0018-9162/99/$10.00 © 1999 IEEE March 1999 31

Computing Practices

Beyond Spreadsheets:Tools for BuildingDecision SupportSystems

Decision support systems (DSSs) aresoftware products that help users applyanalytical and scientific methods todecision making. They work by usingmodels and algorithms from disciplines

such as decision analysis, mathematical programmingand optimization, stochastic modeling, simulation,and logic modeling. DSS products can execute, inter-pret, visualize, and interactively analyze these mod-els over multiple scenarios.

In recent years, the growing popularity of onlineanalytical processing, data warehousing, and supplychain management has led to an increased interest inthe development of decision support systems. Decisionsupport tools could assist decision-makers in problemsinvolving risk management, the allocation of scarceresources, and the need to balance conflicting objec-tives. When well implemented and used wisely, DSSscan significantly improve the quality of an organiza-tion’s decision making.

However, building a DSS requires significant exper-tise in decision analysis, programming, and user inter-face design. A DSS can also need to connect in realtime with other enterprise applications, further com-plicating the task. For these reasons, the broad use ofDSSs—compared with, say, database management sys-tems—has not occurred.

What’s missing is sophisticated tool support for devel-oping DSSs. One current solution is a DSS generator,1

a kit or environment for developing an application-spe-cific DSS. DSS generators provide tools that make it eas-

ier and faster to develop models, data, and user inter-faces that are customized to the application’s require-ments. Using a DSS generator reduces DSS developmentto a decision analysis task—which requires expertise indecision analysis and mathematical modeling—ratherthan a programming task. DSS generators are crucialto the success of DSSs in practice.

In this article, we describe the state of the art in DSSgenerator software, specifically in the realm of decisionanalysis methods (see the “Decision Analysis” sidebar).2

Decision analysis techniques account for the uncertain,dynamic, and multicriteria aspects of decisions.Essentially, they aid the evaluation of alternatives in theface of trade-offs. Well-known decision analysis meth-ods include decision trees and influence diagrams. Herewe briefly describe the features of 11 commerciallyavailable DSS generators that specialize in decision

Com

putin

g Pr

actic

es

The complexity and long development time inherent in building decision

support systems has thus far prevented their wide use. A new class of

tools, DSS generators, seeks to cut the lead time between development

and deployment.

Hemant K.BhargavaCarnegie MellonUniversity

Suresh Sridhari2 Technologies

Craig HerrickUS Navy

.

32 Computer

analysis. Although not a comprehensive, completeanalysis of these tools, this article clarifies the idea ofDSS generators as DSS development environments andpresents an overview of the progress in this area.

WHAT IS A DSS GENERATOR?The word “generator” is somewhat of a misnomer.

Although DSS generators are indeed used to developDSS applications for specific decision problems, theydo not behave like a typical code generator (whichautomatically generates programming code from a setof requirements specifications), or a software devel-opment environment. Programmers use a develop-ment environment to write and then compile sourcecode. The resulting stand-alone, runtime object canbe executed without the development environment.

Using a DSS generator, in contrast, is more akin tousing an interpreted programming language—you mustalways have the interpreter to execute the code. Asanother example, consider a spreadsheet as an appli-cation development environment. Application devel-opers use certain spreadsheet features to create

mathematical models, data structures, macros, inputforms, and other objects. However, spreadsheets don’thave compilers and don’t produce runtime objects.Users must have the spreadsheet system to execute theapplication.

As an application development environment, a DSSgenerator is more like a spreadsheet than a C++ devel-opment environment. Developers work with a DSSgenerator, which provides certain generic functions(discussed later). DSS developers can use these func-tions to build models and data for the application,speeding up the DSS development process. However,DSS generators don’t produce runtime objects, andthe applications they generate can be executed only ifthe user has the generator program as well.

Therefore, a DSS generator is a code generator inonly a limited sense: It provides a visual interface formodel development and data definition. A DSS gen-erator transforms the resulting visual objects into codein a lower level model-definition language. This letsDSS developers focus on decision analysis and modeldevelopment rather than programming constructs—

Decision AnalysisDecision analysis consists of theories, processes, and analytical

methods for decision making that account for the uncertain,dynamic, and multicriteria aspects of decisions. Well-known deci-sion analysis methods include decision trees, influence diagrams,multi-attribute utility models, and the analytic hierarchy process.Figure A shows samples of the first two, the most common meth-ods employed in decision analysis. These methods allow decision-makers to evaluate decision alternatives in terms of the trade-offsbetween multiple decision criteria as well as the risk and uncer-tainties associated with each alternative.

