concernedparties.

If we consider only those situations where uncer-tainty dominates and information systems ingeneral are very appropriate choices, this sug-gests that the impact of data integration on thecosts and benefits of information systems willcome primarily through three potential factors:(1) increased ability to share information to ad-dress subunit interdependence, (2)reducedability to meet unique subunit information re-quirements, and (3) changes in the costs of in-


Data Integration

formation system design and implementation.This is illustrated in Figure 2. The next three sec-tions of this paper look in more depth at each ofthese potential impacts.

Factor 1: Benefits FromIntegrated, Sharable DataIn his discussion of the application of informationtechnology to organizational design, Simon(1973) suggests that "there is no magic in com-prehensiveness .... The mere existence of amass of data is not sufficient reason for collect-ing it into a single comprehensive informationsystem" (p. 271). The major benefits of "a singlecomprehensive information system" come fromthe ability to share or aggregate informationacross many divisions or functions or units of theorganization. The need for sharing informationis greatest where actions and events acrossthose units are interdependent, that is, where theactions of one subunit affect the actions or out-comes of another subunit (McCann and Ferry,1979).

When there is interdependence but not too muchdifference in perspective between the subunits,

a formalized information system (including dataintegration) will be an appropriate mechanism(Daft and Lengel, 1986). We can distinguish least two different impacts data integration mighthave on an organization: (1) improved manager-ial information for organization-wide communica-tion and (2) operational coordination between.interdependent parts of the organization(Goodhue, et al., 1988).

Improved communication acrosssubunitsData integration is necessary for data to serveas a common language for communicating aboutevents the organization faces (Galbraith, 1973).Without data integration there will be increasedprocessing costs and ambiguity of meaning asmessages between subunits are processed. Theresult should be delays, decreases in com-munication, reductions in the amount of sum-marization, and greater distortion of meaning(Huber, 1982). Thus, data integration facilitatesthe collection, comparison, and aggregation ofdata from various parts of the organization,leading to better understanding and decisionmaking when there are complex, interdependent

Data / (-) =integration ~

Ability to sharecorporate-widedata to addresssubunit interdependencies

Flexibility to respondto subunit needs forlocally uniqueinformation

Costs of designingand implementinginformation systems

Costs andbenefits ofinformationsystems

Figure 2. The Impact of Data Integrationon the Costs and Benefits of Information Systems


Data Integration

problems. Two examples from our case studiesillustrate this.

When on-time deliveries (a critical competitiveissue in the semiconductor market) fell to only 70percent, a multi-disciplinary team at Devlin Elec-tronics used organization-wide integrated schedul-ing data to track how production schedules weremade, changed, and adjusted by the many dif-ferent groups involved. They found a number ofinterlocking problems: some plants were not prop-erly updating their inventory levels and equipmentconditions, marketing was overriding the schedulewithout regard to plant capabilities, and plants wereoverriding the system without regard to criticalorder requirements. Organization-wide integrateddata allowed Devlin to understand its problems andtake corrective action. On-time delivery improvedfrom 70 percent to 98 percent.

At Southern Cross, Inc., the executive vice presi-dent expressed concern at hearing that in spite ofmany new accounts, sales were up by only one-half percent. He complained that, "It’s obvious weare bleeding somewhere, but nothing in these stan-dard reports gives me any indication of what theproblem is." He demanded an analysis across allregions, all customers, and all products to deter-mine the cause. Because the information was con-rained on several different (unintegrated) systems,and because products had over time been groupedinto various different (and shifting) categories forreporting purposes, there was no way to do theanalysis in an automated fashion. Using spread-sheet programs and manually backing out prod-ucts that had switched categories and adjustingother inconsistencies between the differentsystems, top-level analysts were able to assem-ble a compatible base of information to answer theEVP’s questions after 40 person-hours of effort.The analysis indicated the sales slump was occur-ring primarily in the old, established, long-timedistributorships (an insight that was not apparentfrom analysis of the non-integrated data). With theproblem somewhat pinpointed, managers couldtake appropriate further action.

One possibility is that the Southern Crosssystems were not well-designed in the first place.However, Southern Cross has for years enjoyedgreat strategic advantage from its systems. Sincethe need for this type of cross-product lineanalysis had not surfaced in the past, its systemswere not designed to address it.

The temptation might be to design enough dataintegration into all our systems to be able toanswer any question that might come up in thefuture. However, as seen later in this paper, thevery real benefits of data integration must bebalanced against equally real costs.

Improved operational coordinationacross subunitsThe actions of separate subunits in an organiza-tion can be coordinated using a number of dif-ferent approaches (Galbraith, 1973; Malone,1987). For example, organizational hierarchiescan be designed such that the most interdepen-dent units are closely linked (Thompson, 1967).However, in order to pass large amounts of in-formation between subunits, a formalized, stan-dardized language is required. Thus, some waysof coordinating subunits will require integrateddata so that unambiguous messages may be sentfrom one subunit to another. Top managementmust recognize that approaches used to coor-dinate subunit action may be constrained by thedegree to which a common language and adatabase of integrated, compatible data exist.

