CHAPTER 4. MARKET SEGMENTATION Along with product positioning, market segmentation is one of the most talked about andacted upon concepts in marketing. Simply put, the basic ideas are: • Market segmentation presupposes heterogeneity in buyers’ preferences (and ultimately choices) for products/services. • Preference heterogeneity for products/services can be related to either person variables (e.g.. demographic characteristics, psychographic characteristics, product usage, current brand loyalties, etc.) or situational variables (e.g., type of meal in which beverage is consumed, buying for oneself versus a gift for someone else, etc.), and their interactions. • Companies can react to (or possibly initiate) preference heterogeneity by modifications of their current product/service attributes including price, distribution, and advertising/promotion. • Companies are motivated to do so if the net payoff from modifying their offerings exceeds what the payoff would be without such modification. • A firm’s modification of its product/marketing mix includes product line addition/ deletion decisions as well the repositioning of current offerings. Market segmentation and product positioning are inextricably related, as buyers andsellers seek mutual accommodation in product/service offerings that best satisfy preference andprofit objectives. This process takes place in a competitive milieu of other brands/suppliers in thesame product category or even other categories of goods competing for the buyers budget.
ADVENTURES IN CONJOINT ANALYSIS 2 In a priori segmentation, the number of segments, their relative size, and theirdescription are known in advance. In post hoc segmentation, these three characteristics are foundafter the fact. In terms of researcher activity, the newer methodology of cluster-basedsegmentation appears to have received considerable user attention in the past decade.The Role of Conjoint Analysis As we illustrate subsequently, conjoint analysis is well suited for the implementation ofselected types of market segmentation. First, the focus of conjoint analysis is squarely on themeasurement of buyer preferences for product attribute levels (including price) and the buyerbenefits that may flow from the product attributes. Second, conjoint analysis is a micro-basedmeasurement technique. Part worth functions (i.e., preferences for attribute levels) are measuredat the individual level. Hence, if preference heterogeneity is present, the researcher can find it.Third, conjoint studies typically entail the collection of respondent background information (e.g.,demographic data, psychographic data). One should bear in mind, however, that buyerbackground variables, particularly demographic ones, do not necessarily correlate well withattribute preferences. Increasingly, background data information is collected on respondents’perceived importance of purchase/use occasions. Fourth, even rudimentary conjoint studiesusually include a buyer choice simulation stage in which the researcher can enter new ormodified product profiles and find out who chooses them versus those of competitors. Two recent trends in conjoint analysis have served to make the method even moreapplicable to market segmentation. First, user friendly and relatively inexpensive PC softwarepackages for conducting conjoint studies appeared during the mid-1980s. The second trend is thedevelopment and application of optimal product and product line positioning models. Optimalproduct design models extend the conjoint analyst’s traditional search for the best profile in a
3 MARKET SEGMENTATIONsmall set of simulated alternatives. Product design optimizers search for the best profile in whatmay be hundreds of thousands (or even millions) of possible attribute-level combinations.Market Segmentation in the Context of Conjoint Analysis Exhibit 4-1 is a schematic diagram of the proposed segmentation approach. We firstconsider the researchers initial focus: buyer background characteristics versus product attributepart worths (as computed from conjoint analysis). All segmentation approaches ultimatelyconsider both facets. However, in some cases we first target the type of buyer we are looking forand then design the best product for that type of buyer. In other cases we use the part worthsthemselves as a basis for clustering buyers’ attribute-level preferences and then design the bestproduct for each resulting buyer segment. _______________________________ PLACE EXHIBIT 4-1 HERE _______________________________ At the next level in Exhibit 4-1, we choose either an a priori or post hoc (cluster-based)method. If