A typical decision analysis problem that involves uncertaintyis the harvesting problem. In this example, a farmer must decidewhen to harvest a crop of, say, tomatoes, to maximize his earn-

ings. An early harvest would realize the current crop but fetch alower unit price. A late harvest of vine-ripened tomatoes is risky:It might fetch a higher unit price and even a higher yield, but someor all of the crop could be lost to frost.

The harvesting problem’s structure and details can be modeledas an influence diagram or as a decision tree. We can transformmost influence diagrams into decision trees. A decision tree rep-resentation, however, loses the conditional probability relation-ship between variables and can quickly become very big. Influencediagrams are compact and preserve logical relationships but areunsuitable for asymmetric problems—problems whose structuredoes not branch out symmetrically. Asymmetric problems com-prise many useful decision analysis problems in the real world.

When toharvest?Decision

Frost

1,000 lbs at $0.50/lb

300 lbs at $1/lb

Outcome

Earnings

$500sure

$300(60 percent

chance)

1,200 lbs at $1/lb $1,200(20 percent

chance)

100 lbs at $1/lb $100(20 percent

chance)

Wait

Now

None

Min

or

Major

Yieldquantity

Yieldquality

Harvestdate

Decision

Price

Frost

Event

Earnings

Outcome

Figure A. Sample (1) decision tree and (2) influence diagrams, components of decision analysis.

(1) (2)

.

data structures, functions, procedures—and the inter-action among them.

Required functionalitySo what does a DSS generator need to do? Its func-

tionality depends on what we expect the DSS to do. ADSS should have the necessary set of models, data,and user interfaces for the application in question. Itmust include the necessary input and output repre-sentations and control mechanisms. Thus, a DSS gen-erator should encompass

• a modeling paradigm, that is, specific methodsfor solving decision problems;

• model and data definition features, for definingmodel schemas and for entering data;

• model and data management features, includinga specialized mechanism for maintaining, stor-ing, retrieving, and executing models and fordefining and manipulating data;

• analytical methods and visual representations,for exploring relationships between outcomevariables and various controllable variables(including decision variables and user prefer-ences) and/or uncontrollable variables that influ-ence outcomes;

• user interfaces, either general-purpose and/or thecapability to create a specialized user interface;and

• model and data interchange features, for com-municating with external programs such as data-bases or external user interfaces.

Commercial productsDSS generators were primarily a theoretical con-

cept in the 1980s, but several commercial productshave emerged in the past few years. These products

provide a fairly comprehensive set of DSS features. Welimit this discussion to products that apply decisionanalysis methods; an earlier survey3 provides a morecomplete list of systems.

These products let a developer build a DSS appli-cation simply by specifying the necessary models anddata. They offer a visual and/or textual language forbuilding model schemas, features for model solutionand analysis, and commands and representations forvisualizing model results. They minimize developmenteffort by offering a generic (but usually inextensible)graphical user interface, generic (but again, inexten-sible) data management features, and generalized solu-tion algorithms and analysis tools.

Using these products, someone who knows the rel-evant modeling paradigm and the problem domaincan develop a DSS application in a few hours ordays—a significantly shorter time than previously pos-sible. Such DSS generators are now commerciallyavailable.

We evaluated the 11 DSS generators listed in Table 1,comparing the decision analysis methods they supportand their strengths and limitations. All of them arebased on one or more decision analysis methods.

EVALUATING DSS GENERATORSWhen we discuss the capabilities, strengths, and lim-

itations of specific products, we do so in an anecdotalsense. Our aim is to give readers concrete examplesand a flavor for the features found in various systems.We organize the discussion according to DSS func-tionality.

Modeling paradigmDecision analysis products employ two major cat-

egories of methods for modeling and solving decisionproblems.

March 1999 33

Table 1. DSS generators considered for evaluation.