Each of Greenfields Products’ five divisions hasits own sales force and distributes its own prod-uct lines in separate trucks. Top managementwanted the ability to present a single face to thecustomer by having only one (not five) salesper-son call on each customer and only one (not five)truck back up to the customer’s delivery dock.They realized that without a single, consistent baseof customer and order data, coordinating the ac-tions of the five divisions to create a singlecustomer interface would be impossible.

Burton Trucking Company completely overhauledits information processing systems based on asingle logical data model for the entire corporation.This allowed them to link not only acrossgeography but also across functions. By usingcommon, sharable data, they found their dispatchsystems (the responsibility of operations) could expanded with little effort to give them a much bet-ter shipment tracking system (the responsibility ofmarketing). Thus, data integration allowed themto capitalize on previously unrecognized in-terdependencies between dispatching and ship-ment tracking.

Our discussion of the impacts of data integrationon the sharability of information leads to a firstproposition. Note that we assume that informa-tion needs are characterized by uncertainty asopposed to equivocality, so that informationsystems in general are appropriate mechanismsfor organizational information processing.

Proposition 1: All other things being equal, asthe interdependence between subunitsincreases, the benefits of data integra-tion will increase, and the amount of


Data Integration

data integration in rational organizationsshould also increase.

Factor 2: Flexibility to MeetUnique Subunit InformationRequirementsKing (1983) notes that organizations wrestlingwith issues of control over computing face "thedilemma of deciding between organizational stan-dardization and departmental autonomy," andthat local autonomy over the design of systems"makes uniform collection of data for upwardreporting more difficult, whereas the impositionof organization-wide standards...diminishesdepartmental autonomy" (p. 329). Organization-wide data integration, as one aspect of the designof organizational computing resources, also im-plies some degree of "logical centralization" andsome central authority with control over thelogical aspects of data (Heimbigner and McLeod,1985). Thus, it may result in a loss of localautonomy in the design and use of data (Shethand Larson, 1990).

Data integration may involve not only a loss oflocal autonomy but also a loss of local effec-tiveness. Over time, different subunits may facedifferent inherent task complexity and differentenvironmental challenges. In order to effective-ly meet the challenges of unanticipated localevents, subunits may need the flexibility tochange their information systems on a local,unilateral basis (Tushman and Nadler, 1978).Gupta and Govindarajan (1986) argue that thescale advantages of resource sharing betweenstrategic business units do not come cost-freeand that, in particular, resource sharing may leadto losses of flexibility in responding to unan-ticipated local events for individual SBUs. For ex-ample, IS champions may see their ability tomove rapidly to achieve first-mover advantagesas being hampered by IS standards and in-frastructure goals, such as data integration(Beath, 1991).

Thus, constraints that forced a subunit facinglocally unique demands to conform to corporate-wide data integration standards might not be op-timal from the total organization perspective.Choosing the appropriate level of data integra-tion in an organization may require trading off im-proved organization-wide coordination against

increased local flexibility and local effectiveness.The next section looks at two ways local flexibili-ty can be reduced by data integration: com-promise and bureaucratic delay.

Compromises in meeting localinformation needsBy forcing the whole organization to conform toa single logical data design, data integration canreduce a local subunit’s knowledge about itsunique local environment and diminish its sen-sitivity to local environmental changes (Ashby,1959; Weick, 1976)., Consider an organizationwith several product divisions, each with its ownproducts. Each has its own different approach tomanufacturing and marketing those productsbased on the different characteristics of the prod-ucts and the markets for them. Top managementcould allow each division to design and imple-ment its own systems, based entirely on best ser-ving its local operational and information needs.The result would be systems that were locally op-timal but not integrated across the divisions, withdifferent definitions, identifiers, and calculationsin each division.

On the other hand, top management could re-quire the divisions to cooperate and design asingle logical data design, which could be usedby all of them. This "common" logical designwould allow easy access to compatible dataacross all the divisions but might not meet thelocal needs of the separate divisions as effective-ly well as separately designed logical datadesigns would.

At Burton Trucking, IS and operations hadcooperated to develop a common dispatchingsystem for its freight terminals across the coun-try. The new system contained integrated dataabout all customers, equipment, and shipments.Salespeople at each local terminal argued that forthe information to be valuable to them, they neededto add additional fields such as permissible deliveryhours, after-hours phone numbers, and special in-struction for drivers. But the salespeople couldn’tagree on exactly which additional fields should beadded. Terminals with close-in satellites (truckse~/ery 2-3 hours) had very different needs fromthose with distant satellites (trucks every 5 hours).IS decided that trying to standardize at this leveldidn’t make sense. They designed 10 extra fieldsthat the local people could use as they sawfit andgave them search capabilities and screens to up-date and query whatever data they needed in thosefields.


