An Anatomy of a Decision-Support System for Developing and Launching Line Extensions Author(S): Morris A

An Anatomy of a Decision-Support System for Developing and Launching Line Extensions Author(s): Morris A. Cohen, Jehoshua Eliashberg and Teck H. Ho Source: Journal of Marketing Research, Vol. 34, No. 1, Special Issue on Innovation and New Products (Feb., 1997), pp. 117-129 Published by: American Marketing Association Stable URL: http://www.jstor.org/stable/3152069 . Accessed: 15/04/2013 10:19

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Marketing Research.

http://www.jstor.org

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions MORRISA. COHEN,JEHOSHUA ELIASHBERG, and TECKH. HO*

For most firms producing fast-moving consumer packaged goods, line extension is central to their new product development (NPD) strategy. The authors present a decision-support system for managing the NPD process in this industry, which explicitly evaluates the financial prospects of new line extension concepts. The system developed is based on an in- depth analysis of 51 new product projects launched over a three-year period at a major food manufacturer. It embodies historical knowledge about the productivity of the firm's NPD process and captures some key research and development resource inputs that can affect this productivity. It also provides shipment forecasts at various stages of the NPD process and thus can be used at new product project review gates to evaluate line extension concepts systematically. Finally, the system also can be used to improve the practice of the NPD process by enabling its users to take a product line perspective, using incremental sales evaluation, and by facilitating cross-functional and inter-project learning. Although the system has been developed specifically for a packaged food company environment, its underlying design principles are generic and applicable to a wide range of industries.

An Anatomy of a Decision-Support System for Developing and Launching Line Extensions

For most industrialand consumer productfirms, success- dustry,characterized by a high rate of productobsolescence ful new products are engines of growth for sales and prof- and a narrowwindow of marketopportunity, depends on the itability.Although there is consensus about the strategicim- interactionof a numberof factors, includingthe level of cap- portance of new product development (NPD), no universal ital investmentto supportdevelopment (e.g., use of comput- formula for success has been prescribed.Firms in different er-aideddesign and engineering), the productivityof the en- industrieswrestle with differentcompetitive imperatives and gineeringresource (i.e., how fast it can improveproduct per- formulatedifferent strategiesto succeed in the marketplace. formance),the numberof engineeringhours allocated to dif- In Cohen, Eliashberg,and Ho (1996), we present an ana- ferent researchactivities in the NPD process, an appropriate lytical model for managing the trade-offs between a target choice of a targetlevel for productperformance, and time to level of product performanceand the time to market, two market. critical factors for success in high-technology industries The challenges faced by firms in the packaged goods (e.g., computers, packaged software). Because objective (e.g., food, detergents)industry are somewhat different.Us- measures for product performance can be determined in ing an in-depthanalysis of 51 new productslaunched over a these industriesfairly easily, the primaryobjective of a de- three-yearperiod at a major packaged goods company, we sign team becomes launching a product with sufficiently observed that it is often impossible to develop objective high performance in the shortest time possible. We also measuresof productperformance. Consequently, the drivers show that the ability to achieve success in this type of in- of success of NPD for packaged goods are different from those in the high-technologyindustry. Firms cannot perfect- determine customer wants ex ante for *Morris A. Cohen is the Matsushita Professor of Manufacturingand ly (especially truly Logistics, Operations & Information Management Department, and new products),so they must invest relatively more time and JehoshuaEl iashberg is Professorof Marketing,Operations and Information effort as they move throughthe design process to capturethe Management, Department of Marketing, Wharton School, University of "voice of the customer,"so that the new product will be Pennsylvania.Teck H. Ho is Assistant Professor of Operationsand Tech- more likely to meet customers' needs (Akao 1992). nology Management,Anderson School of Management,University of Cal- ifornia, Los Angeles. Therefore,the critical input resources are those that help to identify,and laterto influence, customers'wants. We have

Journal of MarketingResearch 117 Vol. XXXIV (February 1997), 117-129

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions 118 JOURNAL OF MARKETINGRESEARCH, FEBRUARY 1997

found that these inputs include the level of experienceof the Figure 1 new product design team leader and the amount of cus- THEPS2'S OVERALLSYSTEM DESIGN tomer-basedinformation collected during the NPD process. In addition,a significant amountof promotionand advertis- dollars ing must be invested to launch such products suc- New Product Function Button Invoked cessfully. Life Cycle of Support System Model (Equation The difficulty and risk associated with determiningcus- Number) tomers' needs have promptedmany firms in the fast-moving consumergoods industriesto rely heavily on line extensions for stimulating demand for their brands. Indeed, a recent study of leading consumer packaged goods companies re- Resource Product vealed that only 5% of new product introductionsare new Scenario Performance Analysis Inprovement brands,6% are brandextensions, and the remaining89% are Function line extensions (Aaker 1991).1 Many believe that line exten- (2.1-2.2) sions will continue to be central to NPD strategy in consumerpackaged goods markets(Aaker 1991). Therefore,the importance of effectively managing the line extension is the lower risk in- Promotion & process quite high. Moreover, despite Multi-stage Forecasting volved in line feel that their NPD Advertising Forecasting Functions extensions, managers Spending process is successful less than 50% of the time (Group EFO (2.3) Limited 1994). One way to improve the NPD process is to rely on decision-support systems (DSSs). The use of computer-based supportsystems to improvedecision making in marketingis Competitive, Cash Flow Cumulative Pricing, & Cos Analysis Profit Function not new. Several successful of DSSs have Assumptions implementations (2.7) been reported,supporting such activities as marketingmix (Little 1970, 1975), sales calling (Lodish 1971), retailing (Lodish 1981), consumer promotion (Keon and Bayer commercializationand communicationof new Cumulative Profit 1986), prod- Break-even Time ucts (Rangaswamy et al. 1987), new product launching (Choffray and Lilien 1986; Ram and Ram 1989), and market trendsdetection (Schmitz, Armstrong,and Little 1990). We describe a DSS designed for packaged goods manu- istics similar to those of existing products.The process of facturersto improvethe evaluationand selection of new line developing it, in cooperationwith the company,has already extension concepts, the allocation of the necessary re- led to changes in the company's NPD procedures.We con- sources, and the promotion of links between research and clude by discussing the actual and likely future impacts of development (R&D) and marketingdecisions. The DSS is the system on the company, its managerialand researchim- model-based and comparable to ADBUDG (Little 1970), plications, and the conditions necessary for its successful CALLPLAN(Lodish 1971), and BRANDAID (Little 1975). implementation. Several empirical relationships, which can be developed OVERALLSYSTEM DESIGN from company historical recordsof line extension launches, are embeddedinto the system. These empiricalrelationships Description of the System are used to predict the impact of alternativeresource input The DSS we developed is called Product Portfolio Sup- mix choices on anticipatedproduct performance and to gen- port System (PS2). The PS2 design is based on data available erate sales volume forecasts at various stages of the NPD in the company and supports marketresearch, brand man- process. As a result, the system can help improve the selec- agers, R&D managers, and manufacturing managers tivity of the NPD process so that inferior concepts can be throughoutthe NPD process. The basic shell of the system screened out early on, therebyenabling superiorconcepts to was created in Hypercard2.0 V2. be developed successfully (Kotler 1994; Wind 1982). In Figure 1, we indicate the overall.design of PS2 and the We illustrate how the system can be used to support a way it supportsresource allocation, as well as selection and food company's line extension strategy.The article is orga- evaluation of new product concepts, throughoutthe NPD nized as follows: First we describe the overall system design process. The system has four features that are worth noting. and its major functions and highlight its unique features. First, NPD is conceptualized here as a product perfor- Next, we illustratehow the system was customized for use mance improvementprocess. Thus, the mannerand speed at by the food company. The system currentlyis being evalu- which productperformance is enhanced throughoutthe de- ated as a basis for a corporatewidemethodology to assess velopmentprocess is modeled explicitly. This dynamic issue line extension concepts that have product-marketcharacter- is often ignored in the marketingliterature. Second, our system allows for continuousassessment of a IA brand extension occurs when a successful product's brand name is new product concept throughout its development process. used to enter a completely differentproduct category (e.g., Ivory shampoo). As more information becomes available, the system pro- an brandname is used to enter a new A line extension occurs when existing vides updatedand more accurateestimates of the prospects market segment in the same product category (e.g., Levi's Wrinkle-Free Dockers slack series). of the concept.

