modeled to bits: decision models for the digital,networked economy

8/13/2019 Modeled to bits: Decision models for the digital,networked economy

1/9

International Journal of

Research in

Intern. J of Research in Marketing 17 (2000) 227-235arketing

www elsevier com/locate/ijresmar

Modeled to bits: Decision models for the digital,networked economy

Gary L. Lilien* i n d RangaswamyThe Smeal College o Business Adminisimtion Pennsyluunia State Uniuersify 402 Business Adminishation Building Uniuersity Pork

PA 16802-3004 USA

Abstract

Leeflang and Wittink [Internal. 1. Res. Marketing 17 (2000) 1051 sketch a future for marketing modeling that differsprimarily in scale and scope from today s environment. We have a different vision: the digital networked economy willinduce significant structural changes in (a) how models are developed and deployed (b) who uses marketing models, and (c)what types of models are developed. To be successful, marketing modelers must adapt by gaining a better understanding ofthe role of marketing modeling in the new environment and by learning how to use emerging information technologies (IT)for developing, deploying, and validating marketing models. Q2000 Elsevier Science B.V. All rights resewed.

Keywords: Marketing modeling; Decision models; Networked economy

1 Introduction

Leeflang and Wittink (2000) provide a usefulhistorical perspective on the field of marketing deci-sion models and offer their vision of developmentsto come. We expand on their discussion of how theInternet and, more specifically, the World Wide Web(WWW), will influence the development, deploy-ment, and use of marketing decision models. TheInternet is bringing about major changes in howbusinesses are conceived and managed, ushering in

the era of e-business. ur goal here is to articulate

Corresponding author. Tel.: 11-814-863-2782; fax 11-814-863-0413.

E-moil address: [email protected] G.L. Lilien).

how model deployment and use are changing be-cause of the Internet: How will the Internet changethe environment fo r marketing modeling? And, whatshould marketing modelers do to better adapt to thisenvironment?

We fust describe three important ways the Inter-net influences marketing modeling: 1) It de-couplesmodel, data, and the user interface, unlike aditionalmodels that integrate these components. (2) It ex-pands the reach of marketing models to a muchbroader, more heterogeneous group of users. 3) Itvastly increases the opportunities for gathering andusing data, information, and insights to support deci-sion-making. We then sketch the new challengesfacing model developers working in the Internetenvironment, and the new kinds of thinking theymust embrace to encourage wider and more effectiveuse of their models.

0167-81 16/0 0/ see front matter 0 2 0 0 0 Elscv~er cicnce B.V. All rights resewedPI[: S O l 6 7 - 8 1 1 6 ( 0 0 )0 0 0 2 1 - 5


2/9


3/9

G.L. Lilien, A. Rnngaswmy/ lnfern . ofRrsenrch in Marketing 17 2000) 227-235 229

aab

Embedded Models( Models Inside )

Standalone Integrated Systems

1)STANDALONE MODELS

Example: Conjoint Analysis Tools(www.sawtooth.com)

Example: Marketing Engineering Tools(www.mktgeag.com)

2)COMPONENT OBJECTS

Example: Automated Software Agents(Price Comparison Agent)

Degree of Integration

4)INTEGRATED SYSTEMS OF MODELS

Example: Group Decision Systems(Group Analytic HierarchyRocess)

3)INTEGRATED COMPONENT OBJECTSExample: Yield Management SystemsExample: Marketing Optimization System

(www.marketswitch.com)

Fig. 2. Marketing models classified by degree of integration and degree of visibility that can be deployed on the he

aspects of the decision environment, at the otherextreme (e.g., single user, multiple tasks; multipleusers, single task). On the vertical axis (Degree ofVisibility), we distinguish between models that areembedded inside systems (i.e., a blackbox modelthat works in the background) requiring few inputsor interactions with the user, and those that arehighly interactive and whose structures are visible.We discuss below four categories of models that fallat the extremes of these two dimensions and indicatehow the networked economy will enhance their use.

(1) Visible standalone models can be put on appli-cation servers (Application Service Providers arealready emerging, e.g., www.marketswitch.com) andaccessed by client browsers. In such an environment,software renditions of marketing models can bemaintained in central locations, minimizing costs ofupdates and distribution. Model users also benefitbecause they will always have access to the latest

versions of the software. Visible models with userinteractions can also become more valuable online.For example, applications ranging from simple com-putational devices, such as mortgage calculatorswww.jeacle,ie/rnortgage), to sophisticated pro-

grams, such as conjoint analysis (www.valueha~est.com), are available on a 24/7 basis. These applica-

tions are enhanced with online technical help (bothwith machine intelligence as well as live support),improved content (help files, tutorials, etc.), andlinked to related applications that are available else-where on the Internet. Many traditional marketingmodels, such as the Bass Model, would also benefitfrom being re-designed or re-implemented in newers o h a r e packages, for deployment over the Internet.