Data Integration

At Ventura Products, the Automotive Products Divi-sion undertook a major strategic planning effort toidentify ways in which its information systems couldassist in meeting its strategic goals. The result wasa consensus within the division that it should movetoward a tighter link with its customers and shoulddesign a new order-entry system as part of that ef-fort. However, there was an historic precedent atVentura of using a common-order entry systemacross all divisions. The corporate IS departmentfought Automotive Product’s proposal on the basisthat it would reduce corporate ability to present asingle face to customers and would make it harderto maintain compatible data on sales across thetotal corporation.

The concept of requisite variety in systems theorysuggests that in order to successfully respond toits environment, a system must be a good modelof that environment (Ashby, 1956; Conant andAshby, 1970). That is, if a division wishes to guideaction to respond to its environment, it must havean information system (and a data model) at leastas complex as the environment on all the essen-tial dimensions that determine what the bestresponse should be. The Burton Truckingscenario is an example of that principle: sales-people needing to respond to different types ofevents in different environments require differentdata models keyed to their unique environments.

Proposition 2a: All other things being equal,as the differentiation between subunitsincreases, data integration will imposemore and more compromise costs onlocal units; therefore, the amount of dataintegration in rational firms shoulddecrease.

Bureaucratic delays that reducelocal flexibilityEven when subunits can agree on a single datamodel and data definitions without undue com-promise, data integration can slow local flexibili-ty. When changing local business conditionssuggest a change to the data collected by onesubunit, all other subunits would have to studythe impacts of those changes on their own opera-tions, and all would have to agree or the changescould not be made. Requests for changes in theuse and semantics of data will necessarily involvebureaucratic delays as those requests are chan-neled through a remote authority (Leveson andWasserman, 1982).

At Superior Manufacturing, a large global corpora-tion, a major effort to integrate corporate data isunderway. As part of this effort, procedures havebeen put in place specifying how changes to thecorporate data model will take place. Requests firstgo to the data resource management office, whichdecides which subject areas are affected, and arethen passed to the relevant subject area datastewards who review, analyze, and recommend ac-tion. This recommendation will then be reviewedby affected division data administrators and theirdata user groups, who will approve or modify therecommendation. Finally, the modified recommen-dation will be presented to the corporate dataresource management policy and steering com-mittee for review and approval. All together, fivegroups of players will conduct four separatereviews of the request.

At Burton Trucking, when the operations groupwanted to automate freight transfer recording us-ing bar codes, it took six months to get the requestthrough the data administrator. According to onemanager, the data administration group didn’tunderstand what they were trying to do and"couldn’t square it with their data model." Aftera delay of six months, the project was allowed toproceed and was quite successful.

Proposition 2b: All other things being equal,firms with increased data integration willexperience greater bureaucratic delay ingetting approval for changes affectingthe data models used by individualsubunits.

Factor 3: IS Design andImplementation CostsThus far two factors organizations must considerin weighing the advisability of increased data in-tegration have been discussed: gains due to im-proved ability to manage interdependence, andlosses in local flexibility. Both of these factorsfocus on the impact of data integration once ithas been implemented. Even when these poten-tial gains outweigh the losses, organizations mustalso consider the cost of designing and im-plementing data integration. If the additional costis too large, rational firms will make do withoutintegrated data. Thus, the impact of data integra-tion on design and development costs is a~so acritical factor.

What is the impact on these costs? The typicalargument in the literature is that the higher costsof more expensive initial design and implemen-


Data Integration

tation are compensated by lower costs for subse-quent modifications of the systems, with a netincrease in productivity (see, for example, Mar-tin, 1982). This is certainly the situation in thefollowing example.

At Dobbs Insurance, over a period of about 10years, the pensions business unit has developedextensive shared databases used for all pensionsprocesses. The IS group readily acknowledges thatdeveloping integrated databases has cost morethan would have a series of stand-alone systemswith stand-alone databases. The biggest problemhas been getting all parties to agree on designissues that have important implications to manyparts of the organization. Dobbs justifies these ex-tra costs based on additional benefits from the in-tegrated data. For example, new systems thatrequire no changes to the existing integrateddatabases can be implemented very rapidly. Thisincludes new reports for cross-functional informa-tion requested by top management.

Though this example supports the typical argu-ment, in general the impact of data integrationon design and implementation costs dependsheavily on the situation. To understand this weneed to look closely at why and under what cir-cumstances up-front costs are higher, and whyand under what circumstances long-term costsare lower.

Why are up-front costs higher?There is little argument that in practice, buildingintegrated databases based on a wide-scope datamodel is more expensive in the short term thanbuilding separate stand-alone systems. A majorreason for this up-front cost is the greater com-plexity of the design problem: arriving at a singlelogical design for use across multiple organiza-tional groups can be a difficult problem (Litwinand Abdellatif, 1986). In general, the moresubunits involved and the more heterogeneoustheir needs, the more difficult it will be to developa single design that meets all needs.