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions Line Extensions 119

Third, we adopt a life cycle perspective to NPD. With a mance improvementis similar, for instance, to the speed of given set of competitive, pricing, and cost assumptions, the knowledge production studied by R&D economists PS2can estimate both the cumulativerevenue and cost of the (Griliches 1984; Kamien and Schwartz 1982). Griliches new product over its life cycle. In this way, the firm is en- (1984) defines a firm's speed of knowledge productionas couraged to take a long-term perspective in evaluating the the number of patents generated by the firm per unit time potentialreturn on the investmentin a new productconcept. and shows thatit increaseswith the firm's R&D investments. Fourth, our experience suggests that management often If we measurethe outputof the NPD process in "unitsof focuses only on the individualnew productproject, ignoring performance,"the speed of the new product performance product line considerations. The system developed here improvementbecomes similar to the speed of knowledge evaluatesthe performanceof the entireproduct line with and production used by R&D economists. The improvement without the new productunder consideration and thus deter- function enables us to identify and quantify the relative im- mines the incrementalimpact of the new product. portanceof the various R&D resource drivers in improving Our overall approachlinks R&D and marketingactivities, the product's performanceover time. As mentioned previ- thus promotinggreater cross-functional integration between ously, the performanceof a productin the packagedfood in- the two functions. Research and development resource industry is often subjective, and it is thus gauged, indirectly, puts are evaluatedin light of their ultimateincremental sales by customers' preference measures (e.g., purchase intent and profit contributionsso that they are allocated to the de- and expected frequencyof consumption)and by perceptions velopment of the most promising productconcepts. Adver- of the new product(e.g., taste and uniqueness). tising and promotion dollars are employed to launch only Let Ri be the level of R&D resource input i per unit time those productsthat score well with respect to productattrib- assigned to the development process (i = 1, .... I). Assume utes and customer reactions, because they are the most like- that product performanceis gauged by a set of preference ly to succeed in the market.As a consequence, we believe measuresXj (j = 1, ..., J) and a set of perceived performance that PS2 is likely to lead to the development of more effec- measuresYk (k = 1, ..., K). Denote the speeds (rates) of im- tive new product strategies in the future. provement of these measures by Xj and Yk. We chose the Cobb-Douglas forms to model the product improvement PS2 's Functions functions (for empirical support, see Bohem 1982; Ho The system has four major functions: (1) productperfor- 1993). They are given by mance improvement and resource allocation analysis, (2) multistage forecasting, (3) cash flow analysis, and (4) con- (I) j = Kx R' R2...R ,j= 1,..., J, cept ranking(see Figure 2). These functions are discussed in the following paragraphsin more detail. (2) = R2 ...R , k = 1,..., K, Product performance improvementand resource alloca- Yk KkRI' tion analysis. The system supports the allocation of neces- Y where Kx and Kk are constants of porportionalityand oaji sary, and often scarce, development resources among com- and 3ki are productivityparameters of resource i.2 NPD This is done trade-offs peting projects. by assessing forecasting. The PS2 system evaluates the a Multiple-stage explicitly through product performance improvement prospects of a product concept during its development function. process. Our forecastingmethod combines three broadcate- The idea here, that the of NPD can underlying dynamics gories of methods to predict product sales and be construed as a (Choffray product performance improvement Lilien 1986): process, is not entirely new. The speed of product perfor- 1.Judgmental methods, which include Delphi-like group judg- mentprocedures, pooling of experts'opinions, and other qual- itative and Winkler 2 forecastingtechniques (Clemen 1993; Figure Wheelwrightand Makridakis 1980). THE FUNCTIONSOF PS2 2. Product testing methods, which include various pre-launch consumerfocus groups,surveys, and test markets.Product samplesare offered for evaluation on a test marketbasis, lim- itedby geographyor a selectedgroup of targetedconsumers. Prospectiveconsumer reactions are recorded and used to forecast sales. 3. Analogmethods, which identify similar product-market situa- tionsand assume that the way the new line extensionwill be acceptedin themarket is comparableto the waysimilar prod- uctshave been accepted. In our DSS, analogies and product testing methods are used to forecast initial post-launchperformance (e.g., sales, shipments to retailers, profits). Judgmental methods, cou- pled with initial post-launchperformance forecasts, are used to forecast subsequentperformance. Let the sales (or ship-

2This formulationdoes not imply that every resourceinput has an impact on every performancemeasure. For example, if aji is 0, then resourceinput i does not enter into the design productionfunction of preferencemeasure j.

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions 120 JOURNAL OF MARKETINGRESEARCH, FEBRUARY 1997