2) Component objects can be deployed morewidely on the Internet because they can be structuredto continuously monitor and optimize various aspectsof how an organization functions. Proctor and Gam-ble's access to purchase data for their products atWal-Mart allows P&G to deploy automated modelsto forecast demand, schedule production and deliv-ery, optimize inventory holdings and even assess theeffectiveness of their promotions.

3 ) Integrared component objects exploit the blur-ring lines between software, content, services, and

applications to deliver more complete decision solu-tions to managers. For example, an integrated seg-mentation system could run not only standard clus-tering algorithms, but could also access data fromelsewhere on the web before model execution, andthen distribute customized communications to cus-tomers in different segments. Yield management sys-


4/9

230 G.L . Lilien A. Rangnswomy/lnfern. I f esearch in a rk t i~g 7 20001 227 235

tems at the world's major airlines dynamically opti-mize schedules, prices and seat inventories and sendmessages to targeted customers about new travelopportunities that they might find attractive. Al-

though such models may be fully automated, or usedby unsophisticated users, the models themselves arelikely to be quite sophisticated (akin to an autopilotfor an aircraft) and require frequent updating andvalidation by highly skilled modelers.

4) Integrated systems of models put a logicallylinked set of models in the hands of decision-makers(possibly geographically separated) who need to sharetheir different knowledge bases for important com-mon decisions. (e.g., negotiation support, bid plan-ning, marketing planning models). For example,Lodish et al. (1988) described a subjectively cali-

brated market response model that required co-loca-tion of the decision-makers. In the Internet world,such subjective data inputs can be obtained onlinefrom managers n different locations, consensus de-veloped using models running on a server, and theresource and planning implications made available toall through a group decision support system.

Note that the Internet tends to drive the prices ofdigital products (i.e., products, such as marketingmodels, that can be distributed on the Net) down totheir marginal production costs, which are near zero(Evans and Wurster, 2000). As a result, many Inter-net-based models w ~l l e available almost free, atleast for hmited use (or offered in exchange forviewing commercials, etc.), making them even moreattractive for analysts and managerial usen alike.Thus, we expect an explosion in the availability ofcustomizable, scalable and (possibly) embedded de-cision models on the Internet, available anytime,anywhere, for anyone.

2.2. More heterogeneous mix ofmodel users: Whereare the analysts in the e-world?

As more consumers and businesses get connected

to the Internet, the clientele for marketing modelsexpands. Not long ago, the only people using mar-keting models were analysts, consultants, and mar-keting academics. This situation is giving way to amore heterogeneous mix of model users, includingstudents and frontline managers. For example, ourMarketing Engineering (Lilien and Rangaswamy,

1998) students come to class, connect their laptops tothe Internet and access data and information relevantto the class in real time. They have access to manyanalytic tools, often in the form of spreadsheet soft-

ware like Excel (along with key add-ins), whichprovide a wide range of readily available capabili-ties. ur students are increasingly employed in star-tups or in flat organizations where rapid decision-making is essential and seasoned analysts are notavailable to help. They are empowered early on intheir jobs to make decisions with the software, hard-ware, and data they can access online. Or, perhaps tolook at it from their point of view, they are literallyabandoned in a world of personal modeling with

little analytical support staff.Fig. gives a perspective on personal (or seif-

service) modeling, which is becoming a more com-mon context for marketing decision models in thenew economy. Personal modeling involves small tomedium scale decisions that must be made n a veryshort time, decisions that may not re-occur (at leastnot for the same individual) and need only moderateskills to perform the associated analyses. Such envi-ronments are characterized by:

limited modeling knowledge and ad-hoc modelingprocesses, without the luxury of quick access totrained human analysts;

nomathematical modeling, where the modelingprocess relies heavily on graphics, web-content,spreadsheets, and canned software to build anappropriate model;severe budget and time constraints, where an ac-ceptable answer is needed immediately, forcingthe modeler or user to rely heavily on assumptionsand judgments about many model inputs;modeling more for general insights (which prod-ucts or policies seem to perform better than theothers) than for specific numerical results.