A critical factor in the speed with which complexsystems can be designed is the degree to whichthe overall systems are "decomposable hierar-chies’ ’--that is, where collections of componentscan be grouped into subsystems with rather in-tense interaction within any subsystem and ratherlight interaction between subsystems (Simon,1981). This idea is commonly implemented instructured systems design through the concepts

of cohesion and coupling (Page-Jones, 1980).The need for a common data model movesdesigners away from a decomposable hier-archies approach, forcing them to deal with thefull complexity of the data needs for the wholeorganization. Whether or not the common datamodel is required by the interdependency needsof the organization, large-scope data integrationwill result in a more complex and more expen-sive design and implementation process. Thefollowing example shows the difficulty of linkingtwo different conceptualizations of the data anorganization needs.

Affer having identified the data requirements forthe first of two critical new systems, the Supportand Service Division of Ventura Products spentmany person-weeks of effort coordinating its newdata definitions with the several thousand datastandards on the corporate data dictionary.Because of the number of corporate standards andthe possibility of subtle differences in meaning, itwas quite difficult to determine which, if any, of thedivision’s desired data elements were exactly thesame as those in the corporate data dictionary.When the division began to design and developits second new system, it decided the time andcosts of coordinating with the corporate standardswere not justified by the benefits to them.

Though empirical research has not focusedspecifically on the effect of size and the numberof participants in data design, there is goodevidence for diseconomies of scale in thedevelopment of large systems in general. Seem-ingly disparate empirical findings on theeconomies of scale for systems development areconsistent with increasing economies for smallprojects and decreasing economies for large proj-ects (Banker and Kemerer, 1989). Further, theeffort to build large systems may increase withthe square of the size of the effort (Banker andKaufmann, 1991). This suggests that the costsof designing and implementing data integrationcould increase rapidly with the size and scopeof the effort.

Part of these diseconomies of scale may stemfrom the fact that as the size of a project in-creases, there is an increased difficulty in achiev-ing conceptual integrity across differentperspectives and different design requirements(Brooks, 1975). This is consistent with Daft andLengel’s (1986) assertion that when there is greatdifferentiation between subunits, much equivo-


Data Integration

cality must be reduced before common under-standings can be reached. If a system were tobe used by only a single subunit, or if all subunitsinvolved were identical in their needs, a commondata model could be designed with input fromonly one subunit. This would simplify the designproblem.

However, when multiple subunits with differingneeds are involved, the organization may face atrade-off between imposing a compromise solu-tion that does not meet all local needs or engag-ing in a more demanding design effort to trulyunderstand and meet each subunit’s uniqueneeds. Even if that ideal creative solution can befound, it will usually require better analysts, takelonger, and cost more than if fewer distinct re-quirements need to be met.

At Dobbs Insurance, even though the pensionsbusiness unit has integrated its own data with greatsuccess and many benefits, it is moving only veryslowly to consider integrating data with the grouphealth unit (a similar market and similar businessneeds) and has no plans to integrate with the lifeand casualty business units (a very different marketand very different business needs).

We propose that the number of subunits involvedand the heterogeneity of their data needs shouldmake a major difference in the cost of develop-ing an agreed-upon design for integrated data;disparate information requirements would com-plicate the design. Where there are numeroussubunits with quite different products, approach-es to marketing, or competitive challenges, efoforts at data integration could involve long anddifficult searches for acceptable compromises.

Proposition 3a: All other things being equal,as the number and heterogeneity ofsubunit information needs increase, thedifficulty of arriving at acceptable designcompromises increases, and the cost ofthe resulting design will increase morethan linearly. Thus, rational firms will in-tegrate less extensively when there aremany heterogeneous subunits involved.

What about long-term costs?Once integrated data has been put in place, firmsmust respond to turbulence in the organization’senvironment, with the resulting unanticipated in-

formation requirements. Where market condi-tions are rapidly changing, competitors are takingnon-routine actions, or new technologies arebecoming available that change products or pro-cesses, the need for flexibility will be high. If thechanges needed to the information systems arecompletely consistent with the existing datamodels and databases, modifications to thesesystems can be designed and implemented veryrapidly because all the data design work will havebeen already completed.

Unfortunately, there are many examples wherethe data needs of the organization change as wellas the business processes. When this happens,system designers must re-enter the difficult andtime-consuming process of redesigning (re-negotiating) a new data model that meets all par-ties’ changed needs. The more often the datamodel needs to be redesigned, the higher theaverage annual costs. Thus, an important ques-tion is whether business environments reallychange enough to require changes to theorganization’s data model, and how often thismight occur. Two examples demonstrate thatchanging business environments can indeedforce changes to the data model.

At Rolling Freight Incorporated, the business hasfor years been oriented around managing theequipment needed to continue smooth railroadoperations. Under this view of the business,customers and shipments were a secondary focus.With hard competition from trucking and otherrailroads, a new perspective is forming that putscustomers and shipments on the center stage. Thiswill require a major rethinking of the way in whichdata is conceptualized, captured, and managed.

In the pensions business unit of Dobbs Insurance,the customer that is the focus of marketing activi-ty has changed from an organization with manyemployees to the individual employee who makeshis or her own decisions about what companiesand what products he or she wishes to invest in.This has forced a major change in the data struc-ture needed by this business unit.