ments) volume forecasts (in units) in the first time period - D St = Dt_ A+ TA0(t ) after the product is launched (e.g., year or quarter) be (5) St denoted by Sl. Denote the level of marketingspending on a particularadvertising and promotionmethod I by Ml (1 = 1, A - ..., L). Then, we can invoke, for example, the widely TAo(t 1) T accepted linear relationship: A + TAo(t) .t=2. TAo(t - 1) TAo(t - 1) (3) S = + +...+ + Yl+... + 00 OXXI OXJOXJ +OYK where T is the sales cycle (e.g., end of the product life + OMMI+... + OML, cycle). Let rt = A/TAo(t)= St/(Dt - St) (from Equation4), rto= [TAo(t)- TAo(t- l)]/TA0(t- 1), and gt = (Dt - Dt _ )/ where Xs and Ys are consumers' preferencesfor and perfor- Dt- 1. Then Equation5 can be writtenas mance of the new and perceptions product concept proto- rt_ + I type. These measuresare collected sequentiallyas the prod- (6) St = St- (I + gt) ,t = 2,...,T. + I + uct flows throughthe NPD process. In the next section, we rt_1 rt? use Equation3 to obtain three best-fit models as more infor- the recursive 6 and mation(Xs, Ys, and Ms) becomes availableat the three suc- Using Equation SI (from Equation3), we can estimate for t > 1. Note thatrt is the relativeattrac- cessive developmentstages. St tion of the new line extension vis-a-vis all other productsin The company's collected consumer preference measures the category, rotis the growth rate of the total attractionof for the new concept include purchaseintent, which product all other products in the category, and gt is the growth (or is widely used in practicefor sales forecastingfor new prod- decline if it is negative) rate of the size of the total product ucts at the concept stage. Tauber(198 1) suggests thatthe use category at time period t. rotand gt can be determined,as in of purchaseintent to develop sales forecasts will continue to Little (1970), by using judgmental methods (e.g., pooling dominate in practice because of its simplicity despite the estimates from the appropriatebrand managers). In Equa- discrepancy between intent and actual behavior. Tauber tion 6, rt - I is computed using values of St _ l and Dt_1 (1975) shows that a concept's purchaseintent relates posi- obtained in the previous recursion. tively to new productawareness and trial purchasebut not to We chose to rely mainly on judgmentalmethods to obtain its repeat purchase.He concludes that a concept's purchase ro and gt because the use of analog methodsto forecastsales intent should not be used alone to make go/no go decisions in the subsequenttime periods presentssome problems.On at new productreview gates. In addition to a concept's pur- the one hand, productsthat have been launchedrecently are chase intent,consumers' perceptions of taste and uniqueness likely to be more representativeof new productconcepts un- for the concept, purchaseintent for the prototype,and first- der consideration,but sales for the distantfuture have yet to year advertisingexpenditure are employed in our sales fore- be realized. On the other hand, older products have all the needed sales but tend to be less of casting equations. data, they predictive sales for the new productconcepts under consideration,because The sales of the new productin the subsequenttime peri- the competitive environmentfor the new products is likely ods (St, t > 2) can be estimated using the estimated first-pe- to be differentfrom that of the old products.It is also worth riod sales level (SI) and managers' subjective judgments noting that PS2 consists of multiple forecasting equations about the market environment. Denote the (unobservable) and incorporatesconsumer reactions as they become avail- attractionof the new A. It is assumed that A is product by able at the different stages of the NPD process. The system fixed at launch and remains unchangedthroughout the life automaticallyadapts and improves forecasting accuracy by cycle. The total attractionof all other products in the cate- incorporatingnew informationitems into a revised forecast- gory (i.e., all productsexcept the line extension under con- ing equation.Thus, new productconcepts can be reassessed sideration)is denoted by TAo(t).Note that the lattervariable for their potential payoff throughout their development may vary over time.3 Let Dt be the size of the productcate- process. gory at time t. Using the classic attraction model (Bell, Cash flow analysis. The system also supports the con- Keeney,and Little 1975), we express the sales volume of the structionof the so-called "well-curve":the cumulativeprof- new productunder consideration at time t, St, as it projection of a product over a certain planning horizon given a set of competitive, pricing, and cost scenarios (see (4) S = D? ,t = 1,2,...,T. Figure 3). With the well-curve, it is possible to determinethe A + TAo(t) cumulative profit of the productover the planning horizon. Thus, products can also be evaluated on the basis of their - = * + + Dividing Equation4 by St_ Dt A/[A TAo (t 1)] and life-cycle profits.The cumulativeprofit realized by time pe- rearrangingterms, we have riod t is

T (7) Flt = (Pt - ct)St - DC - MS, 3TAovaries over time because competitorstoo extend their productlines A and and the firm may launch more line extensions subsequently. TAo where Pt and are the unit and cost of the at could also change as a result of changes in marketingmix. We assume A to ct price product be fixed over time here because our firm did not allocate much money for time t, DC is the development cost, and MS is the market marketspending to supportthe line extension after the first year. spending needed to launch the product.The specification of

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions Line Extensions 121

Figure 3 ucts in the same product category. Thus, consistent with a THEWELL-CURVE, OR CUMULATIVEPROFIT PROJECTION product line perspective, it is important to analyze the UNDERTWO COMPETITIVE SCENARIOS sources of sales volume to determinethe incrementalimpact of the line extension. For matureproduct categories such as packagedfood products,there is often little or no growth in Cumulative the category volume. Thus, the sales volume of the line extension can come from either cannibalizationof the firm's existing product line or reduced sales of competitive products. Our system enables users to develop an estimate of the 4 Less Intense / degree of cannibalizationand provides a forecast of the in- Competition \ crementalsales. Let TA,(t) be the total attraction of the firm's current productline (i.e., excluding the new concept) in the category in period t, and TS(0) be the firm's total shipmentvolume Break-evenTime I in period 0. Then, the total shipment volume of the firm in Iv period I if the line extension under consideration is not Fixed launched,TSn?(l), is Cost (8) TSno?() = Dl * TAf(l)

= DI TAf() TAo(0) TAo(0) TAo(l) Development Sales Cycle Cycle = (I + gl) TS(O) I + r Il+rl? If the line extension is launched,the total attractionof the firm's products in period 1 becomes A + TAf(O). Here the the pricing and cost scenarios are based on judgmentalesti- firm's total shipment volume in period 1, TSYes(l),is mates obtained from managers. Anticipated competitive A + TA() reactions affect sales volume (9) TSYes(l) = D (capturedby r?t) St (see Equa- A + TAo(l) tion 6), and hence the cumulativeprofits at time t. In Figure 3, we show how this feature can enable the A = - + D TAf(0) TAo(I) brand manager to evaluate the new product concept under A+TA(l) TAo(l) TAo((I)+A two competitive scenarios.As a result of more intense competition, the new product is likely to have a longer break- = S, +(l+ g)' TS(O)? I - . even time and a smaller cumulativeprofit. The construction of the well-curve also makes it possible to generate several useful new productperformance metrics, such as the break- Note thatall terms on the right-handside are either known even time and rate of returnof the project. Firms frequently or can be elicited from managers.The incrementalshipment use these performancemetrics to manage NPD. For exam- in period I due to the line extension is given by ple, Hewlett Packard'sBET/2 strategyis directed towardre- ducing productbreak-even time (BET) by one-half for all its (10) IS(I) = TSYes(l) - TSN?(1). new products(House and Price 1991). Equations 8 and 9 assume that Luce's choice axiom Ranking of concepts. Different criteria for evaluating a (which states that the ratio of the marketshares between two developmentproject can be applied at differenttimes during existing products remains before and after the the NPD process in order to rank new product concepts, unchanged line extension) holds. Ho and (1995) a which involves using, for example, a concept's consumer- Tang provide proce- dure for incremental sales when Luce's axiom based scores, first-yearsales, or total This estimating life-cycle profits. does not hold. feature permits the firm to change the evaluation criteriaat The returnfrom a new in terms of total differenttimes to gain a balancedassessment of a new prod- productconcept sales, incremental sales, or should be as- uct concept. For example, a superior new product concept life-cycle profit sessed against its required level of development resources (i.e., high concept score) may be hard to operationalizeand and We can formulatethe overall thus may score poorly as a prototypeor have high expected marketingexpense. easily resource allocation of the entire line as a unit cost. Such a concept may be rejectedas a result of such problem product an evaluation.4 constrainedoptimization program.The objective of the resource allocation is to maximize the overall ex- For many line extensions, total sales may not be adequate problem return of the line to the re- for determining the success of the NPD process, because pected product subject given source constraints. most line extensions cannibalize the sales of existing prod- User Interface 4To support managerialdecision making further,the system also offers The PS2 system is designed with the following principles several graphicaltools to allow visual comparisonof new productconcepts. in mind:

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions 122 JOURNAL OF MARKETINGRESEARCH, FEBRUARY 1997