DSS generator Vendor name URL

Aliah Think! Aliah Inc. http://www.aliah.comAnalytica Lumina Decision Systems Inc. http://www.lumina.comCriterium Decision Plus InfoHarvest Inc. http://www.infoharvest.comDecide Right for Windows Avantos Performance Systems Inc. Not availableDecision Analysis by TreeAge (DATA) TreeAge Software Inc. http://www.treeage.comDecision Pro Vanguard Software http://www.vanguardsw.comDecision Programming Language (DPL) Applied Decision Analysis http://www.adainc.comExpert Choice Professional Expert Choice Inc. http://www.expertchoice.comLogical Decisions for Windows (LDW) Logical Decisions http://www.logicaldecisions.comPrecisionTree Palisade Corp. http://www.palisade.comWhich & Why Arlington Software Corp. http://www.arlingsoft.com

.

34 Computer

The first category—decision making under uncer-tainty—includes methods for dealing with uncertaintyabout data values, user preferences, and other aspectsabout the decision problem, and for analyzing deci-sion alternatives in terms of both return and risk. Thesemethods include decision trees (employed by DecisionPro and PrecisionTree), influence diagrams (Analytica),or some proprietary variation of these. DecisionAnalysis by TreeAge (DATA) allows users to develop

a general influence diagram and convert it to a detaileddecision tree. Decision Programming Language (DPL)provides users a way to represent problems as a hybridof decision trees and influence diagrams.

The second category—multicriteria decision mak-ing (MCDM)—includes methods for making trade-offs between multiple decision objectives, and foranalyzing relationships between the ranking of deci-sion alternatives and importance of different decisionobjectives. MCDM modeling can rely on traditionalmethods, which include multi-attribute utility theoryand analytical hierarchy process. Aliah Think!,Criterium Decision Plus, Expert Choice Professional,and Logical Decisions for Windows (LDW) employthese traditional methods. Products like Decide Rightand Which & Why use proprietary variations of them.

Model and data definitionBuilding decision models and specifying the necessary

data is the most important aspect of using a DSS gener-ator. Since the DSS generator applies its remaining func-tions (such as sensitivity analyses and the user interfacemechanisms for using them) to the model schema, thesefunctions become available without further cost oncethe model has been specified. Not having to providethese functions frees developers to concentrate on devel-oping the application’s model schemas.

Model schema definition. Again, the tools can bedivided along the lines of the paradigm they employ:decision making under uncertainty and MCDM.

All modern products that employ decision makingunder uncertainty offer a visual modeling languagefor model development. This means that model build-ing involves manipulating icons (for different types ofmodel nodes), edges (for connecting nodes), andobjects from drop-down menus. The capabilities thatDSS generators provide are extremely useful: Theymake model development easy and also minimize typ-ing errors.

In Analytica, DSS developers can create an influ-ence diagram model by defining nodes, connectingthem, and entering formulas in the appropriate boxes.To help in entering formulas, Analytica makes rele-vant variables and mathematical functions available indrop-down lists, as shown in Figure 1a.

Similarly, DATA lets DSS developers select built-inicons that represent decision nodes, chance nodes, andso on. PrecisionTree allows developers to build a deci-sion tree graphically. It also provides a spreadsheetinterface for viewing and manipulating data.

Products that support multicriteria decision mak-ing also provide a visual modeling language for modeldevelopment. Initially, DSS developers visually set upthe hierarchy of criteria. Then they can use severalvisual tools and representations to rate the alterna-tives against the criteria and enter the criteria weights.

Figure 1. Model development and data input in (a) Analytica and (b) CriteriumDecision Plus. In Analytica, DSS developers can drag suitable icons from the toolbarand manipulate them to build an influence diagram model. In Criterium Decision Plus,DSS developers can construct the hierarchy via graphical icons and then evaluatethem via different mechanisms provided on menus.

(a)

(b)

.

The interface in Criterium Decision Plus, shown inFigure 1b, provides icons that DSS developers use toconstruct hierarchies. The generator automaticallydefines relevant mathematical variables and programelements such as data structures. The DSS generatorsautomatically derive mathematical relationships andequations involving these criteria and alternatives.

All products help develop a fairly realistic and com-plex decision model within hours. They differ in themodeling paradigm they support and their modelinglanguage. Here, a key differentiator is support for analgebraic modeling device called indexing.4 Indexinglets DSS developers group a collection of objects intoa class (an index or an array), define variables andmathematical relationships that apply to all index ele-ments, and declare aggregation functions (such assummation) over elements of an index variable.