our initial focus is on buyer background characteristics, the user either defines a set ofa priori target segments or clusters the battery of background characteristics to find segments. Ineither case, once this step is done, the product design model is used to find the best product foreach segment (defined, illustratively, as the product profile that maximizes contribution tooverhead/profits). The segmentation procedure is somewhat different when we focus on the part worths. Inthe a priori approach, the researcher may segment buyers in terms of their part worths for one (ormore) product attributes. Examples include sensitivity to price, most preferred brand, andpreferences across selected features. In the post hoc approach. it is the part worths (or some
ADVENTURES IN CONJOINT ANALYSIS 4function of them) that are clustered to obtain buyer segments having preference similaritiesacross the full set of attributes. However, the main distinction between the buyer characteristics and part worthsegmentation approaches is in the fifth branch, labeled “stepwise segmentation.” In thatprocedure, each buyer is considered a “segment of one.” The product design optimizer is used tofind the best single product, for the firm in question, that maximizes contribution to the firm’soverhead/profits. This can be done in two basic ways. First, the optimizer cm be used to find thebest replacement for the firms current product. Alternatively, the optimizer can be used to findthe best product addition. That addition maximizes the sum of contributions across all productsin the firms line (and, hence, cannibalization as well as competitive draw is taken into account). In a stepwise way, other products can be added, each based on the preceding criterion.Unlike the other segmentation branches, stepwise segmentation does not design optimal productsto rnatch specific segments (a priori or post hoc, as the case may be). However, in either thetargeted or stepwise approach, multiple products can he designed; in the former approach aspecific new product is simply designed for each target segment. As noted from Exhibit 4-1, the stepwise selection procedure ultimately induces a buyersegmentation in the sense that a final pass in the model identifies the background characteristicsof the buyers who choose each product in the array (including competitive products). All five branches in Exhibit 4-1 eventually produce two sets of outputs: • Product profiles with associated returns to the firm under study. • A size and background description or each buyer segment choosing one of the product profiles (or perhaps a competitive product) from the resulting array of choices.
5 MARKET SEGMENTATION Which approach is “best” becomes a managerial question, once more subjective criteriasuch as reachability, substantiality, and actionability, are introduced.Additional Considerations Three additional considerations underlie the schematic framework work of Exhibit 4-1.First, we assume that once a product profile has been designed optimally for a pre-specifiedbuyer segment, it becomes available as a potential choice option for all buyers. We do not “walloff” buyers by constraining the availability of each of the firm’s products to selected subsets ofbuyers. The model does not require free buyer access to all options (including competitiveproducts). It can be adapted to handle the “walled-off” approach. However, in our experience wehave found that most firms consider it more realistic to permit all competitive items in the firm’sproduct line to be available to buyers. Hence, the buyer is free to select the option he or she findsmost attractive. Second, more subjective criteria (segment reachablity, etc.) can be handled in part byresearcher-assigned weights on various background characteristics if the buyer-focus option ischosen. Weights can be assigned to either buyer characteristics, levels within characteristic, orboth. Often, these researcher-supplied weights will reflect information on advertising audience,demographic characteristics, and the like. Whatever the source, the weights provide a differentialattraction score for each buyer. That score, in turn, affects the composition of the optimal productprofile. Third, we emphasize that the principal criterion adopted here is a financial one – findinga set of products whose overall contribution to the firms overhead/profits is optimized. As canbe surmised, the approach of Exhibit 4-1 places less emphasis on statistical criteria (e.g.,goodness-of-fit measures in cluster analysis) and greater emphasis on financial return to the firm.