1. Modular: PS2 is modular in that additional modules can be Figure 5 added easily. Additionalempirical relationships can be added THEFIRM'S NPD PROCESS to PS2 by using functional buttons (see the next item). This will reduce the lead time for modifying the system. 2. Button-driven:We use data and functionbuttons. Data buttons Process enable the firm to develop different levels of aggregationto Concept Product 4 Advertising Development-- Development - Development Development --- view the data. For instance, a data buttonfor unit productcost andSelection andTesting andTrial andPromotion can be created if it is useful to break this aggregatecost into Production its components. There is no limit to the level of aggregation allowed. The function buttonsinvoke differentembedded empirical models to analyze and evaluate productconcepts. The models are "best-fit" de- embedded empirical relationships t t I + rived from the historicaldata of similar productsin the prod- * ConceptScores * PrototypeScores UnitProduct Sales uct category (see multistage forecasting and resource alloca- * ProjectTargets * ExpectedCost Cost Profits tion analysis). 3. Single-screen:The use of functionand data buttonsenables us to adopt a single screen design. By "calling" the name of a Variousconsumer-based tests are then productconcept, the system providesa "card"that captures all furtherdevelopment. relevantaggregate data about the productconcept. Additional conducted to evaluate customer acceptance for the product details of a data item for the concept can be viewed by click- prototype.These include "taste and feel" surveys at central ing the associated data button. In Figure 4, we show how the locations, such as a shopping mall, and direct consumption MarketSpending data button is brokeninto its components. tests at consumers' homes. The tests generate information about consumerpreferences (Xj) and perceptions(Yk) of the IMPLEMENTATIONISSUES proposed new product. The process development and trial productionstep tests Overall NPD Process the feasibility of the proposed productfor mass production In this section, we describe how we customized the PS2 and provides a more realistic unit cost estimate. The adver- system for a major food processor. Figure 5 is a schematic tising development and promotionstage includes formulat- of the firm's NPD process. The developmentprocess, which ing plans for productlaunching. The plan includes selecting interfaceswith the screen illustratedin Figure4, consists of the appropriatelevel and mix of advertisingand promotion four majorsteps: (1) concept developmentand selection, (2) to supportlaunch of the product. prototypedevelopment and testing, (3) process development and production,and (4) promotion,advertising, and sales. New Product PerformanceImprovement The concept developmentand selection step entails com- Knowledge about the actual performanceof a new prod- municatingand expressing the new productconcept in terms uct within a firm is often subjective and not always well thatcustomers can understandand relateto (e.g., as a picture defined. Thus, development teams do not know, a priori, and/ora verbal description)to gauge marketreaction. which formulation of the product will best suit the target The prototypedevelopment and testing stage is quite la- group of consumers. Indeed,the majorobjective of the NPD borious and often occurs iteratively. First, initial product process is to determinethe definitionof a high-performance prototypes are assessed with an experimentaldesign using product.This is usually done by differentproduct tests con- focus groups. The reactions collected from the focus groups ducted duringproduct development. These tests gatherfeed- help to narrowthe numberof productprototypes targeted for back and increase the chance of meeting the customer's wants. Using the data made available to us by the company, we Figure 4 developed two performance measure improvement func- THESINGLE-SCREEN DESIGN tions for the NPD process. We collected six independent variables (project scope, number of prototypes used in the focus group, sample size of the focus group, experience lev- . *..* . .* ...... ?*-.* .~~~~~~~~~~~T . .. els of the lead and secondary investigators,and total labor U LAT UPDATE I SOf P |\i | | 91I4 | hours spent on the tests)5 and analyzed their predictiveroles MAMNUF|Goot Food | PRICE(WCASE) | | NT COST($ICAE E) on consumer perceptionsof the developed prototype.Using SCORES ^ a of 51 NPD launchedover a *.ii CONCEPT CLT HUT sample projects three-yearpe- |i DATE 614693 I DATE I I DATE 12/493 | riod, three independentvariables were found: to- 42 significant |"i DVB 30 DVB DVB ,j tal labor hours on the and FREQ = FREQ _ FREQ ___ En (RL) spent prototypedevelopment UNQ 40 TQ UiQ_Q the size of the focus (RN), and the ex- TABTE 21 TAXTE TAXTE testing, sample group VALUE VALUE __ VALUE perience of the primaryor lead investigator(RE). Because NEED N____ EED N____ EED the main of the focus is to understandthe dri- Advertisig: 6000 BB 82 Iu- purpose group vers of success, as perceived by the consumer, the MIPricBBeP?| o| | Actioa Std. | DATEOF LAUHCH 7/14/94 product CouponsluIreutCost: r____ yO I 2 3 4 5 I AllOtherMaketn: -6000 -57781 130101 1899|1 244531 287231 5Projectscope is measuredon a nine-point scale and is judged subjec- Market Spdg.(K$)1 -1 | CompetL Set I tively by the R&D director. Experience levels of the lead and secondary measured their of in the food indus- | FILE FINANCIAL RESOURCE MARKET i e( > a investigatorsare by years experience labor hours of all ...... try. Total labor hours spent on the tests is the sum of the ...... ll. . . .. e X. . .. ll ...... J. members in the team. The sample size is measuredby the total numberof subjects interviewed in the focus groups.

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions Line Extensions 123

total number of subjects used in the tests is a reasonable Figure 6 measurefor the extensiveness of these tests. RE is measured THE FIRM'S NEW PRODUCT PERFORMANCE IMPROVEMENT by the numberof years of experience of the primary(R&D) FUNCTIONS investigator;presumably, more experience leads to better- Log(SC) = (.05)*** + .0348 Log(R) quality information. O 4.3495 .0348Log(RL) The two dependentvariables in the performancemeasure Log(SC) = ' (.0245)*** (.0175)** improvementfunction are preference measures.6The first measuresconsumers' average liking for the product(i.e., the + .0348 Log(RE) + .0 123 Log(RN) "scoreboard,"XI = SC), and the second measures the con- (.0033)*** (.0037)*** sumers' intent the "defi- purchase (i.e., percentagechecking (Adj R-sq = .34) nitely will buy" X2 = DWBp) after a direct consumption test 2.8874 .3263 is administered. (ranges from 0 to 100%)is measured LogDBp = + Log(RL) DWBp Log(DWBp) = (.0245)*** (.1860)* by consumers' willingness to purchase the product when asked to evaluate the productin their own home. SC (ranges .3611 Log(RE) .1188 Log(RN) from 10 to 90%) is a measure of how much the consumers (.0923)*** (.0392) ** like the product.7 Because all new products are developed (Adj R-sq = .33) under a one-year schedule to conform to an industryprod- uct-launchingseason, new productperformance at the time *Statisticallysignificant at 10% level. of launchis used as a measurefor the rateof the **Statisticallysignificant at 5% level. productper- ***Statisticallysignificant at 1% level. formanceimprovement during that year.The total personnel +Numbers in the parenthesesare standarderrors of the estimatedcoeffi- hours expended is equivalent to the rate of the labor input. cients. Equation 1, when applied to our data set, is thus given by SC scoreboard measures how much subjects like the prototype (10-90) (11) SC = KXRL RE R(N, DWBp percentageof subjects checking top box in a five-point purchase intentquestion (0-100%) RL total R&D labor hours involved in prototype development (in person-years) (12) DWBp = K2R-L RN . RL2E RE years of experienceof the lead investigatorinvolved in prototype In Figure 6, we show the results of the regressionanaly- development RN number of subjects involved in the focus group for soliciting ses fitted empirically to Equations 11 and 12. The parame- feedback in prototypedevelopment ters of the three independentvariables were found to be sta- tistically significant.8 The total personnel year variable is less critical relative to the extensiveness of productguidance tests and experience of the primaryinvestigator.9 RE (a2EWN .3611WN Because all the regression coefficients are positive, the (13) RN Oa2NWE .1188WE R&D resource inputs complement ratherthan substitutefor each other. This design "production"function enables us to where a2E and Oa2Nare parameter estimates of the second find the optimal mix of R&D resource inputs to be allocat- regressionequation given in Figure 6. ed to a NPD projectto achieve a given level of output,either DWBpor SC. To illustratethis, let us focus on the two more MultistageShipments Forecasting significant resource inputs, RE and RN, and the production Nine independentvariables were available for launched of DWBp. If the constant marginal costs for RE and RN are projects that went throughall stages of the NPD process, to given by WE and WN, then it can be shown that the optimal establish regressions for first-yearshipments. They include levels of REand RN obey the following ratio (Varian 1984): consumer preference and perceptual measures generated during the concept test (purchaseintent, expected consumption frequency, product uniqueness, expected taste, per- 6Unfortunately,the firm did not collect any perceptual measures about ceived value), consumers' preference measures collected at the productduring the prototypedevelopment and testing stage. Some per- the test intent, and the measures about the prototype (purchase scoreboard), ceptual concept (e.g., taste and uniqueness) were col- firm's lected during the concept development and screening stage, however. We marketingspending decisions (media spending and could thus determine the impact of R&D resource inputs only on the pref- all other marketingspending). o The independentvariables erence measurescollected during the prototypingstage. that turned out to have significant predictive power are the 7Customerswere asked how likely they would be to buy the line exten- concept's purchase intent = the concept's per- sion at a store where at a certain for a cer- (XI DWBc), they normallyshop price point ceived taste = Taste), the concept's tain packaged size. A five-point scale is used (I = Definitely would buy it; (Yt perceived uniqueness = the intent = 5 = Definitely would not buy it). DWBpis the percentageof subjectscheck- (Y2 Uniq), prototype's purchase (X2 ing the top box (i.e., choosing 1). They were also asked to pick an expres- sion best describing how much they liked the product overall on a nine- point scale (I = dislike extremely;9 = like extremely). Scoreboardis com- 10Purchaseintent in the concept test is measuredin the same way as that puted from subjects' responses to this question. measured in the prototype test. Expected consumption frequency is mea- 8Linearand semilog functional forms were also investigated.They did sured on a five-point scale. Productuniqueness is the percentageof respon- not provide a good fit of the data. dents checking the top box in a three-pointscale question. Expected taste 9R-squareddrops from .34 to .29 and .33 to .30, respectively when RL is and perceived value are percentagesof people checking the top boxes in a removed from the regression. five-point scale question.