Are decisions made in such environments worsethan when formal, objective decision models are puttogether? While evidence seems to suggest (not un-equivocally) that formal models built by skilled ana-lysts work better, the more appropriate question iswhether the systematic use of models based on judg-mental inputs works better than pure judgmental


5/9

G L . Lilien A. Rangaswumy/htern. ofResearch in Marketing 17 2000) 227-235

Criterion Personal models High end models

User training Moderate to high

Scale of problem

Time availability(for setting up m odel)

Costsmenefits

Technical skills for Low lo moderatesetting up model

Short

ow to medium

Small to medium

High

ow to moderate

High

Small to large

Renurence of problem Low orhigh*ow

Fia. 3 Characteristics of two extremes of m arketing decision models: personal versus high-end models. These m odels can fall in any of thef categories described in Fig 2 although managers would most likely develop simp lestanda lone personal models. ow for one-timeshidies e.e.. Marriott Conioint Shldv. Wind et al., 1989) and hi& for models in continuoususe ( e .~ . ,American Airlines Yield Management.System, Smith et al., 1992). Source: adapted from Powell, 1997.

decisions alone. The answer here appears to be yesas well (Russo and Shoemak er, 1989).

A well-trained marketing scientist or an alyst may,for good reasons, he skeptical about providing so-phisticated modeling tools to unsophisticated users.However, demand for help at the point of decision-making in a networked economy will force us toimprove the design and delivery of models, so thatthey can be used by people with widely varyingmodeling skills. So, what do we have to look for-ward to? As learning and work become more cou-temporaneous, managers will expect access to infor-mation, knowledge, and mod els that are just a few

clicks away on the Internet hat we will havemodeling on demand (Fig. 2, Box 1).

2.3. Data, data eu erywkere: What s a m odeler todo?

It is no news that the Internet is heling anexplosion of marketing data way beyond what wehave seen over the past few years (e.g., scannerdata). Web sew ers record clickstream data hatpeople are looking at, for how long, in what se-quence, and what products they order (if any). Chat

groups and forums record every message posted atthose sites. Ad delivery systems(e.g., Doubleclick)record where every banner ad is displayed. Compa-nies, such as binate.com, are recording user experi-ences regarding web shopping. The list goes on. W eare seeing marketing data collection on a scale thatwas unimaginable just a few years ago. A single

successful portal (e.g., Amazon.com) collects over30 gigabytes of data per day (about 2 3 millionpages o f text) in its server logs.

Most large web-databases currently available con-tain mostly behavioral data bey capture whatpeople do. To understand the reasons for the ob-served behaviors, we have to infer those reasonsthrough m odeling efforts, sup plementing clickstreamdata, for example, with data on past purchases anddata from surveys. Even with purely behavioral data,we need new models to capture the richness of themarketing data sources. For example, Wu and Ran-gaswamy (1999) develop a consideration set model

to explore whether online search and sort behaviorchanges how many brands consumers include intheir consideration sets (i.e., the bran ds they wouldconsider buying). Understanding such behaviorwould enable marketers to design strategies to in-crease the chances that their brand wouldbe one ofthe brands considered by consumers before theymake a purchase. As another example, Fader andHardie (1999) develop a model to determine whetherrepeat purchase rates are increasing, decreasing, orstaying steady, and to understand the drivers ofrepeat purchase rates. Mahajan and Venkatesh (2000)

summarize the modeis and findings that are begin-ning to emerge.In addition to the exploding quantities of market-

ing data, we are also beginning to see data that covera wider range of marketing phenomena. For exam-ple, with online grocery shopping data(Degeratu etal., 2000), we not only have the information that is


6/9

3 G.L. Lilien. A. Rongoswamy/Inlern. J . ofResearch in Marketing 17 12 1 227 235

typically available in traditional scanner panels (e.g.,price and promotion), but also detailed data about thenavigation process of users during a web visit. Withsuch data, we can develop richer models that arelikely to offer more valuable diagnostics about thewhy as well as the what of shopping behavior.

Other types of data, such as those captured bye.piphany.com, which records customers' search be-havior, should also produce new marketing modelsthat incorporate customer decision processes and al-low for heterogeneity in decision-making. New de-velopments should include productcustomization us-ing interactive conjoint models, and dynamic andcustomized promotions (constructed on the fly) basedon hierarchical Bayes models or Bayesian networkmodeling.

3 Becoming better modelers in the digital, net-worked economy

There is clearly much that marketing modelersshould do to adap t to, and exploit, the new m odelingenvironment. We suggest three actions here: (1)become more familiar with information technologies(IT); (2) build more (decision) relevant models, and3) do more real-world assessments of the value and

impact of their models.