ADVENTURES IN CONJOINT ANALYSIS 6Illustrative Application An empirical example should help clarify the proposed approach. Our applicationinvolves a pharmaceutical firm (herein called Gamma) that produces an antifungal medicationfor the treatment of various female disorders. (The product class and attribute descriptions aredisguised.) Gamma currently has a modest share (14%) of the market. Alpha and Beta, two lower-priced but less efficacious brands, have shares of 6% and 10%, respectively. The “Rolls Royce”of the marketplace is Delta, whose share is 70%. Because of Delta’s dominant position in themarketplace, other competitors tend to compare their entries with Delta’s brand as a referenceproduct. Table 4-1 illustrates this point. Clinical cure rate, rapidity of symptom relief, andrecurrence rate are each expressed in terms of departures from Delta as a reference point.(Physicians also use Delta as a basis for comparing competitive brands.) As shown in Table 4-1,the antifungal therapeutic class is described in terms of eight attributes related to efficacy, sideeffects, dosage regimen, and patient cost over the course of therapy. _______________________________ PLACE TABLE 4-1 HERE _______________________________ Table 4-2 shows the current brand profiles of each of the four competitors, as well as theircurrent market shares. Gamma and Delta are priced the same. Gamma is superior to Delta interms of clinical cure rate, rapidity of symptom relief, and recurrence rate, whereas Delta isbetter than Gamma in terms of side effects and dosage regimen.
7 MARKET SEGMENTATION _______________________________ PLACE TABLE 4-2 HERE _______________________________Market Survey Gamma’s managers felt that their current pharmacological research efforts could produceproduct improvements in attributes on which it was currently deficient in relation to Delta. Someof those improvements would necessitate higher production costs, however. Managers decided tocommission a conjoint-based research study to determine what the demand effects of variousproduct improvements might be. A sample of 320 physicians were contacted by a nationally-known marketing researchfirm. Conjoint data were collected at the individual-respondent level by personal interviews.Respondents received an honorarium for their participation. In addition to the conjoint exercise,physician background data (including psychographic data) were obtained. As background, Exhibit 4-2 shows average part worths for the total sample obtained fromthe conjoint exercise. To reduce clutter, only the “best” level is labeled; Table 4-1 givesdescriptions of all levels. We note from Exhibit 4-2 that cure rate and cost of therapy are highlyimportant attributes on average. _______________________________ PLACE EXHIBIT 4-2 HERE _______________________________ Gamma’s managers were able to estimate variable costs at the individual-attribute level.Their estimates were crude, but generally followed a pattern that one would expect – moreattractive levels (on efficacy, side effects, etc.) would entail higher production and qualitycontrol costs. With cost estimates at the attribute level (and price data), we can compute a
ADVENTURES IN CONJOINT ANALYSIS 8contribution to overhead and profit for each profile combination that is composable from theeight attributes. For illustrative purposes, we assume that Gamma’s managers want to retain the firm’scurrent brand profile but are interested in extending its line with the addition of two newproducts. The new products could cannibalize the firm’s current brand, but might also draw sharefrom competitive products. As illustrative options. we consider five ways of selecting two newproduct additions for Gamma. 1. Buyer-focused a priori segment selection 2. Buyer-focused post hoc segment selection 3. Part worth-focused post hoc segment selection 4. Importance-weight-focused post hoc segment selection 5. Stepwise segmentation.Buyer-Focused Segmentation Three demographic/psychographic characteristics were available for segmentation. 1. Physician practice (solo vs. group-based practice) 2. Physician specialty (gynecology, internal medicine, general practice) 3. Psychographic profile (six different segments obtained from a previous cluster analysis of 24 psychographic variables).For illustration, we chose the first background variable – type of physician practice. The samplebreakdown was 48% solo versus 52% group practice. We then found the best product for eachseparate segment, conditional on Gamma’s current product remaining in the line. This analysisillustrates the a priori approach.
9 MARKET SEGMENTATION To implement the post hoc (or cluster-based) approach, we used a two-step procedure.First, multiple correspondence analysis was applied to the characteristics (type of physicianpractice, specialty, and psychographic segment) to obtain a coordinate representation of thephysician respondents in a common space. The respondents then were clustered by a k-meansprogram. Four different starting configurations were used and split-half replications of theclustering were done to obtain the most highly replicable two-cluster solution. The product-optimizing program was again used to find the best product for each of thetwo clusters, conditional on Gamma’s current product remaining in the line. As a final step forboth the a priori and post hoc procedures, all six products (four original and two additional) wereentered into the optimal product design program. Returns were computed for each Gammaproduct and identification numbers were recorded for all respondents choosing each product,including competitors’ brands.Part Worth-Focused Segmentation We applied two different cluster-based approaches (using the same split-half method justdescribed) to these data. First, we clustered respondents according to the part worths themselves,after centering the data around each respondent’s mean. Second, we clustered attributeimportances, as obtained from the conjoint model, by the same procedure. These twoapproaches, in general, produce different clusterings (which was the case here). In each case, twoclusters were found. Next. the same procedure was used to find two new product additions. These productswere entered and returns were computed for each of Gamma’s first-choice products, as well ascompetitors’ brands.