Indexing structures4 are very useful in mathemati-cal modeling, and their absence in most DSS generatorsis an important limitation. However, Analytica doesexplicitly support indexing in user-defined models.

Data definition. MCDM problems require decision-makers to rate the alternatives for each criterion. Thusa DSS generator requires a user to define this data insome fashion.

In some cases, these ratings are subjective and canbe entered directly. Figure 2 shows an example fromExpert Choice Professional. A DSS generator can alsoindirectly compute a rating via a user-specified pair-wise comparison of alternatives (as in analytical hier-archy processing). In other cases, ratings may bederived as a function of available raw data values anda user-specified utility function.

Certain MCDM products (Decide Right, Which &Why, and Expert Choice Professional) accept onlysubjective user ratings. This restriction forces a userto implicitly determine a utility function, a limitationwhen measures are well defined and raw data is avail-able. Here, LDW is unique in that it supports bothmethods as well as other multi-attribute functions, rat-ing techniques, and elicitation tools (those that elicituser preferences).

To account for uncertainty in the decision problem,users can enter data values as single data points, tables,or as probabilistic distributions. (Commercial DSSgenerators support most parametric distributions.)The decision tree programs and a few of the MCDMprograms accept data in this way. Analytica is dis-tinctive in its use of multidimensional tables for dataentry, display, and visualization.

From a user interface perspective, most productsoffer visual, direct-manipulation interfaces for enter-ing model data such as probabilities, decision alter-natives, utilities, and other values. Users typicallyaccess input forms for data by navigating the corre-sponding model schema element. Figure 1 and Figure

March 1999 35

Figure 2. Data input in Expert Choice Professional. Once thedecision model is set up, users can choose several methodsfor weighting criteria (or for grading alternatives): a (a) matrixmethod, (b) graphical view, and (c) textual method.

(a)

(b)

(c)

.

36 Computer

2 both show samples of these input forms. The draw-back of this approach is that it fails to give the DSSdeveloper a comprehensive view of the data.PrecisionTree, a spreadsheet add-in, is unique in thisregard because it offers the familiar spreadsheet inter-face for data entry and manipulation.

Model and data managementMost commercial products work with one model

of the decision problem at a time. Essentially, eachmodel corresponds to a file or more than one file tostore relationships, data, and so on. In this way, mostproducts leave model management to the operatingsystem, since only one model can be used at a time.Most don’t offer additional functionality.

Analytica, however, is one exception. Users canbuild models by integrating existing models as mod-ules in a “super” model. Several models can thus reusethe same modules.

Data management features within these productsare fairly straightforward, and certainly less powerfulthan in database management systems. Data man-

agement in DSS generators doesn’t enforce databasedesign principles such as normalization, for example.In most cases, once data is entered, it is difficult toquery or extract subsets of the data.

Analytical methods and visual representationEffective use of models involves executing them

repeatedly under different data scenarios and thenanalyzing the sensitivity of the results from an indi-vidual execution. The DSS generators discussed hereoffer a rich collection of commands and generic rep-resentations for analysis.

Basic output representations and dealing with uncer-tainty. In decision tree models, the basic output iden-tifies the recommended decision branch and expectedvalues (or other chosen measure) for each alternative.For example, in DATA, users simply click a “rollback”button to mark the superior decision branch on thetree (see Figure 3a). Users can deal with the uncer-tainty in input variables by using sensitivity analysis—Figure 3b displays an analysis of the probability ofmarket behavior. Users can also have the application

Figure 3. Decision treeanalysis in DATA:Results of (a) tree roll-back and (b) sensitivityanalysis.

Figure 4. Different views of output in MCDM systems: In the (a) stacked bar chart (from LDW), each bar represents the score ofan alternative, and stacks show the contribution from various criteria; (b) a spider chart (Aliah Think!) displays two or morealternatives based on various criteria.

(a) (b)

(a) (b)

.

distribute values probabilistically. Average-, worst-,and best-case results are available. Again, once a DSSdeveloper specifies the model schema, these capabili-ties automatically become available to users of the spe-cific DSS.

Programs for multicriteria problems typically dis-play results as a bar chart showing the relative totalscores for decision alternatives. A stacked bar chartshows, for each alternative, the contribution from eachcriterion. Scatter diagrams and spider diagrams pro-vide additional formats for users in reviewing output.Figure 4 shows samples of two of these types of output.

