Table of Contents
Go Back
Guiding New Product Development & Pricing in an
Automotive High Tech SME:
When customer preferences are critical to strategic decisions using Conjoint
Analysis adds needed precision to model formulation & validation
Markus J. Schmidt
PA Consulting Group, Inc.
One Memorial Drive, Cambridge, Massachusetts 02142 USA
Voice: +1 617-252 0245, Fax: +1 617-225 2631
Markus.Schmidt@ PA Consulting.com
Michael Shayne Gary
Australian Graduate School of Management
UNSW Sydney NSW 2052, Australia
Voice: +61 (2) 9931 9247, Fax: +61 (2) 9663 4672
Shayneg@ agsm.edu.au
Please do not cite or quote without the authors’ approval.
*The authors would like to acknowledge the support of PA Consulting Group and particularly Alan
Graham for his helpful comments on this paper. We are also grateful for the valuable comments and
suggestions from George Backus, John Sterman, Carmine Bianchi, and two anonymous reviewers.
Abstract
This paper describes the use of system dynamics in combination with conjoint analysis
to assist a high tech SME in evaluating different policy options in a context where
customer preferences were critical to strategic decision making. Conjoint analysis
served an important role in eliciting customers’ underlying multi-attribute choice
preferences, and had a significant impact on both the structure and parameterization of
the final simulation model. The combination of methods was quite powerful in this case
and we feel could be successfully applied to a broad class of problems where behavioral
policies of decision makers include tradeoffs among multiple attributes. In such cases,
conjoint analysis- or other methods developed to address the multi-attribute choice
problem- can add needed precision to model formulation and validation. The alternative
is to use SD as a stand-alone approach and employ formulations that are not empirically
derived or grounded in the extensive choice theory literature. We suggest this
alternative is not viable when choice preferences are important for guiding strategic
decisions, and more generally we contend that appropriately integrating relevant
methods can substantially improve our system dynamics models and policy analysis.
Key Words: System dynamics, conjoint analysis, multi-attribute choice, new product design, automotive
industry, product preference, multi-attribute utility function, Small & Medium Size Enterprises (SME)
1. Introduction-T he Power of Combining Methods
Developing and introducing successful new products is crucial for the survival of most
firms and particularly for small and medium sized enterprises (SME's). This paper
describes the use of system dynamics in combination with conjoint analysis to assist a
high tech SME in designing a new product for launch and in analyzing a number of
different pricing strategies for the company’s existing product. The system dynamics
model was utilized within the company to rehearse these crucial strategic decisions.
Conjoint analysis served an important role in formulating and estimating customers’
decision policies and in guiding specification of model structure. The combination of
methods was quite powerful in this case and we feel could be successfully applied to a
wide range of issues where customer preferences are critically important in evaluating
strategic alternatives.
The primary contribution of this paper is to provide a specific demonstration of how
integrating modeling methodologies can significantly improve the formulation of
system dynamics models and thereby improve policy analysis in supporting strategic
decisions. It is our belief that system dynamics is too often employed as a stand-alone
approach, and that there are substantial synergies from appropriately integrating relevant
methods. We looked through each issue of the System Dynamics Review for the last
nine years, and found very few articles discussing anything other than the application of
system dynamics models in isolation. A related point, expressed by others previously,
is that we should always strive to use numerical data and the appropriate statistical tools
to ground our model formulations (Sterman, 2000 p. 854; Homer, 1997; Oliva and
Sterman, 2001). We cannot cover the entire range of potentially complementary
methods in one paper, so instead we describe one specific application, where combining
methods proved valuable, that we believe can be applied to a broad class of problems.
Specifically, we will discuss an application in which the behavioral policies of actors
include making a choice involving trade offs between multiple attributes- a multi-
attribute choice problem such as customer preferences for one company’s products or
services over rivals’. We argue that when multi-attribute choice preferences are critical
to strategic decisions or important to overall model structure and/or behavior, modelers
should make use of the extensive body of theory and empirical methods developed to
address multiple criteria tradeoff problems.