ADVENTURES IN CONJOINT ANALYSIS 10Stepwise Segmentation The last approach involved stepwise segmentation. First, the optimal design model wasapplied to the total sample to find the highest return product for Gamma, conditional on itscurrent product remaining in the line. The new product was added to the array. The model wasused again to find a second optimal product for Gamma, conditional on the first two productsremaining in the line. A similar procedure was used to find Gamma/s shares and returns andrespondents’ selections for the six products in the total competitive array. In sum, five different approaches were used to select two new products for Gamma. Allnew products were selected so as to maximize return to Gamma’s whole product line (i.e., thepotential for cannibalization was taken into consideration).Results of Analysis We first discuss the findings on market shares and returns received by Gamma under eachsegmentation and product design strategy. We then consider the segments themselves in terms ofrespondent background characteristics.Market Shares and Returns By design, each of the five segmentation strategies produces two new product profiles forGamma. The first finding of interest is that for three of the product attributes (duration of sideeffects, severity of side effects, and cost per completed therapy), the results are the same: oneday, mild, and $65.20, respectively (see Table 4-3). That is, virtually all respondents wanted thesame side-effect profiles in terms of duration and severity. Not surprisingly, the high cost($65.20) was not desired by most respondents. However, because of the costs necessary toachieve highly desired efficacy and side-effect profiles, the highest price turned out to be optimalfrom Gamma’s standpoint.
11 MARKET SEGMENTATION _______________________________ PLACE TABLE 4-3 HERE _______________________________ Table 4-3 gives comparative results for the segmentation strategies, based on the fivevaried attributes. Also shown are cumulative market shares for Gamma’s three products(including its status quo product) and return to the company expressed as an index value with abase of 100. The first point to note from Table 4-3 is the result for the first strategy, whereby buyersare segmented according to their type of practice (solo vs. group). The new product profiles areidentical between the two segments. Not surprisingly, this strategy gives Gamma the lowestshare and return of all five strategies (because the second product is redundant with the first). Clearly, type of physician practice is not a useful segmenting attribute in terms of newproduct design for our dataset. What happens is that buyers in the two segments are reasonablyhomogeneous when it comes to the best product for Gamma to market. Of course, they coulddiffer in product preferences that would entail less attractive products for Gamma, but evidentlydo not differ in terms of its best product strategy. This result illustrates the value in couplingproduct design with segmentation strategy. Not surprisingly. buyer similarity in preferencedepends on which products are being offered. The other four strategies provide differentiation between products 1 and 2. For example.in the case of buyer-focused post hoc segmentation, the two products differ in four of the fiveattributes shown in Table 4-3. However, in this case the best segmentation is provided by thestepwise approach, with a return index of 111.
ADVENTURES IN CONJOINT ANALYSIS 12 Still, the buyer-focused post hoc strategies each show a return index of 109, withcumulative market shares that are only slightly lower than that associated with the stepwisesegmentation approach. All of the preceding results are tempered by (at least) the following assumptions. 1. Gamma can produce the appropriate attribute levels at the costs used in the model. 2. Competitors do not retaliate by changing their profiles and/or adding new products. 3. The list of attributes and levels is reasonably exhaustive of the important attributes in the therapeutic class. 4. The sample is representative of the relevant population and parameter estimation error is relatively small. 5. Firms are at a rough parity in advertising, promotion, and distribution. 6. Physicians’ preferences for product attributes remain reasonably stable over the firms planning horizon. 7. The share and return estimates are based on “steady-state” attainment (i.e., the time path by which these are reached is not considered). 8. Segments are reachable, actionable. and substantial. We examine the last assumption in more detail by summarizing physician profiles of brandselectors.Background Profiles At the users request, the optimal design model records who chooses which brand/servicein the array. These files can be cross-tabulated with other variables, in this case the threebackground characteristics – type of practice, physician specialty. and psychographic segments.