3.1. Become more familiar with IT

Most traditional marketing models (at least thosedeveloped by academ ics) have focused on the modef -ing components themselves and not on factors thatenhance how the models will he used (exceptionsinclude models, such as those based on the DecisionCalculus approach, Little, 1970). Increasingly, mod-els that do not design in features that take advantageof the distributed and data-rich context provided bythe Internet (previous section) will become irrele-vant: they will not get used, and will have dimin-ished importance to future developm ents in the mod-eling field. To develop models that do get used,modelers must pay attention to the IT-infrashuctureunder which their models will be used.

Many important decision modeling opportunitiestoday are related to IT-based modeling, where mod-els are developed and driven by newer types ofmarketing data that are generated and made available

through corp orate (e.g., web logs), syndicated (e.g.,scanner data), and commercial databases (e.g.,MediaM etrix data: www.mediametrix.com). Thesedevelopments have spawned new modeling ap-proaches designed for IT-intensive environments,where the models will be used by different types ofusers (e.g., consu ltants, managers, analysts, etc.).Researchers in computer science, finance, physics,and other areas are developing and testing newermodeling techniques, such as Bayesian networks,neural networks, and data mining. However, market-ing modelers (especially those n academia) have yetto embrace these modeling approaches, and we riskbecoming marginalized as thought leaders if vendorsand practitioners deploy mod els based on these new ermethods without rigorous and independent assess-

ments of their validity and value. We do not suggestthat marketing scientists become IT experts atherwe marketing scientists must learn to work with ITexperts who can help us leverage our models forwider deployment to determine what works, whatdoes not, and to identify areas for future modelingresearch.

3.2. Build more relevant decision models

In an environment characterized by rapid changeand constant experimentation, data, by themselves,are of little value unless they can be deployed quickly

to drive decisions that impact company performanceand profits. Marketing models can help transformdata into actionable insights in four ways: 1) byhelping us understand and leverage customer atti-tudes, preferences, and choices; 2 ) by helping usunderstand and leverage marketplace (competitor)behavior, 3) by harnessing managerial insights andjudgments and 4) by deploying models more widelywithin the firm. Leeflang and Wittink focus mostlyon the first area; we outline some ideas about theothers.

3.2.1. Understanding and leveraging marketplacecompetitor) behav ior

The Internet has fostered a network-centric ap-proach to business. Increasingly, successful firmsfind that they are becoming focal players in a com-plex ecosystem (a node connected to many subn-odes). At one time, 1BM dominated the computerindustry and drove the pace of development there.


7/9

G L ilien A. Rnngmwamy/lnfern J ofResenrch in arketinr 17 2000) 227 235 33

The company often waited to introduce a new prod-uct until its R D department had developed thenext generation product. In the networked world,where knowledge is distributed across an intercon-nected set of players, IBM is more of a focalplayer that leverages developments wherever theyoccur. For example, recently, Ariba, i2, and IBMannounced a joint venture to co-develop and deploye-commerce systems. Likewise, Amazon.com basrecruited a large number of Associates to drivetraffic to its site. As marketplace behavior is drivenby such complex interactions between players, thereis an increasing need for the development of simula-tion models to help managers evaluate strategic op-tions.

Simulation models facilitate anticipatoxy learning

to enable managers to t y out alternative scenarios ina safe environment before committing real resources.While there are several successful simulation modelsin marketing (e.g., the Assessor new product devel-opment model, Silk and Urban. 1978, conjoint simu-lators, etc.), most of these models are aimed atsimulating consumer behavior. There are also someteaching tools to help simulate real-world competi-tive behavior (e.g., Markstrat, Capstone (www.capsim.com), and Photowars). However, in the net-worked economy, there is an increasing need formodels that simulate competitive and marketplacebehavior using real-time data feeds from multiplesources. For example, a decision system designed tocontinuously monitor, measure, and simulate bannerad effectiveness across a network of partner sitescould quickly suggest ways to optimize ad deploy-ment. If a news event at a golf tournament suddenlyincreases traffic at golf sites, triggering improvedeffectiveness of banner ads presented there, an adver-tiser for whom this target segment is important canthen redirect more of its advertising expenditures tothose sites. Such a capability will become moreimportant as fums try to gain competitive advantageby rapid adjustment of effort deployment to reflectthe continuously changing marketing response pat-terns characteristic of a networked economy.

3.2.2. H arnessing managerial insights and judgment .~

No matter how much data and how many modelsare available, managerial judgments will play a deci-

sive role in most decisions, especially those that havestrategic (as opposed to operational) consequences.Indeed, if more models and data are available, modelusers are likely to have to make more, rather thanfewer, judgments. Hard data represent the past andjudgments reflect beliefs about the future. And moredata will force more careful assessments of moreaspects of the future. Increasingly, successful modelswill be those that encourage and help bring outindividual and collective managerial judgments (e.g.,judgments about the range of a model parameter).