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions 124 JOURNAL OF MARKETINGRESEARCH, FEBRUARY 1997

Figure 7 Figure 8 THE SALES FORECASTING FUNCTIONS PRODUCT X'S RESOURCE INPUTS AND CHARACTERISTICS

Item Level or Score Concept Stage DevelopmentResources Sales = -932.54 + 14.65(DWBc + Taste) + 21.48Uniq *TotalR&D labor hours involved in prototype (RL) .6 person-year (292.8 1)** (4.34)** (5.27)** development *Yearsof experience of the lead investigator involved in development(RE) 1.5 years (Adj R-Sq = .44) prototype *Numberof subjects involved in the focus group for soliciting feedback in prototypedevelopment (RN) 60 subjects

PrototypeStage Concept Stage (Subjects were presented with a picture) *Percentageof subjects checking top box in a Sales = -931.30 + 11.41(DWBc + Taste) + 18.51Uniq five-point question about purchaseintent (DWBc) 30 (337.13)** (5.16)** (5.71)** *Percentageof subjects checking top box in a three-pointquestion about purchaseintent (Uniq) 40 + 10.86 DWBp *Percentageof subjects checking top box in a (4.63)** five-point question about purchaseintent (Taste) 21

(AdjR-Sq = .51) PrototypeStage (Subjects tried the prototypeat home) *Percentageof subjects checking top box in a five-point question about purchaseintent (DWBp) 42 *Averageliking of subjects for the prototype(SC) 82 MarketingStage Promotionand Advertising Sales = -872.38 + + Taste) + 13.37Uniq 12.85(DWBc *Mediaspending in million of dollars (Media) $6.0 million (320.93)** (4.90)** (5.82)**

+ 2.85DWBp + 79.39Media (1.48)** (30.88)** used to forecast the first-year shipment after both the concept and prototypetests are completed. Finally, if the media R-Sq = .66) (Adj efforts are known, then the third regressionequation can be *Statisticallysignificant at 5% level. used accordingly.Note that the goodness of fit increases as **Statisticallysignificant at 1%level. more informationitems become availableover development +Numbers in the parenthesesare standarderrors of the estimated coeffi- process. At the marketingstage, we achieved an adjustedR2 cients. of value .66.12 At the were with a of the con- concept stage, subjects presented picture An Illustrative Example cept. The following were collected: DWBp percentageof subjects checking top box in a five-point ques- In this section, we illustratethe use of PS2 with data avail- tion about purchaseintent (0-100%) able from one real new product(X) that was launchedby the box in a Taste percentageof subjects checking top five-point ques- firm the for a of interactionswith the tion about expected taste (0-100%) (see Appendix sample Uniq percentage of subjects checking top box in a three-point decision supportsystem by the user). In Figure 8, we show question about uniqueness(0-100%) the levels of various resource inputs, productand consumer At the prototype stage, subjects consumed the prototype at home. The testing scores, and the media spending of the product. following was collected: With the levels of resource the of box in a inputs, using regression DWBp percentage subjectschecking top five-point ques- are tion about purchaseintent (0-100%) equationsgiven in Figure6, the predictedSC and DWBp After first-yearof launch, the following were collected: 81 and 29 (the actual scores were 82 and 42, respectively). Sales first-yearshipment volume (in thousandsof cases) It follows from the regression equations given in Figure 7 Media media spending during the first year of product launch (in that the first-year shipment forecasts at the three different millions of dollars) stages are 671K, 847K, and 913K cases, respectively. The actual shipmentsfor the first year were 840K cases. In Fig- set estimates of DWBp), and media spending (MI = Media).11We experi- ure 9, we show an illustrative of subjective mented with various functional forms and found that the growth rates in productcategory, competitive, pricing, and best-fit models are the linear regression models. We report costing scenarios. In addition,the volume of the category in on three empirical regression equations (i.e., Equation 3 period 0, Do, is estimated to be 3,530K cases. With these with increasingly more independent variables as they are subjective estimates, we can construct the well-curve for a collected), which can be used to generate shipment five-year sales cycle (T = 5; note that the developmentcycle prospects for the first-yearafter launch (SI) throughoutthe is one year). NPD process (see Figure 7). The first regression equation can be used to generate the 12Regressionparameters have been rescaled to preserve confidentiality. first-yearshipment forecast afterthe concept test is done for R&D resource inputs play a significant but not as importantrole as media a new line extension. The second regressionequation can be spending through DWBp. We provide two caveats here. First, our sample consists entirely of product launches, whose DWBp and SC scores are above certainthresholds. This "biased"sample might limit the effects of the IIDWBCand Taste are highly correlated;we group them into a single two measuressomewhat. Second, we are unable to determinethe effects of measure. The reliability of the construct appears reasonable (Cronbach's R&D inputs on other productmeasures (e.g., Taste and Uniqueness) in the alpha = .8 1). prototypestage because these measuresare not collected.