13 MARKET SEGMENTATIONIn each segmentation approach, we found that the respondents who selected Gamma’s newproducts 1 or 2 had similar background attribute levels. In particular. the modal attribute levelswere (1) group practice, (2) internal medicine specialty, and (3) a psychographic segmentidentified as “primary interest in drug efficacy, information seeker, and proneness to brandswitch.” Exhibit 4-3 shows the profiles for four of the segmentation approaches. In the buyer-focused a priori approach, the two new products turned out to be the same. (Their modalbackground profiles were also the same as those found in the other four segmentationapproaches.) _______________________________ PLACE EXHIBIT 4-3 HERE _______________________________ From Exhibit 4-3 we see that the profiles are fairly similar across new products 1 and 2and across segmentation approaches. The stepwise segmentation approach seems to produce themost dissimilar background profiles for products 1 and 2, particularly in the percentagesclassified as group practice and internal medicine specialty. However. the differences are notextreme. Though the finding is not shown in Exhibit 4-3, respondents who chose Alpha, Beta, andDelta were drawn primarily from the solo practice group in all five of the segmentationapproaches. Modal profiles for specialty and psychographic characteristics do not differ fromthose found for Gamma. Other datasets, of course, may not show such high agreement acrossbackground attribute classification. In the illustrative case, Gamma might do well to emphasizeproduct attribute levels that distinguish its new products from competitors’ products (and letbuyer self-selection take over).
ADVENTURES IN CONJOINT ANALYSIS 14Recapitulation The case example shows how different segmentation approaches can lead to differentproduct positionings. In our example. stepwise segmentation produces the highest return forGamma (as measured across all three of its products). We also note that the buyer-focused apriori approach fails to discriminate between solo and group practice physicians in terms of bestnew products. In the other four approaches, attribute-level differences are noted across products, eventhough the returns are fairly close. The three post hoc clusterings produced clusters ofapproximately the same size. The clustering of the part worths produced somewhat differentresults than clustering only on the importance component of the part-worths. In our example, thepart worth-based clustering produced a somewhat higher product line return for Gamma. Though stepwise segmentation should, in principle, do very well in terms of market share(because its product selection potential is less restricted), the researcher should also considerreachability and other aspects of its segmentation. This more general objective accounts for thelast step in the segmentation strategy shown in Exhibit 4-1.Caveats and Limitations The advent of conjoint-based product line optimizers has led to a new tool for selectedtypes of market segmentation. As Exhibit 4-1 shows, segmentation and product positioning areinterrelated. The emphasis of thus dual approach is on constructing and using an operationalmeasure of segmentation that addresses share/return. For example, post hoc clustering isevaluated less by statistical discrimination tests of the clustering results than by how well the
15 MARKET SEGMENTATIONassociated new product positioning strategy is forecasted to perform in terms of corporatefinancial return. We believe the suggested approach can be helpful in real world applications (and hasalready received limited application), but several caveats and limitations must be mentioned asareas for future research.Measurement and Parameter Estimation Issues Parameter estimation in conjoint analysis is subject to error. Also, the model might beincomplete – important product attributes and/or important buyer characteristics could beomitted. To some extent, focus groups and survey pretests can be used to reduce modelspecification errors, and those preliminary steps are undertaken routinely by experiencedconjoint analysts. Cost estimation is