Multi-attribute utility measurement has been combined with system dynamics in only a
few previous studies available in the public domain. Nuthman (1994) discusses
cognitive algebra and the use of judgment in model formulation, and identifies
information integration as an area of research that system dynamicists should investigate
to ground our cognitive formulations. Others have used multi-attribute utility
measurement in field work to evaluate model generated policy options (Gardiner and
Ford, 1980; Reagan-Cirincione et al, 1991); an application of the techniques that we feel
should be more widespread. Homer (1996) mentions that he uses conjoint analysis, or
data derived through conjoint analysis research conducted by the company, in a
modeling project to explore product positioning in the pharmaceutical industry.
However, he does not explain the use of conjoint analysis in the project since that was
not the focus of the paper. We will try and fill this gap by describing in some detail
how we combined system dynamics and conjoint analysis to significantly improve the
formulation of our system dynamics model and thereby improve the subsequent policy
analysis.
The next section provides additional background about multi-attribute choice problems,
acommonly employed SD formulation for this class of problems, and an introduction to
conjoint analysis. Section 3 focuses on a detailed discussion of the steps involved to
effectively integrate SD and conjoint analysis by describing a modeling project with a
high tech SME in the automotive industry. Section 3 describes a series of simulation
experiments designed to help the SME’s management team design a robust pricing and
product development strategy. Finally, Section 4 discusses our findings and
conclusions.
2, Background- Formulating an Attractiveness Index
Researchers from a variety of disciplines- economics, operations research, psychology,
statistics, and marketing- have studied aspects of the multi-attribute choice problem. As
a result, there are a variety of techniques to analyze choice preferences for situations in
which a decision maker has to choose among options that simultaneously vary across
two or more attributes. An enormous range of complex decision problems involve
multiple conflicting objectives where the fundamental issue is one of value tradeoffs,
and over the years a number of system dynamics models have addressed multi- attribute
choice problems in various topic domains. Examples include models dealing with such
diverse issues as new product diffusion in consumer and industrial contexts, urban
dynamics, competition for market share in a variety of industries, competition in
Tecruiting high quality staff, and the growth of alternative modes of transportation
(Forrester, 1969; Piatelli et al, 2002; Backus et al, 2001; Mayo et al, 2001; Maier, 1998;
Paiche and Sterman, 1993; Doman et al, 1995; Ford, 1995).
There is a long tradition in our field of incorporating attractiveness or relative
attractiveness variables in our models to capture behavioral policies of actors trading off
multiple attributes. These formulations are typically guided by the judgment of the
modelers themselves, opinions of experts such as experienced managers in the industry,
and readily available numerical data. This information is used to identify attributes
important in the choice decision, and to specify formal mathematical relationships for
5
the value tradeoffs across all attributes. These formulations often involve nonlinear
attribute utility functions that are combined in a multiplicative attractiveness index. As
one indication of how pervasive this piece of structure has become in our field, the
Product A ttractiveness Molecule’ operationalizes this structure into a building block for
the use of novice and experienced model builders (Hines, 1995). We do not suggest
here that this formulation is wrong, and furthermore we applaud the efforts of those who
have developed the molecules building-blocks as a resource library. However, when
multi-attribute choice preferences are critical to strategic decisions or important to
overall model structure and/or behavior, we contend that modelers should ground these
formulations in the extensive choice theory literature and empirically derive underlying
choice preferences. The consequences of not grounding these formulations are: 1)
misspecifying model structure by ignoring important preference differences across
decision makers, 2) inaccurate attribute utility functions that do not capture underlying
preference structures, and 3) erroneous conclusions derived through policy analysis
based on a flawed model. Employing conjoint analysis or another appropriate method
developed to address the multi-attribute choice problem can add needed precision to
model formulation and parameter estimation when choice preferences are critical to
strategic decisions.
Conjoint analysis is actually a family of techniques and methods, all theoretically based
on the models of information integration and functional measurement. The theoretical
foundations for conjoint analysis are found in the seminal psychological research of
1 Molecules are available from http://www.vensim.com/molecule.html. From the tutorial distributed with
Molecules 1.4: “Molecules are the building blocks of good system dynamics models.
[Molecules]... provide a framework for presenting important and commonly used elements of model
structure...”
Luce and Tukey (1964), and thousands of applications of conjoint analysis have been
carried out over the past three decades by marketing scholars and practitioners.