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions LineExtensions 125

Figure 9 pany. It has also led to several managerialand researchfind- A SET OF SUBJECTIVEESTIMATES ings, which are described in this section. Actual and LikelyImpact lime Period Although PS2 has not been fully implemented,its princi- Item 1 2 3 4 5 ples and qualitativeimplications have had an impact on the Grow Rate of Category food manufacturer for which it was customized. Our Demand, gt 0% 0% 0% 0% 0% has provided this company with several links to Rate of approach Grow Total in the Attraction,TAo(t),rt 20% 20% 20% 20% 20% best practices following ways: 19.2 19.2 19.2 Unit Price, pt(in $) 25.2 20.2 1. The process of developing the system, with the active partici- Unit Cost 12.3 11.1 10.5 10.0 10.0 ct(in $) pation of functional managersand the disseminationof its results, has led to significant changes in the firm's concept and prototypescreening proceduresfor new products.Commonly held beliefs the relative value of differ- If we the estimate for from the last concerning predictive adopt SI regression ent variables, based on consumer data, collected at = response equation, then SI 913K cases. With this estimate, using different stages in the development process, have been chal- Equation6, lenged by the empirical results. Overall, the disseminationof 913 +I the system concept and its empiricalfoundations have helped to enhance the level of rigor and consistency in the firm's = 913 (I + 3,530 - = 795K. S2 0) 913 NPD process across both productand geographicaldivisions. + I .2 2. Productstrategy has been affected. The company's approach 530 - 913 3, for acceptance and rejection of concepts and prototypes involves the use of a "hurdlerate" for an aggregatemetric based = 688K, = 593K, and = 475K. Because Similarly, S3 S4 S5 on variousconsumer responses. Only those concepts and pro- St, Pt, and ct are known, using Equation 7, we can compute totypes that clear the hurdle are accepted or continued. The the cumulative profit. The project is expected to break even PS2-basedapproach naturally leads to an overall productline in the middle of the second year (1.5 years from the time the perspectivein which line extensions are evaluatedin terms of product concept was conceived). The cumulative profits by their incrementalfinancial impacton the existing productline the end of five years is $28.7 million (see Figure 10). (i.e., incrementalrevenue). As a consequence, the company The firm had 78% of the market share in period 0 (i.e., has added incrementalsales volume as an input to the screen- prior to the time period at which the new product is under ing process for productconcepts. Ultimately, with PS2, a life financialreturn will become for each launching consideration). Thus, TS(0) = 78% * 3,530 = cycle required stage gate review of a development project. 2,753K. Using Equation 8, we have TSn?(l) = 1 * 2,753 3. The developmentof the system also has led to the design of a 1/1.2 = 2,294K cases. From 9, we obtain TSYes(l) Equation unified global productdevelopment protocol, which combines = + - = Thus 913 2,294{ 1/[l+ (913)/(3530 913)]} 2,614K. best practicefrom within the company with the normativein- the incremental shipment volume in period 1, IS(1), is 2,614 sights of our model-basedsystem as well as with observations - 2,294K = 320K cases. Consequently, 35% of the sales vol- of best practice by other leading package goods companies. ume for the line extension in the first year come from reduced sales of competitive products, and 65% come from cannibalization of the firm's existing product line. Figure 10 Wind, Mahajan, and Cardozo (1981) suggest that the di- THEPRODUCT X'S agnostic power of a forecasting model should be taken into WELL-CURVE account in assessing its practical value. In addition to estimating the degree of cannibalization, PS2 provides diagnos- Cumulative tic insights into the relative impact of various R&D resource ProfitsA strategies on anticipated product performance. For example, the performance improvement functions given in Figure 6 suggest that if the experience level of the lead investigator is five years instead of one and one-half years, then new pre- 0 dicted SC and DWBp are 84.6 and 59.6, respectively. In ad- E dition, the system generates conditional well-curve esti- ?l mates under alternative competitive and environmental scenarios. These alternative scenarios can be captured by dif- 1.5 Years ferent values of ro and gt. Finally, with reliable probability estimates of these scenarios, the user can construct the risk \YearO Year I Year 2 Year3 Year4 Year5 profile associated with the rate of return for launching the $6.0 million line extension. DISCUSSION We describe the design and development of a decision to enhance the NPD Development Sales Cycle support system the management of Cycle process in fast-moving package goods industries. The approach described here has had an impact on the practice of NPD within the specific application site of the food com-

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions 126 JOURNAL OF MARKETINGRESEARCH, FEBRUARY 1997

This global protocol, to be used worldwide, is currentlybeing products can be evaluated for the long term on a financial rolled out by the company. basis. Simple and powerful performancemeasures, such as 4. PS2 can easily generate multiple performance metrics for break-eventime, rate of return,and net presentvalue, can be product development. The company now has expanded its constructedfrom these cash flow estimates at each stage of evaluationfor new to include some of these metrics. products a project. Consequently,stage gate reviews can be carried For example, rate of returnand break-eventime derived from out the developmentprocess, thereby increasing the well-curve are currentlyused routinelyby the senior man- throughout the effectiveness of scarce resources. agers in new productplanning. development 5. The role of cross-functional teams for product development Finally, a productline perspectiveis adoptedso that users has long been recognized a key factorof best practice.The de- can evaluateline extension concepts with respect to their in- velopment of the PS2 system indicated that there were chal- crementalimpact on the entire productline. This is particu- lenges within the company concerning such integration. In larly relevantin consumerpackaged goods, where cannibal- particular,coordination between R&D and marketing and ization often representsa significant source of sales volume among different divisions has not always been effective; for the line extension. hence, gaps between concept and product (in terms of consumer evaluations)sometimes can arise. The system supports Research Implications the use of a common databaseand the sharingof information across functions and divisions. It thus supportsmore effective From a researchperspective, we also make severalcontri- team performanceand is likely to lead to the elimination of butions. First, we provide furtherempirical evidence for the such gaps. This common database is currentlybeing imple- existence of crucial concepts underlying the NPD process, mented by the company. such as the performanceimprovement function. In Cohen, Eliashberg,and Ho (1996), this function was introducedat Other likely impacts of the system include the following: the conceptuallevel and shown to be a majordriver of strate- decisions such as time to market.Here we demonstrate 1. The system provides a means to specify resource allocation gic trade-offsexplicitly. For example, the design productionfunc- that such functions are estimable. tions given in Figure 6 can be used to determinethe optimal Second, our modeling approachidentifies research areas mix of R&D resource inputs to be allocated to a NPD project. that have been considered separatelyin the past. For exam- In addition,the relative value of each resource input can now ple, variables such as total labor hours and purchase intent be determinedby its relative contributionto product perfor- were traditionallyconsidered to belong to the operationsand mance enhancement(see Equation 13). marketingfunctions, respectively.They are considered here 2. As indicated in "An IllustrativeExample," the system can be simultaneously,and, hence, the analysis contributesto the used to generate diagnostic insights concerning the impact of evolving marketing/operationsinterface discipline. and environmentscenarios. alternativeresource, competitive, Third, our though preliminary and can be used to better allocate the scarce devel- empirical findings, These insights some for and em- resources. suggestive, provide guidance formulating opment some research For 3. The system, by design, encouragesthe applicationof different pirically testing interesting hypotheses. the focus forecastingand business propositionformulations at different example, the trade-offbetween the sample size of project stages. Our empirical results indicate that the signifi- groups employed and the experience level of the lead R&D cance of sales forecastingpredictor variables changes with the employee raises an intriguingempirical question about the development stage and that informationand knowledge are extent to which the voice of the customercan be substituted importantdrivers of product performanceimprovement The by an experienced projectteam. system also illustrates the role of forecasting a product's life Finally, our preliminaryfindings suggest that differential cycle performanceon project management.It, therefore,can attention should be given to alternate predictor variables be used as an effective learningtool. uniqueness, definitely will buy) at different stages of of failure and suc- (e.g., 4. The system supportspost-mortem analyses the NPD process. We can argue that scoring too high or too cesses and thus can enhance organizationallearning within low on a uniqueness scale may be detrimentalto the actual the company across projects and over time. performanceof the new product.This suggests another interesting research avenue: defining the "optimal"level of Managerial Implications uniqueness and the stage (e.g., concept, prototype)at which From a generic managerialperspective, our work has sev- it should be consideredas the most critical to the NPD team. eral implications. First, it provides a frameworkfor R&D and brand managers to understandthe linkages and impact Conditionsfor Applicability of their decisions on the success of a NPD project. With The PS2 system was developed with a particulardomain design productionfunctions, R&D managerscan now focus in mind. The new productconcepts of interest here are not on key resource drivers for product performanceenhance- revolutionaryin nature,so thathistories of NPD projectscan ment. Similarly, sales forecasting equations enable brand be used effectively to forecast sales of the new product.This managersto focus their attentionon critical productperfor- assumptiondoes not hold if a particularnew productis truly mance measuresfor improved sales. new and revolutionary.Thus, our system will work well for Second, our framework integrates information inputs line extension projects that frequently occur in packaged from multiple stages of NPD and thus provides a systemat- food companies. Our approachis restrictedby the data set ic way to ensure that good productconcepts are not "killed" needed to derive the empirical relationships. Many firms arbitrarilyand that only "the best of the best" productsare may not collect the needed data to implementsuch a system; launchedat the end of the NPD process. however, we believe that the benefits of a systematic and Third, the designed system estimates the cash flow of the disciplined approachto NPD will outweigh the costs of such project over a life cycle planning horizon, and thus new data collection. Last, the approachrequires brand managers