Facilitating insight. An important purpose of deci-sion support systems is to help users gain insight intothe relationships among variables in the problem.Users must understand how changes in various con-trollable and/or uncontrollable variables influence thevalues of outcome variables.

Figure 5 displays two examples involving sensitiv-ity analysis. In Figure 5a, LDW depicts the sensitivityof the final score to the weight of the price criterion.The graphic also displays the rankings of the alterna-tives and the cutoff points where these rankingschange.

Figure 5b helps explain the way in which ExpertChoice Professional provides dynamic sensitivityanalysis: As users adjust the criteria weights (by mov-ing the sliding teal bars at left), the software automat-ically updates the total scores of the alternatives (right).This gives users instant and graphical feedback on therelationship between final scores and criteria weights.

Finally in this category, we consider the output rep-resentation that Decide Right uses. As shown in Figure6, Decide Right displays grades for each alternative foreach criterion; color helps users quickly spot the com-parative strengths and limitations of each alternative.

Not all people are comfortable with numbers andgraphs or gleaning insight from several seeminglyindependent output representations. Textual out-puts—interpretations of the numbers rather than meretranslations into graphs or text—would be valuableto many users. Besides, in many situations, textualoutput is more portable than a collection of figures.Decide Right is distinctive in its attempt to translateand interpret quantitative analysis results into a tex-tual representation.

User interfacesThe interfaces of the DSS generators we reviewed

varied greatly, though most are relatively easy to learnand use. In most cases, however, and especially forcomplex problems, user interface design shouldaccount for at least two categories of DSS users:

• the DSS developer or analyst, who uses the gen-erator to build a specific DSS; and

• the user or decision-maker, who must use the spe-cific DSS created by the DSS developer.

You might expect user interface requirements—therepresentations, tools, commands, functions, and soon—to differ for these two categories of users. In addi-tion, for many applications it can be desirable to builda user flow (a sequence of operations) into the userinterface. This flow would guide the user at each step,rather than allowing only a laissez-faire approach tosystem use.

Unfortunately, most products do not distinguishbetween these two user roles and instead provide ageneral-purpose, but inflexible, user interface. DSSdevelopers may find the user interface unsatisfactoryfor some applications. If they wish to extend it or toimpose a certain user flow, they have no means ofdoing so, unless the product offers a good macro-programming language or can interface with externalprograms. Even then, imposing a user flow is usuallyquite difficult. As a result, commercial tools oftenchoose to have users interact with the system via thesame interface as the DSS developer. Not surprisingly,this interface can be too complex for users.

March 1999 37

Best

Worst

Utility

0 100

Percent of Weight on Price Measure

Dodge Ram-50Chevy S-10MitsubishiFord RangerToyotaDodge Dakota

Figure 5. LDWprovides (a) sensitiv-ity analysis, andExpert Choice Profes-sional provides (b)dynamic sensitivityanalysis.

(b)

(a)

.

38 Computer

A few products have modest tools for creating a sep-arate user interface for users. DATA, for example, letsdevelopers provide a custom interface. Such interfacescan contain a few buttons that users can use to enterdata or to choose analysis types and graphs. Thus,users do not have to learn the complete (usually com-plex) interface provided for DSS developers.

Analytica offers a simple means of generating a verybasic interface containing one-click buttons to obtainresults and graphs. Also, products such as Preci-sionTree operate within a spreadsheet, giving users aninterface similar to the familiar spreadsheet user inter-face. In such a spreadsheet-coupled product, DSSdevelopers also have access to all the usual spread-sheet macros and user interface design tools. Theseare significant advantages in creating custom inter-faces for a specific application.

Model and data interchangeIt is generally not possible to export model schemas

from one product to another. This is not surprisingsince there are no standard formats for storing modelschemas for different decision analysis model types.Some products do offer a form of version support.DATA and Decide Right, for instance, allow users tosave different versions of a model and then recall themlater. In most products, users must use the “save as”command to save different versions of the model asdifferent file names.

On the other hand, most products can interchangedata. Most DSS generators we reviewed couldimport and export data as well as create ASCII datafiles. Some also connected to popular desktop prod-ucts such as spreadsheets and databases. These capa-bilities are useful, given the limited data design andmanagement features in the DSS generators them-selves.