How often such changes occur might vary frombusiness to business and from one time in historyto another. Following Huber’s (1984) predictionabout the nature of the post-industrial environ-ment, we might guess that such changes will takeplace increasingly often. In more turbulent en-vironments, firms will be faced with frequentlyredesigning their data models. This will heightenthe average annual cost of information system


Data Integration

design and implementation, whether or not firmshave extensive data integration.

When subunits are fairly homogeneous and facethe same environments and same challenges, anintegrated data environment will probably be ad-vantageous because a single redesign will ad-dress all new needs. Further, it is clear that thealternative (subunit-specific data models) wouldinvolve redundant design efforts. Thus, for organ-izations with homogeneous subunits, turbulencemay make data integration more attractive.

On the other hand, when subunits are quite dif-ferent and the environment exhibits manydegrees of freedom, with different types of newchallenges facing each subunit, less data integra-tion and more local flexibility may be preferredso that local data models can be highly tuned tonuances in the local environment. If they did tryto integrate, the subunits’ heterogeneous needswould make each separate effort at redesigninga common data model quite difficult. In addition,even though only selected subunits would facechanges in their information needs, all subunitswould have to engage in the redesign effortbecause all would be affected by the changeddesign. Thus, organizations with heterogeneoussubunits and turbulent environments will be fre-quently faced with the short-term higher costs ofdata integration and will seldom have the abilityto enjoy its long-term benefits.

Proposition 3b: As organizations face greaterinstability in their environments andtheir information requirements, the im-portance of proposition 3a will increase.In turbulent environments, firms withmany heterogeneous subunits will beeven less likely to integrate extensive-ly, and firms with homogeneous sub-units will be more likely to integrateextensively.

Why the "Rational" Model IsOnly Part of the PictureThe model of the impact of data integrationshown in Figure 2 is consistent with a rationalperspective on organizations. Though muchunderstanding and guidance can come from therational model, there are some important com-plications that this model does not explicitlyaddress,

Interdependence with strategy andstructureIn an organization, the technologies employed,the structure, the strategy, individual roles, andmanagement processes are all tightly interdepen-dent (Leavitt, 1965; Rockart and Scott Morton,1984). No one of these can be changed withouthaving an impact on the others. Changes in dataintegration (a technology) may have powerful in-teractions with other organizational variables. Forexample, data integration might be a necessaryprerequisite for a shift in strategy or structure. Itmight create the potential for changes in infor-mation flows, thus affecting individual roles andorganizational structure. Decisions on the ap-propriate amount of data integration are intimate-ly tied to decisions about other key corporatevariables and to corporate strategy in general.

At the Van Buren Bank, after the IS departmentachieved en.ough data integration to providecustomer and profitability information, it was notimmediately clear that the account managerswould take advantage of the new information. Theculture of the bank, including incentive schemesand understandings about the way in whichbusiness was run, ran strongly against accountmanagers using masses of quantitative data intheir daily work. At first, only the administrativeassistants were heavy users of the system. Ac-count managers claimed the information was "notuseful."

In a practical sense, no organization has theresources to pursue all attractive technologicalpossibilities available to it. In this paper it hasbeen asserted that interdependence is the driv-ing force for data integration benefits, but in-terdependence that is not recognized by topmanagement, or that is not integral to topmanagement’s current business thrust, will prob-ably not have a large claim on a firm’s discre-tionary resources. It takes more than arecognition by an IS group that parts of a firm arehighly interdependent. Those data resourcemanagement efforts (and data integration efforts)that succeed are those driven by business re-quirements clearly understood and championedby top management (Goodhue, et al., 1988).

Power and pofiticsThe rational model in Figure 2 assumes thateveryone in an organization shares the sameoverall goals and can agree about how to value


Data Integration

various benefits. There is certainly enough prece-dent in research on organizations in general (e.g.,Pfeffer, 1981) and on the development of infor-mation systems in particular (e.g., Markus, 1983)to suggest that such a rational assessment ofcosts and benefits is incomplete. Even with largenet benefits to the organization as a whole, dataintegration may distribute those benefits andcosts in an uneven way, reducing the localautonomy of some divisions, changing the levelof access to critical information, or changing thepower balance in some other way.

The disparity between "who pays the costs" and"who gets the benefits" can lead to less than op-timum solutions for everyone involved, as Thornand Connolly (1987) found when they looked the motivation for subunits to contribute privatedata to a public "discretionary database." In ad-dition, even the possibility of changes in thepower balance can cause resistance to a data in-tegration effort by those concerned they mightlose out, regardless of arguments taking a totalorganizational benefit point of view. The ques-tion of who gets the benefits and who pays thecosts is not an idle one.