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions Line Extensions 127 to provide estimates for the size of the product category, its (see Screen S5). As shown, the revised forecast for SI at the growth rate, and the evolution of competitive product per- prototypestage is 847K, of which 296K are incremental.The formance over time. These estimates may be subjective, but user keys in planned advertisingdollars by clicking the mar- their inclusion forces managers to consider relevant strategic ket spending button (see Figure 4). Another forecast of the factors. sales volume can be obtained by clicking button7 of the forecasting menu (see Screen S6). We believe that our PS2 system can have a wider applica- 4. Business Proposition: A major managementtool for project The can be customized to suit other bility. system easily evaluation is the use of a well-curve. The user constructsthe packaged goods Our system will also be valu- companies. project's well-curve by clicking the FINANCIAL BUTTON; able to other industries that have the following characteris- the table in Figure9 appearson the screen. The user enters al- tics: ternative market growth, competitive, pricing, and cost scenarios. the GO button will cumulative *Objective product performance measures are difficult to ob- Activating generate over a horizon tain. The NPD process, therefore, is a process of continuously profits six-year planning (see Figure 4). testing the extent to which consumers' wants and needs are met. REFERENCES *The major investmentin NPD occurs at the launch of the new Aaker, D. A. (1991), Brand on the product (i.e., advertising and promotion) so that new product Managing Equity: Capitalizing Valueof a Brand Name. New York:The Free Press. concepts can still be killed after the prototypestage. This will ed. Function increase the value of the sales forecasting equations. The po- Akao, Y., (1992), Quality Deployment: Integrating tential impact of the system will be smaller if the cost of devel- Customer RequirementsInto Product Design. Norwalk, CT: opment prior to launch is a substantialpart of the total cost of ProductivityPress. the new product. Bell, D., R. Keeney, and J. Little (1975), "A Market Share Theo- Line extension representsa substantialportion of the new prod- rem,"Journal of MarketingResearch, 12 (August), 265-75. uct activities. This will increase the predictive value of the em- Bohem, B. W. (1982), Software Engineering Economics. Engle- pirical relationshipsderived from the existing products. wood Cliffs, NJ: Prentice Hall. Choffray,J. and G. Lilien (1986), "A Decision-SupportSystem for EvaluatingSales Prospectsand Launch for New Prod- APPENDIX Strategies ucts,"Industrial Marketing Management, 15, 75-85. We illustrate the use of PS2 with a concrete example (see Clemen, R.T. and R.L. Winkler (1993), "AggregatingPoint Esti- Figure 11). The four major steps, listed in likely order of mates: A Flexible Modeling Approach,"Management Science, usage pattern, follow: 39, 501-15. Cohen, M., J. Eliashberg,and T.H. Ho (1996), "New ProductDe- 1. Concept Test: When PS2 icon is clicked or activated (recall velopment: The Performance and Time-to-market Tradeoff," that PS2 is Hypercard based), the user is presented with ManagementScience, 42 (January/February),173-86. Screen S 1. Assume that X has been and Concept developed Griliches, Z. (1984), R&D, Patents, and Productivity.Chicago: tested with customers. The user wants to create a potential The Universityof Chicago Press. record to these test scores and evaluate its capture concept Group EFO Limited (1994), 1994 Innovation Survey: Report on sales volume the button in Screen prospect. Clicking concept New Products,Weston, CT. S I brings up a concept menu, one item of which is labeled Ho, T. H. (1993), "New Product Development Strategy Analysis: NEW. Because this is a new concept, the user clicks NEW and The Marketing-ManufacturingInterface," doctoral dissertation, is presented with a record template given in Screen S2. The Universityof Pennsylvania. user enters the concept name (in this case it is called X) and Ho, T. H. and C. (1995), "When Is Product Line Extension the test scores for the concept (DWBC, Taste, and UNIQ). Tang Beneficial?" UCLA. Screen S3 shows the sales volume forecast that is obtained by working paper, C. H. and R.L. clicking the MARKET buttonand Button 1 in the forecasting House, Price (1991), "The Return Map: Tracking menu, which consists of seven options based on information ProductTeams,"' Harvard Business Review,(January/February), inputs. Screen S3 shows that the forecast of first-year ship- 92-100. ments, Sl at the concept stage, is 674K, 35% of which are in- Kamien, M. and N. Schwartz (1982), MarketStructure and Inno- cremental. vation. Cambridge:Cambridge University Press. Keon, J. and J. Bayer (1986), "An Expert System Approach to 2. Resource Allocation: If the concept is selected for prototype Sales Promotion Management," Journal of Advertising Re- development at the review gate, the user can test alternative search, 26 (3), 19-28. resource allocation strategies on product performance mea- Kotler, P. (1994), Marketing Management, 8th Ed. Englewood sures (SC and DWBp) by clicking the RESOURCE button. Cliffs, NJ: Prentice-Hall. After levels for the the user clicks entering planned resources, Little, J. (1970), "Models and Managers:The Concept of a Deci- the GO button to obtain the and SC predicted DWBp (see sion Calculus,"Management Science, 16, B466-B485. Screen S4 underthe HUT [home use test] column). At the giv- (1975), "BRANDAID: A Marketing Mix Model, Part I: en levels of resource inputs, the and SC turn predictedDWBp Structure;Part II: Implementation," Research, 23, out 29 Operations to be and 81, respectively. 628-73. 3. PrototypeTest: At any time during the NPD process, new in- Lodish, L. (1971), "CALLPLAN:An InteractiveSalesman's Call formationcan be added to the record by clicking the Concept Planning System," Part2, ManagementScience, 18 (4), 25-40. buttonand Concept X, which is now partof the concept menu. (1981), "Experience with Decision Calculus Models and Assume that Concept X has been tested in a home-use test. Decision Support Systems;',"in MarketingDecision Models, R. The test reveals that DWBp = 42. The user enters this score. Schultz and A. Zoltners,eds. New York:North-Holland, 165-82. At this point, the user can provide a better sales volume fore- Ram, S. and S. Ram (1989), "ExpertSystems: An EmergingTech- cast that makes use of both concept and prototypetests scores nology for Selecting New ProductWinners," Journal of Product by clicking MARKET and button 4 in the forecasting menu InnovationManagement, 6 (2), 89-98.