3.2.3. Deploying models more widely in irmsHaving data and infomation is not enough to

improve organizational performance; they must bedeployed when and where they are needed, whichmay require mass customizing reports, dynami-cally tailoring them to the needs of individual man-agers. To encourage broader use, models can belinked to internal databases and Enterprise ResourcePlanning ERP) systems, thus linking them to ITsystems already in wide use in companies. But be-fore models can be widely adopted, managers musttrust the insights they produce (cf., Mooman et al.,1992). Managers must also feel comfortable thatmodels complement, and not ovemde, their judg-ments. Another way to increase model use is to makethem an integral part of how an organization func-

tions by embedding them in larger IT systems (e.g.,a forecasting model embedded in an order processingsystem). If the overall system performs well, theembedded model may be deemed to be successful,and is likely to he used on a regular basis. In areassuch as yield management, direct marketing, andsegmentation there are numerous opportunities to putMarketing Models Inside existing business sys-

tems.

3.3. Do more real world assessments of the ualueand impact of marketing models

To drive both our research and our teaching,marketing academics need a better understanding ofhow and why models either get used or get rejectedby managers. By ail accounts, many fewer modelsare actually used than are developed. Is the problemwith our models, or just with their interfaces and


8/9

234 G.L. Lilien A. Rangarwamy/Infern. J. ofRmearch in Marketing 17 2000) 227-235

distribution mechanisms? If it is the latter, we needresearch to find better ways to get our models used.If it is the former, we should explore alternativeapproaches, like Artificia l Intelligence (AI), th at arebeyond our traditional toolkits derived from statis-tics, operations research, and psychology.

As more models are deployed on the Internet,model usage will become easier to monitor andmeasure. Who ar e using marketing models? How arethey using the models? What featuresare they usingmost? Using least? Answers to these questions willhelp develop models that are m ost likely to be used.We envision altemative versions of models beingavailable on the web, in different forms, so that wecan observe what model features and structures winin the marketplace.

Models take time and often cost substantialamounts to develop. Much marketing analysis isviewed by finns as expense; we need frameworksand measurement procedures to assess the businessand financial benefits associated with model use. Wesee the value of models both in changing actions(instrumental view short-term earnings ) as wellas in changing minds (larger view optionsperspective). The Leeflang and Wittink paper framesmodeling as an activity focused on faithfully captur-ing marketing phenomena. That view by itself willnot support a vibrant future for our field. We alsoneed more studies on the impact of models hedemand side of the equation (see Wierenga and vanBruggen, 2000) o better understand what works,what does not and why nd to apply that knowl-edge to provide modetmg on demand for the net-worked economy. Past studies on the effectiveness ofDecision Support Systems have mostly been con-ducted in laboratory with few exceptions (e.g., Gen-sch et al., 1990). Although such studies generally~nd icate hat D SSs imurove decision verformance.

research, developm ent, use, and effectiveness relateto one another.

4 Conclusions

In this commentary, we have focused on how thedigital, networked econom y provides both challengesand opportunities for marketing modelers. Our keychallenge is one of retooling: traditional marketingmodelers who are not net-savvy risk having theirwork becoming marginalized and irrelevant in thedigital age. Our opportunities are enormous: in thenetworked world, anyone, anywhere, anytime can b ean active (e.g., consultant, decision-maker) or pas-

sive (e.g., implemen tor) participant in a mark etingdecision system. The potential demand in the areasof research, teaching, and practice for what we arecapable of providing has neverbeen greater; Let ussee if we are both clever and dedicated enough toexploit these opportunities and meet the associatedchallenges.

References

Benbasat, L Nault, B.R., 1990. An e~ lu at io n f empir ical re-search in manag erial support systems. Decision Sup port Sys-tems 6, 203-226 (August).

Bhargava, HK. , Krishn an, R., 1998. The World Wide Web andits implications for OR /MS (feature article). INFO RMS Jour-nal on Computing 10, 359-383 (Fall).

Degeratu, A., Rangaswamy,A. Wu, R., 2000. Consumer choicebehavior in online and traditional supermarkets: The effects ofbrand name, price, and other search amibutes. InternationalJournal of Research in Marketing 17 (I), 55-78.

Evans, P., Wurster, T.S., 2000. Blown to Bits: How the NewEconomics of Information Transforms Sttatew . Harvard Busi-

modeled to bits: decision models for the digital,networked economy

Documents