At Spectrum Electronics, each of nine differentplants had separate manufacturing systems, mak-ing it difficult to coordinate purchasing, schedul-ing, and spare parts inventories across plants.Over a period of more than five years, the VP ofmanufacturing, the VP of materials, and the direc-tor of MIS lobbied and positioned themselves tomove to a single set of integrated databases andcommon manufacturing systems to be used by allplants. A few plant managers resisted so stronglythat only their early retirement allowed plans to pro-ceed. Once implemented, the new integrateddatabases and common systems were believed tobe a major factor in Spectrum’s successful com-petition with its Asian rivals.

ConclusionWidespread data integration is an expensiveproposition. While much of the literature focuseson some attractive benefits, we have tried tobalance the view by looking at both positive andnegative impacts. The arguments for this newperspective have been primarily theoretical.Though anecdotal evidence was presented tobolster the plausibility of the theoretical proposi-tions, future empirical work will be needed tovalidate them. Figure 3 summarizes the proposed

impact of data integration on an organization.This new perspective should give managers fac-ing data integration decisions a more realisticbasis for understanding its impact.

The major implication of this analysis is that ingeneral it will not be cost-effective to integrateall of an organization’s data. If this is true,organizations will need help in their efforts to"partially integrate" to achieve the most impor-tant benefits and avoid the most burdensomecosts. For the MIS field to provide this help, weneed to "unbundle" our concept of data integra-tion. It is not an all-or-nothing choice, nor is it thepreferred approach. As researchers and practi-tioners, we need to change our thinking so wecan provide guidance to organizations on im-plementing "partial integration."

Evidence from our case studies gives some in-itial suggestions on several ways in which par-tial integration can be implemented. One obviousapproach is to limit the scope to only certainsubunits, as was done at Dobbs Insurance. Therethe pensions group implemented data integrationonly for its own business unit. Even across awider scope, there are at least three different par-tial integration approaches evident in the casestudies. One is to require all subunits to use aselection of "global" or organization-wide ap-plication systems (such as payroll, order entry,or purchasing), allowing them discretion on otherless critical application systems (Van Rensselaer,1985). These organization-wide applicationsystems include a common data model and, byimplication, commonly defined data.

Another approach is to develop selectedorganization-wide databases (such as thoserelated to customers or products) and require allapplication systems that use these entities to ac-cess and update the common databases. Anynon-common databases could be designed withlocal discretion. A third partial integration ap-proach is to identify a selection of critical dataelements and hammer out agreed-upon defini-tions for those across the entire organization.These standard definitions can then be enforcedin all systems development. Firms have ex-perimented with anywhere from 20 to a thousandsuch standard data elements.

Additional conceptual work and empiricalresearch is needed to suggest which of these orother partial integration approaches will be most


Data Integration

Greater Data Integration

In the Presence of: Will Tend to Lead to:Interdependence Better communications

between subunits Better coordination

Differentiationbetween subunits

More compromise or more design costs.More bureaucratic delay

Differentiation andenvironmental turbulence

Even more compromise, orEven more design costs

Figure 3. The Impact of Data Integration in Uncertain (as Opposed to Equivocal) Environments

effective, and why. A second and equally press-ing need is for methods that can help anorganization determine which data should be in-tegrated and which should not. Existing informa-tion engineering methodologies have overlookedthis question because they assumed that all datashould be integrated, though some other IS plan-ning approaches, such as critical success factors(Rockart, 1979), may be applicable for the par-tial integration problem. The choice presumablydepends heavily on the strategic direction of theorganization. Firms that attempt to integrate aninappropriate subset of the data, or on a widerscope and in more detail than is appropriategiven their organizational situation, will probablyface stiff resistance to either the large costinvolved or the loss of local flexibility ofheterogeneous subunits.

The emphasis in this paper has been on central-ly controlled data integration for large internal in-formation systems. However, the ideas presentedhere can be applied in either more macro or moremicro settings. For example, partners in elec-tronic data interchange clearly have interdepen-dence interests but must also be cognizant of theneeds for local flexibility, especially in the faceof industry turbulence. End-user computing at thework group level is a realm where local flexibilityis of paramount importance, but there are alsoreasons to impose some level of commonlanguage so that data can be shared between in-terdependent work groups. Thus, the sameissues of interdependence, need for local flex-ibility, and design costs clearly apply in these ad-ditional realms.

Though this paper has pointed out some negativeaspects to data integration, many organizationstoday are striving for a more global conscious-ness of their businesses and a more globalresponse to customers and markets. Huber(1984) suggests that managers in post-industrialorganizations will need to access more informa-tion about more aspects of the business, fromeven more parts of the organization, and inshorter time spans. This implies both greaterinterdependence and greater reliance oncomputer-based information. These types ofchanges in the organizational climate will pro-bably shift the balance toward the need forgreater (but not total) data integration in manyfirms, heightening the practical and academic im-portance of this area of research.

AcknowledgementsThe authors wish to express their appreciationto Judith Quillard for important contributions tothe ideas and presentation of this work. Theresearch was supported by the MIT Center forInformation Systems Research, the CarlsonSchool of Management, the Grant-in-Aid programof the University of Minnesota, and the IBM Pro-gram of Support for Education in the Manage-ment of Information Systems.