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions 128 JOURNALOF MARKETINGRESEARCH, FEBRUARY 1997

Figure 11 USAGESCREENS ILLUSTRATION

SOUP | LASTUPDATE [ | MAMUF PRODUCTPORTFOLIO SUPPORT SYSTEM (PS)2 | (CApE) | PRICE($CASE) L Uot COST [ Version 1.02 SCORES ? 1993 M.Cohen,J. Eliashberg, T. Ho CONCEPT CLT mHUT DATE DATE DATEAE ~__ DVB - DVB DVI FREQ FREQ ____ FEQ ___ |UNE UONQ UNIQ _ TAETE ____ TASTE ____ TATE i IVALUE ____ VALUE ____ VALUE concept portfolioportfolio managenentmnaomnt MEED ____ HEED N____ EED

DATE |Develop Co(K $)I Action Std. OFLAUNLH| Z i v,.oTr. 1 2 3 4 5 Center for Manufacturingand Logistics Research CUM.LPROFITS (K$) I olo I I I The Vharton School Market SpfI.(K$)I | CompeL Set I

IFILE |FINANCIAL RESOURCE j MARKET | O ...... , ......

ScreenS 1I ScreenS2

.:-:...... :.::, .::: ? ' ."''v''=:: :, :..... SOUP Ix I LA!TUPDATE | --- I MA. , CONCEPT HUT ADVERT.,,[, SCORES SORES SCOR {KS . CONCEPT F'T-"--.- DATE 649'.-x DVB 30 . X FREQ 4 ! x x I _40UNK ' x x - TAOTE _21.- T | - - Tx ...... x _ J VALUE __ Y __ 'EEDI--~~~~~I i ' ' MARKET 0 Yr.l % Develop Cost(K $)1' ---- ' ActionmSales(K cases): 674 100 I 1 1I IncrSalles 236 35 [ lr.O_m._.0 I YE0CarnaL: 1 338 65 CUM. PROFITS (K $) I o01 ol Market} arktI .,,,~ ; m~, I,, CompeLCom,.,,t NetRevene(K 0:Io FILE FINANCIALSpdg.(K$)I|Z3 . MA Profit(K s)

I FILE | FINiANCIAL |RESOURCE |MARKE-T| W>^ 3 r .,L1\ V Dene,~,

ScreenS3 ScreenS4

- , , : !?? :14- : " -.: ;., :: ~::: :: 4: .:. ~ : ,. :-~ : : :: %I ..: I:- .':: :-::...... '...::: :::: : ::.~ SOUP Ix I LASTUP DATE I I SOUP HI LArT tDATEII

MAHUFL -.-..--CONCEPT HUTIT_. AV 4MAHMtF ....CONCEPT HUT...AVERT. ETEl--I SCORES SCORES "1-E= SCORES SCORES (i$) SCORES .Q.(.S (... CONCEPT, CONCEPT t DATE U413 r112/14193 I DATE h16493 2..L 1214, I V DVB 30 I-' - JI DVB 30 2 x FREQ 4 4 x x FREQ X--- XX UMH) 40 X TASTE 21 r6 x x TASTEUHQ 21 6-X x ..... xX--- VALUE VALUEzIFr- X X X !--7 x x ?- . HEED P- ..... - 1- ...... rx MAR... - ----E T.. NEED I--' - '-'.... MARKET 0 I}____~~~~ ~~Yr.l % Yr.l % Develop Cost(K $)IL 1 | Action Sales(Kcases): 847 100 Develop Cost(K $) 1 I Action Sales(Kcases): I 913 1 :.:v~'n~ Yr.O bu~IhcrSabes: 296 35 I ~~'~IncrSales: lnT 1 . 1 320 35 } ~ I ~~~~~~Yr.0 ~I? CaumaL:". 551 65 Tr.0 I CCamaRob.: ~ I 593 65 CUM. PROFITS (KS) o l CUM. PROFITS (K S) I -600 -6000. etRe ): Market Spdg.(K $)-lZo I CompeL NetRe(K): MarketSpdgt.(K $)I 6ooo CompeL FILE ...FI...CI - jRESOURCEC ProfitK: Cure. Pr-ofit(KS): S ILE FINANCIAL RESOURCE MARKET IFILE FINANCIAL RESOURCE MARKET >, S ,-'V~~~~~~~~~~~~~~ Done,- - .~ ^~~~~~~~v - Dorner. .^ J ...... I...U...U.^ .I.I.UJ.I.I.U...... - 1...... I ...... I I....I. I I.I. ...I...... l . .1.1 .. .11.1. N*^.< I------I - .. ., ...... Screen... S6...... Screen S5 ScreenS6

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions LineExtensions 129

Rangaswamy,A., R. Burke, Y. Wind, and J. Eliashberg (1987), New ProductForecasting, Y. Wind,V. Mahajan,and R. Cardozo, "'Expert Systems for Marketing" Report No. 87-107. Cam- eds. Lexington, MA: D. C. Heath and Company. bridge, MA: MarketingScience Institute. Varian,H. (1984), MicroeconomicAnalysis, 2d ed. New York:W. Schmitz, J., G. Armstrong,and J. Little (1990), "CoverStory: Au- W. Norton & Company. tomatedNews Finding in Marketing,"in DSS Transactions,Lin- Wheelwright, S. and S. Makridakis(1980), Forecasting Methods da Bolino, ed. Providence, RI: TIMS College on Information for Management.New York:John Wiley & Sons. Systems. Wind,Y. (1982), ProductPolicy: Concepts,Methods, and Strategy. Tauber, E. (1975), "Predictive Validity in Consumer Research," Reading, MA: Addison Wesley. Journal of AdvertisingResearch, (October), 59-64. , V. Mahajan,and R. Cardozo, eds. (1981), New-Product (1981), "Utilization of Concept Testing for New-Product Forecasting.Lexington, MA: D.C. Heath and Company. Forecasting:Traditional Versus MultiattributeApproaches," in

a I will I

Sol -U-011IN ------AMA's recent books will keep you current.

DictionaryofMarketing Terms OAMA MarketingEncyclopedia Issuesand Trends Peter D. Bennett,editor Spin the Fure PennsylvaniaState University Lfmm. . n. JeffreyHeilbrunn, editor T his completely revised and edition is an essential expanded W ith an emphasis on recent referencefor business and emerging issues, a professionalsand studentsalike. Fully , - H. panel of leading marketing cross-referenced for ease of use, this experts surveys the key areas of marketing, resourcelists comprehensive more than identifies currenttrends, and provides realisticprescriptions 2500 definitionsof up-to-date today's most important for the future.Contributors include industryleaders and top- terms. both marketing Covering the day-to-dayterminology name academics, as well as pioneers and inventorsof key and the specialized vocabularyin corporateand academic use, marketingstrategies and techniques:Tony Alessandra,John the Dictionaryhelps everyonefrom newcomersto senior-level Hauser,Philip Crosby,Philip Kotler,Jack Trout,Al Ries, Don executives a more marketing gain thoroughunderstanding of E. Schultz, Jagdish Sheth, and many others. The past, present, critical marketingconcepts. and futureof marketingis shaped in this renownedbook.

$32.95 AMA Members/$39.95 Nonmembers $42 AMAMember/$47.95 Nonmembers 1995. 1995.

CALL TOLL-FREE 1-800-AMA- 1 150 TO PLACE YOUR ORDER. HAVE YOUR CREDITCARD AND AMA MEMBERSHIPNUMBER HANDY. REFER TO JR297 WHENYOU CALL.

AM E R I C A N MI ARKETING ASSOCIATION 250 S. Wacker Drive, Suite 200, Chicago, IL 60606-5819 Ill l I I I I l I I II I I III I I I I I IIIIIIIIIIIIlll IIII ......

This content downloaded from 128.91.110.146 on Mon, 15 Apr 2013 10:19:14 AM All use subject to JSTOR Terms and Conditions