Conjoint analysis is, by far, the most used marketing research method for finding out
how buyers make trade-offs among competing products and suppliers (Green, Krieger
and Wind, 2001). To establish a model of customer judgments, conjoint analysis
endeavors to unravel the value, or part-worths, that customers place on the product or
service attributes from experimental subject’s evaluation of profiles based on
hypothetical products or services (Green and Wind, 1975; Green and Srinivasan, 1978
and 1990). The experimental design and the assumptions conceming the model form
and types of relationships among variables are more important than the choice of
estimation technique. To this end, conjoint analysis places more emphasis on the ability
of the researcher or manager to theorize about the behavior of choice than it does on
analytical technique for estimating part-worths.
Despite its popularity by marketing scholars and practitioners, conjoint analysis does
not capture the ‘market’ dynamics of competition based on the underlying choice
preferences. Different product concepts are ‘tested’ by parameterizing each concept
along the full range of product attributes, and product attractiveness is computed based
on the estimated part-worths. Predicted market share is then a function of the choice
preferences, the hypothetical and/or actual product attributes, and scaling parameters to
correct for intended versus actual purchase probability. Product attributes are not
endogenous and the diffusion process over time is ignored entirely. Capturing the
dynamics of the competition between firms is clearly important for evaluating policy
options available to any individual firm in the industry, and this makes a persuasive case
for combining conjoint analysis and system dynamics to get the best out of both
approaches. We are not suggesting that every project with some aspect of choice
preferences should use conjoint analysis or a similar method. Instead, we maintain that
when model behavior is sensitive to choice preferences and policy recommendations are
not robust in the face of relatively small changes in preference formulations, then
conjoint analysis can add needed precision to formulations and increase confidence in
the system dynamics model. The next section discusses, in detail, the steps involved in
combining conjoint analysis and system dynamics in a modeling project with a high
tech SME.
3. Combining System Dynamics and C onjoint Analysis
In Spring 2001, we conducted a modeling project with a high tech SME in the
automotive industry. The SME, founded in 1994, was a fabless semiconductor
manufacturer specializing in System-On-Chip networking solutions for information and
entertainment systems to the automotive market. As the first company specializing in
the nascent in-car network market, the SME enjoyed early success by introducing its
technology, Digital Databus (D2B) Optical, at Mercedes-Benz in 1995. Further
adoption by carmakers of D2B stagnated until 1998 when Jaguar Cars adopted the D2B
technology for the new X-Type. However, the competitive landscape changed
dramatically in 1998 as a rival company emerged and successfully introduced their
technology (MOST) at BMW, Audi, Volvo, SAAB, and Daimler Chrysler. At the time
our project began, there were two critical questions under evaluation by the SME’s
management team:
1) Should the price of the D2B solution be decreased in an effort to increase
adoption of the technology by carmakers and what is the impact on the bottom-
line?
2) What product attributes should the Company develop in their next generation
platform and what is the likely market adoption of this new product?
Our modeling project got off the ground to help progress thinking about these issues
and, ultimately, to propose specific policy recommendations that had the potential to
substantially increase the value of the company.
3.1 Overview of the system dynamics model
The sector map in Figure 1 provides a high-level overview of the simulation model
developed with the SME’s management team to explore a variety of new product design
and pricing strategies. There are four sectors representing: 1) the automotive industry,
2) our SME, 3) the competitor and 4) the market for in-car networks.
Car manufacturer's demand
Timing of new vehicle programs.
Annual product volume
Time to adopt new technology
Time to develop/launch new car
Our client Producti ber segment Competitor
Produ “Huber of aplesins pra Product
+ Performance rice elasticity * Performance
+ Price + Price
* Consumer ublity * Consumer utlty
+ Revenues + Revenues
Costs Costs
cocs Market + cos
oe . +e
1 heeoune. Relative ‘our client’ / competitor 2 account
Management * Product performance Management
+ Program * Pricing + Program
Management 3 Management
ae Ce conrs ant
Valuati Valuati
eueren Market share heen
Installed base by technology
Figure 1: Model Sector Map
The automotive industry currently operates in a saturated market. In a struggle to
differentiate car models, carmakers are expected to introduce a variety of new electronic
devices that bring information and entertainment services, currently only available at the
home or office, into the car. Some of this technology, designed to attract potential
buyers, has already been adopted by some carmakers such as radar assisted cruise
control, night-vision system, or voice control. Many industry experts suggest that
within a couple of years new technologies will be standard in the vehicle - as common
as the car-radio today. Examples of such technology include navigation and voice
control systems and the ability to let passengers watch satellite fed movies or browse the
Intermet. As more of these in-car multimedia devices get bundled into the car, the
automobile must be equipped with a digital network to facilitate information exchange
between them.