FUTURE DIRECTIONSFuture decision support systems will probably sup-

port group decision making and will handle systemcomponents that are distributed over a client-server net-work. Only a few commercial products now supportgroup decision making: Which & Why and ExpertChoice Professional (via an extension called TeamExpert Choice) let multiple users rank their preferencesand combine these preferences into a group decision.These products also let groups assign weights to dif-ferent decision-makers. Group features can be desir-able, especially in higher level decision making, wheredecisions are frequently made by group consensus.

New client-server and distributed computing tech-nologies on the Web can aid the remote access of DSSgenerators, providing decision support on demand.5,6

Here, the user’s interface is a Web browser (the client),and separate components on a remote Web server pro-vide data and model management features. Web-enabling a DSS allows a company to share complexand expensive DSS technologies at negligible extracost, without customization for specific operating plat-forms. A few DSS generators already offer some formof Web-based interface and most vendors are payingmore attention to this area.

Maturing DSS technologies and decisionanalysis methods have led to the develop-ment of DSS generators that are both

usable and powerful. Yet the ultimate deployment ofDSSs will depend on the ability of users to use themto their advantage. Clearly the needs of these usersare important, so we are investigating this issue andhave published preliminary results.7,8 Overall, wehope this article will spur the interest of IS managers,DSS vendors, researchers, and users in the use anddevelopment of DSSs and DSS generators. ❖

Figure 6. DecideRight provides thiswindow to help usersunderstand compar-ative strengths andweaknesses of alter-natives. Color letsdecision-makers seethat the “State Uni-versity” is a strongchoice, weak only onthe “small college”criterion.

.

References1. R. Sprague, “A Framework for the Development of Deci-

sion Support Systems,” MIS Quarterly, June 1980, pp.1-26.

2. A.L. Golub, Decision Analysis: An Integrated Approach,John Wiley & Sons, New York, 1997.

3. D. Buede, “Aiding Insight III,” OR/MS Today, Aug.1996, pp. 73-79.

4. A.M. Geoffrion, “Indexing in Modeling Languages forMathematical Programming,” Management Science,Mar. 1992, pp. 325-344.

5. H.K. Bhargava, R. Krishnan, and R. Muller, “DecisionSupport on Demand: Emerging Electronic Markets forDecision Technologies,” Decision Support Systems, Vol.19, 1997, pp. 193-214.

6. S. Sridhar, “Decision Support Using the Intranet,” Deci-sion Support Systems, Vol. 23, 1998, pp. 19-28.

7. C.L. Herrick, “A Survey of Software for Decision Analy-sis,” master’s thesis, Naval Postgraduate School, Mon-terey, Calif., Mar. 1997.

8. H.K. Bhargava, C.L. Herrick, and S. Sridhar, “DesirableFeatures for Decision Analysis Software,” Proc. FourthConf. Int’l Soc. for Decision Support Systems, Int’l Soc.for Decision Support Systems, pp. 41-45.

Hemant K. Bhargava is a visiting professor at the HeinzSchool of Public Policy and Management, CarnegieMellon University. He is on temporary assignmentaway from his position as an associate professor ofinformation technology at the Naval PostgraduateSchool, Monterey, Calif. His research interests includedecision support systems, e-commerce, and Internet-based electronic markets for decision technologies.Bhargava has a PhD in decision sciences from TheWharton School, University of Pennsylvania. He is amember of the IEEE Computer Society, the Institutefor Operations Research and the Management Sciences(INFORMS), and INFORMS’ Computing Society.

Suresh Sridhar is a consultant at i2 Technologies Inc.His research interests include decision support sys-tems and distributed computing. Sridhar has a PhDdegree in management information systems from Van-derbilt University. He is a member of the IEEE Com-puter Society and the ACM.

Craig Herrick is a lieutenant commander in the US Navy.His research interests include decision support and exec-utive information systems. Herrick has an MS degree ininformation technology from the Naval PostgraduateSchool. He is a member of the US Navy Supply Corps.

Contact Bhargava at Carnegie Mellon University, TheHeinz School of Public Policy and Management,2109B, Hamburg Hall, Pittsburgh, PA 15213; bhargava@computer.org.

39

.