also a difficult undertaking. The firm’s cost accounting group isassumed to be able to estimate independent, direct, variable costs at the individual-attribute level.If future investment outlays are also required, they must be estimated and assignable toindividual products. As would be surmised, the proposed approach appears to be most applicableto cases involving recombinations of current attribute levels as opposed to radically newproducts. Concomitantly. we assume that the firm’s engineers can produce the desired level ofeach attribute as dictated by the model.Part-Worth and Cost Stability Over Time Conjoint analysis is essentially a static, steady-state preference measurement technique(though some conjoint applications have involved parameter estimation over a series of timeperiods). The market share and return changes noted in our example obviously would not be
ADVENTURES IN CONJOINT ANALYSIS 16expected to occur instantaneously. Rather, time trends would have to be introduced to make themodel more realistic as a forecasting technique. Some research is underway to make conjoint analysis more “dynamic.” Procedures entaila variety of techniques, ranging from having respondents estimate the anticipated share of theirbusiness that a product profile would obtain over the next (say) two years to analyses of timepaths and diffusion patterns of previous new brand introductions in the same product category.Competitive Retaliation For ease of presentation, our example does not include competitive retaliation. However,the model is capable of including action/reaction sequences. Consider the following examples. 1. Delta, having observed Gamma’s new product introductions, could in turn optimize its product. assuming status quo attribute-level conditions for all competing products. This action could be followed by the actions of Alpha. Beta, and so on. 2. Delta, in conjunction with Alpha, could offer a joint new product, designed to provide the highest net contribution to their current products.Other retaliatory actions are also possible. However, the measurement problems associated withthose product extensions are considerable. If Gamma wants to forecast Delta’s response, it mustbe able to estimate Delta’s attribute-level costs and must assume that Delta’s information aboutbuyers’ part worths is the same as Gamma’s. Moreover, our model does not provide help onwhen competitive reactions might take place. Models based on game theory ideas have been proposed recently, but their application toreal world problems is still in its infancy.
17 MARKET SEGMENTATIONIncomplete Optimization The proposed approach has been designed for conjoint data and, hence, applies primarilyto product/service attributes and price. A more comprehensive model would incorporateadvertising expenditure levels, message content, media mix, sales promotional expenditures, anddistribution outlays. In principle, such additions could be made, but the measurement problemsare formidable. For the short run at least, applications of the proposed approach will continue totreat those elements of the marketing mix outside the conjoint model.Predictive Validity Above all, the manager wants to know how well the model predicts. Our applications ofthe proposed model have emphasized pharmaceuticals, high tech products (such as computersand telecommunications), and consumer financial services such as credit cards. We have foundthat managers view the model primarily as a planning and sensitivity analysis tool for exploringalternative product and pricing strategies. In sum, research on conjoint-based segmentation/positioning is still in its early stages.Though the approach shows promise for the development of buyer- and part-worth-focusedsegmentation strategies, much additional research is needed before its potential is realized.Appendix 4-A Throughout our discussion of the case example, we employ an optimal product designmodel called SIMOPT (SIMulation and OPTimization model). The SIMOPT model (andcomputer program) is designed to provide a systematic search for product profiles that maximizeeither share or return for a user-specified brand/supplier.