ReferencesAshby, W.R. An Introduction to Cybernetics,

Wiley and Sons, New York, NY, 1956.Banker, R. and Kaufmann, R.J. "Reuse and Pro-

308 MIS Quarterly/September 1992

Data Integration

ductivity in Integrated Computer Aided Soft-ware Engineering: An Empirical Study," MISQuarterly (15:3), September 1991, pp.375-401.

Banker, R. and Kemerer, C.F. "Scale Economiesin New Software Development," IEEE Trans-actions on Software Engineering (15:10), Oc-tober 1989, pp. 1199-1205.

Batini, C., Lenzerini, M., and Navathe, S.B. "AComparative Analysis of Methodologies forDatabase Schema Integration," ACM Com-puting Surveys (18:4), December 1986, pp.323-364.

Beath, C.M. "Supporting the InformationTechnology Champion," MIS Quarterly (15:3),September 1991, pp. 355-372.

Bonczek, R.H., Holsapple, C.W., and Whinston,A.B. "Aiding Decision Makers With aGeneralized Data Base Management System:An Application to Inventory Management,"Decision Sciences (9:2), 1978, pp. 228-245.

Brooks, F.P. The Mythical Man-Month: Essayson Software Productivity, Addison-Wesley,Reading, MA, 1975.

Chen, P.P. "The Entity-Relationship Model--Toward a Unified View of Data," ACM Trans-actions on Database Systems (1:1), March1976, pp. 9-36.

Conant, R.C. and Ashby, W.R. "Every GoodRegulator of a System Must be a Model ofThat System," International Journal ofSystems Sciences (1:2), 1970, pp. 89-97.

Daft, R.L. and Lengel, R.H. "Organizational In-formation Requirements, Media Richness,and Structural Design," Management Science(32:5), May 1986, pp.. 554-571.

Date, C.J. An Introduction to Database Systems,Addison-Wesley, Reading, MA, 1981.

Emery, J.C. "Cost/Benefit Analysis of Informa-tion Systems," in Advanced System Develop-ment/Feasibility Techniques, J.D. Couger,M.A. Colter, and R.W. Knapp (eds.), Wiley,New York, NY, 1982.

Finkelstein, C. An Introduction to InformationEngineering, Addison-Wesley, Sydney,Australia, 1990.

Galbraith, J. Designing Complex Organizations,Addison-Wesley, Reading, MA, 1973.

Goodhue, D.L., Quillard, J.A., and Rockart, J.F."Managing the Data Resource: A Contingen-cy Approach," MIS Quarterly (12:3),September 1988, pp. 373-392.

Goodhue, D.L., Kirsch, L.J., Quilllard, J.A., and

Wybo, M.D. "Strategic Data Planning:Lessons from the Field," M/S Quarterly (16:1),March 1992, pp. 11-34.

Gupta, A.K. and Govindarajan, V. "ResourceSharing Among SBU’s: Strategic Antecedentsand Administrative Implications," Academy ofManagement Journal (29:4), December 1986,pp. 695-714.

Heimbigner, D. and McLeod, D. "A FederatedArchitecture for Information Management,"ACM Transactions on Office InformationSystems (3:3), July 1985, pp. 253-278.

Hoffer, J.A., Michaele, S.J., and Carroll, J.J. "ThePitfalls of Strategic Data and Systems Plan-ning," Proceedings of the Twenty-Second An-nual Hawaii International Conference onSystems Sciences, 1989, pp. 348-356.

Huber, G. "Organizational Information Systems:Determinants of Their Performance andBehavior," Management Science (28:2),February 1982, pp. 138-155.

Huber, Go "The Nature and Design of Post-Industrial Organizations," ManagementScience (30:8), August 1984, pp. 928-951.

Huber, G. "A Theory of the Effects of AdvancedInformation Technologies on OrganizationalDesign, Intelligence, and Decision Making,"Academy of Management Review (15:1),1990, pp. 47-71.

IBM. Business Systems Planning, IBM Manual#GE20-0527-3, White Plains, NY, July 1981.

King, J.L. "Centralized versus DecentralizedComputing: Organizational Considerationsand Management Options," ACM ComputingSurveys (15:4), December 1983, pp. 319-349.

Leavitt, H.J. "Applied Organizational Change inIndustry," in Handbook of Organizations, J.March (ed.), Rand McNally, Chicago, IL, 1965,pp. 1144-1170.

Lederer, A.L. and Sethi, V. "The Implementationof Strategic Information Systems PlanningMethodologies," MIS Quarterly (12:3),September 1988, pp. 445-461.

Lederer, A.L. and Sethi, V. "Critical Dimensionsof Strategic Information Systems Planning,"Decision Sciences (22:1), Winter 1991, pp.104-119.

Leveson, N.G. and Wasserman, A.I. "LogicalDecentralization and Semantic Integrity in aDistributed Information System," in Distribu-ted Data Sharing Systems, R.P. van de Rietand W. Litwin (eds.), North Holland, Amster-dam, 1982, pp. 243-253.