ADVENTURES IN CONJOINT ANALYSIS 18 In the case example, the total number of possible attribute-level combinations is 46 - 32 =36,864. This problem is a relatively small one for SIMOPT; in this case the program evaluatedall profiles in a few seconds. For larger problems (e.g.. in which the number of combinations exceeds 1 million),SIMOPT employs a divide-and-conquer algorithm that iteratively optimizes subsets of attributesuntil the program converges. This heuristic works very well in practice. In many cases however,complete enumeration (as used here) is practical.SIMOPT Features SIMOPT is designed to work with large-scale problems entailing up to 1500 respondentsand as many as 40 attributes, with up to 10 levels per attribute, and up to 20 competitivesuppliers. Its features include: 1. Market share and/or profit-return optimization. 2. Total market and/or individual segment forecasts. 3. Sensitivity analysis.as well as optimal profile seeking. 4. Cannibalization issues related to product complementarity and line extension strategies. 5. Calibration of results to current market conditions. 6. Constrained optimization, through fixing of selected attribute levels for any or all suppliers. 7. A decision parameter (alpha) that can be used to mimic any of the principal conjoint choice rules (mar utility, logit, BTL). The alpha rule assumes that the probability of buyer k selecting brand s is given by
19 MARKET SEGMENTATION S Π ks = U ks / ∑U ks α α s =1 where Uks is the utility of buyer k for brand s, α is an exponent (typically greater than 1.0) chosen by the user, and S is the number suppliers. 8. Sequential competitive moves, such as line extensions or competitor actions/reactions. 9. Capability for designing an optimal product against a specific competitive supplier. 10. Provision for accepting part worth input that contains two-way interaction effects, in addition to the more typical main effects. 11. Preparation of output files containing ID numbers of buyers selecting each competitive option. 12. Computation of the “Pareto frontier”; the frontier consists of all product profiles that are not dominated by other profiles in terms of both market share and return.The SEGUE Model In addition to SIMOPT, a complementary model (and program) called SEGUE has beendesigned. SEGUE has two principal functions. First, it provides the user with descriptivesummaries of part worths and attribute importances for user-composed target segments. Second,it prepares a respondent weights file that summarizes each buyer’s “relative value” in meetingsegment desiderata. This buyer weights file is input to SIMOPT to obtain optimal products (etc.)for user-composed target segments. Table 4-4 summarizes the input/output aspects of each program, as well as several of theoperations that each program performs.
ADVENTURES IN CONJOINT ANALYSIS 20 _______________________________ PLACE TABLE 4-A1 HERE _______________________________
21 MARKET SEGMENTATIONTable 4-1. Attribute Levels Used in Conjoint SurveyClinical cure rate in comparison with Delta 10% below Equal to Delta 10% above 20% aboveRapidity of symptom relief in comparison with Delta 1 day slower Equal to Delta 1 day faster 2 days fasterRecurrence rate in comparison with Delta 15% above Equal to Delta 15% below 30% belowIncidence of burning/itching side effects 17% 10% 5% 2%Duration of side effects 3 days 2 days 1 daySeverity of burning/itching side effects Severe Moderate MildDosage regimen: 1 dose per day for 14 days 10 days 5 days 2 daysDrug cost per completed therapy $65.20 $58.85 $44.60 $32.40
ADVENTURES IN CONJOINT ANALYSIS 22Table 4-2. Current Drug Profiles of Four Competitors Attribute Alpha Beta Gamma DeltaClinical cure rate in comparison with Delta 10% below 10% above 10% above EqualRapidity of symptom relief in comparison with Delta 1 day slower 1 day faster 1 day faster EqualRecurrence rate in comparison with Delta 15% above Equal 15% below EqualIncidence of burning/itching side effects 17% 10% 5% 2%Duration of side effects 2 days 3 days 2 day 1 daySeverity of burning/itching side effects Severe Moderate Moderate MildDosage regimen: 1 dose per day for 14 days 10 days 5 days 2 daysDrug cost per completed therapy $44.60 $44.60 $58.85 $58.85Current Market Share 6% 10% 14% 70%