Data Integration

Litwin; W. and Abdellatif, A. "Multidatabase In-teroperability," IEEE Computer, December1986, pp. 10-18.

Litwin, W., Mark, L., and Roussopoulos, N. "ln-teroperability of Multiple AutonomousDatabases," ACM Computing Surveys (22:3),September 1990, pp. 267-293.

Malone, T.W. "Modeling Coordination inOrganizations and Markets," ManagementScience (33:10), October 1987, pp.1317-1332.

Malone, T.W., Yates, J., and Benjamin, R.I."Electronic Markets and Electronic Hierar-chies," Communications of the ACM (30:6),June 1987, pp. 484-497.

Markus, M.L. "Power, Politics and MIS Im-plementation," Communications of the ACM(26:6), June 1983, pp. 430-444.

Marschak, J. "Remarks on the Economics of In-formation," in Contributions to ScientificResearch in Management, University ofCalifornia, Los Angeles, CA, 1959.

Marschak, J. "Economics in Inquiring, Com-municating, Deciding," American EconomicReview (58:2), May 1968, pp. 1-18.

Marschak, J. "Economics of InformationSystems," Journal of the American StatisticalAssociation (66), March 1971, pp. 192-219.

Martin, J. Strategic Data-Planning Methodolo-gies, Prentice Hall, Englewood Cliffs, N J,1982.

Martin, J. Information Engineering, SavantResearch Studies, Carnforth, Lancashire,England, 1986.

McCann, J. and Ferry, D. "An Approach forAssessing and Managing Inter-Unit In-terdependence," Academy of ManagementReview (4:1), January 1979, pp. 113-119.

McCarthy, W.E. "The REA Accounting Model: AGeneralized Framework for AccountingSystems in a Shared Data Environment," TheAccounting Review (57:3), July 1982, pp.554-578.

Mendelson, H. and Saharia, A.S. "Incomplete In-formation Costs and Database Design," ACMTransactions on Database Systems (11:2),June 1986, pp. 159-185.

Page-Jones, M. The Practical Guide to Struc-tured Systems Design, Yourdon Press, NewYork, NY, 1980.

Pfeffer, J. Power in Organizations, BallingerPublishing Company, Cambridge, MA, 1981.

Rockart, J.F. "Chief Executives Define Their Own

Data Needs," Harvard Business Review(57:2), March-April 1979, pp. 81-93.

Rockart, J.F. and Scott Morton, M.S. "Implica-tions of Changes in Information Technologyfor Corporate Strategy," Interfaces (14:1),January-February 1984, pp. 84-95.

Sheth, A.P. and Larson, J.A. "FederatedDatabase Systems for Managing Distributed,Heterogeneous, and Autonomous Data-bases," ACM Computing Surveys (22:3),September 1990, pp. 184-236.

Simon, H.A. "Applying Information TechnologyTo Organization Design," Public Administra-tion Review (33:3), May/June 1973, pp.268-278.

Simon, H.A. The Sciences of the Artificial, MITPress, Cambridge, MA, 1981.

Thompson, J.D. Organizations in Action,McGraw-Hill, New York, NY, 1967.

Thorn, B.K. and Connolly, T. "Discretionary DataBases: A Theory and Some ExperimentalFindings," Communication Research (14:5),October 1987, pp, 512-528.

Tushman, M. and Nadler, D. "Information Pro-cessing as an Integrating Concept inOrganizational Design," Academy of Manage-ment Review (3:3), July 1978, pp. 613-624.

Van Rennselaer, C. "Global, Shared, Local,"Datamation, March 15, 1985, pp. 105-114.

Weick, K.E. "Educational Organizations asLoosely Coupled Systems," AdministrativeScience Quarterly (21:1), March 1976, pp.1-19.

About the Authors

Dale L. Goodhue is assistant professor of MISat the University of Minnesota’s Carlson Schoolof Management. He has worked as a businessanalyst for American Management Systems, andreceived a B.S. from Brown, an M.B.A. fromCarnegie Mellon, and a Ph.D. from M.I.T. He iscontinuing to study data management by con-ducting case studies on planning methods, dataarchitectures, and organizational costs andbenefits of data integration. Other research in-terests include measuring the impact of systemcharacteristics, policies, and design methods onuser evaluations of IS.

Michael D. Wybo is assistant professor of MISon the Faculty of Management at McGill Univer-sity. He received his Ph.D. from the University


Data Integration

of Minnesota and has consulted for a number ofyears on systems development projects in WestAfrica. His research interests include the impactsof alternative data management strategies on firmperformance and the role of information systemsin the design of organizations.

Laurie J. Kirsch is a doctoral candidate in MISat the University of Minnesota. Her research in-terests include management of systems develop-ment projects and data management. She has10 years of experience in the information systemsfield.


The Impact of Data Integration on the Costs and Benefits of … · 2016-01-29 · Data Integration...

Documents

Transcript of The Impact of Data Integration on the Costs and Benefits of … · 2016-01-29 · Data Integration...