23 MARKET SEGMENTATIONTable 4-3. Profiles of New Gamma Products from Optimization Program (five attributes) Clinical Cure Rapidity of Recurrence Incidence of Dosage: Segmentation Strategy Rate Relief Rate Burning/Itching 1 Dose PerBuyer: A Priori Product 1 20% above 2 days faster Equal to Delta 17% 10 days Product 2 20% above 2 days faster Equal to Delta 17% 10 days Gamma share 74.9% Return (Index) 100Buyer: Post Hoc Product 1 10% above 2 days faster Equal to Delta 2% 10 days Product 2 20% above 2 days faster 15% above 17% 14 days Gamma share 80.6% Return (Index) 109Part Worth: Post Hoc Product 1 Equal to Delta 2 days faster Equal to Delta 2% 10 days Product 2 20% above 2 days faster Equal to Delta 17% 10 days Gamma share 81.8% Return (Index) 109Importances: Post Hoc Product 1 20% above 2 days faster Equal to Delta 17% 10 days Product 2 20% above Equal to Delta Equal to Delta 2% 10 days Gamma share 79.2% Return (Index) 103Stepwise Segmentation Product 1 20% above 2 days faster Equal to Delta 17% 10 days Product 2 Equal to Delta Equal to Delta Equal to Delta 2% 10 days Gamma share 83.1% Return (Index) 111
ADVENTURES IN CONJOINT ANALYSIS 24Table 4-A1. Characteristics of Computer Programs Used in Case StudySIMOPT • Individual part worth files • For any set of competitive profiles, • Market share/return for each • Individual’s importance weight the program computes supplier file share/return for each supplier • Individual supplier selection file • Demographics (background) file • All shares/returns are automatically • Optimal product description for • Current market shares for all adjusted to base-case conditions total market or selected suppliers • Sensitivity analyses can be segment • Each supplier’s profile performed at the individual • Sensitivity analysis results by • Value of alpha and demographic attribute level level within attribute attribute weights • Optimization can be carried out by • Control parameters for supplier or for groups of suppliers; organization attribute levels can be fixed for • Attribute-level cost/return data conditional optimization • Analyses can be conducted at the total market or selected target segment levelSEGUE • Individual part worths file • For any target segment • Attribute importance, level • Individual’s importance weight composable from the background desirabilities, and ideal levels, file variables (with weights supplied by by selected segment • Demographics (background) file the user), the program computes • Profile utilities by selected • Segment attribute weights size of segment, ideal levels, segment attribute importances, and attribute • Respondent weights file desirability levels summarizing each individual’s • Both additive and conjunctive relevance to the target segment segments can be created (input to SIMOPT) • The user can also input any trial product profile and find its total utility compared to the best profile • A respondent weights file is prepared for later use in SIMOPT
25 MARKET SEGMENTATIONExhibit 4-1. Market Segmentation in the Context of Conjoint Analysis Initial Researcher Focus Buyer background characteristics Product attribute (including use occasions) part worths Segmentation Approach A priori Post hoc A priori Post hoc Stepwise segmentation User selects target User clusters buyers User selects target User clusters part segment background on set of background part worths for buyer worths of attribute characteristics characteristics segmentation importances Optimal Product Design Optimal Product Design Model Finds Model Finds Best K Best Product for Each of the Segments Products Sequentially Total Contribution to Overhead/Profits is Computed Background Profile is Found for Selectors of Each Competitive Product
ADVENTURES IN CONJOINT ANALYSIS 26Exhibit 4-2. Average Part Worth Values from Conjoint Model (see Table 4-1).ScaleValues 0.6 Unreadable 0.5 • 0.4 0.3 Unreadable Unreadable • 0.2 • Unreadable • Unreadable .01 • • • • • • • • • • • • • • • Cure Rapidity Recurrence Incidence of Duration of Rate of Relief Rate Side Effects Side EffectsScaleValues 0.6 Unreadable • 0.5 0.4 0.3 Unreadable 0.2 • Unreadable .01 • • • • • • • • • Severity of Rapidity Recurrence Side Effects of Relief Rate
27 MARKET SEGMENTATIONExhibit 4-3. Profile Charts of Background Attributes by Segmentation TypesSegmentation Approach Percent Group Percent Internal Percent Efficacy/ Practice Medicine Seeker/Switcher Buyer Post Hoc Part Worths: Post Hoc Importances: Post Hoc Stepwise Segmentation 0 100 0 100 0 100 Product 1 Lengths of bars refer, respectively to percent of segment classified as group practice, internal medicine specialty, and psychographic segment: Product 2 efficacy/seeker/switcher