The diffusion of these in-car multimedia networks across all carmakers is the primary
area of concern for the project reported here. Since the focus of this paper is on the
integration of system dynamics and conjoint analysis in this modeling project, we will
only discuss the details of the model relevant to this integration process. Adoption of
in-car networks by carmakers is crucial for the SME’s strategic issues and is the point
where the combination of methods proved quite valuable. The adoption process is
discussed in the next section.
3.2 Adoption of in-vehicle networks
Carmakers make the decision to adopt an in-car network solution as part of the
development process for a new vehicle. The pool of Potential Automotive Partners
Tepresents carmakers worldwide yet to adopt in-car networking technology, as shown in
the stock and flow diagram in Figure 2. These potential partners may, over time, adopt
an in-car networking solution and, upon adopting, become Partners in Development.
Once the decision to adopt an in-car network for a new vehicle has been made, there is a
10
substantial development delay before vehicle production. This delay is due to the
development cycle of a vehicle, which usually takes up to three years in Europe, four
years in the US and two years inJapan. The development partners, upon completion of
the new vehicle development, become Partners in Production as the cars are launched
into production. The Technology Replacement Rate represents carmakers phasing out
vehicle programs and becoming potential customers again as they replace the
technology in their next generation of cars.
Technology Replacement Rage
Avg Technology Lifeti
Avg Time for
Carmakers New Program
Vehicle Introductions Development
Potential af a)
Partners in Ea Partners in
Automotiv > >|
Partners Adoption Rate Development Development Production
a. Completion Rate
Product
Attractiveness Program Managment
Effectiveness
Figure 2: The in-car network technology adoption process
The Adoption Rate is dependent upon Product Attractiveness and New Vehicle
Introductions. New Vehicle Introductions represents an assumption regarding the
timing of new vehicle programs across all carmakers. There is considerable uncertainty
about the launch dates of new vehicle programs, and we therefore assumed the market
entry of new vehicle programs follows a normal distribution with mean and standard
deviation derived from interview data with carmakers about the likely timing of new
vehicles. Of course, carmakers may still not adopt an in-car network for their new
vehicles if product attractiveness is below acceptable levels. Product Attractiveness is a
11
multi-attribute measure of carmakers’ preferences regarding the in-car network
solutions available at any given time. Carmakers evaluate each in-car network based on
multiple product attributes in deciding whether to adopt an in-car network and if so,
which solution to adopt. These product attributes and the underlying choice preferences
specifying the importance of each attribute were identified through market research,
employing conjoint analysis, and are discussed in the next section.
3.3 Market research into product preferences
Suppliers in the automotive industry typically rely on concept testing to gauge
prospective customer reactions to prototype product ideas. This approach yields
valuable insights into the decision-making process of customers and reactions to a
particular prototype product, but it does not provide quantifiable trade-offs for
combinations of product attributes. For example, should the SME develop a 50Mbps
product and sell it for $5 or a 100Mbps product and sell it for $10? Which combination
creates more value?
In order to identify the product attributes carmakers considered in their choice of in-car
network solution and to assess the value tradeoffs among these attributes, we employed
the full profile conjoint analysis procedure to identify and measure carmakers’ product
attribute preferences for in-car networks. Surveys were sent to a large sample of
industry decision makers, and respondents were presented with a set of product concepts
containing the full range of product attributes. Respondents were asked to rank the
product concepts according to their “attractiveness”. The steps in a conjoint analysis
are: 1) Identifying the relevant product attributes and designing the survey
questionnaire, 2) Administering the survey and data collection, and 3) Estimating the
12
underlying customer utility functions and identifying market segments. Each step is
described in sequence in the following sections with details about how we implemented
conjoint analysis in our modeling project with the SME.
3.3.1 Identifying Relevant Product Attributes and Survey Design
The first step in conjoint analysis is to identify the set of independent product attributes
that are important to customers in making their choices about which products to adopt.
A preliminary list of the important product attributes emerged from interviews with
managers and employees of the SME who knew the product and market. These
‘experts’ identified five criteria that were subsequently discussed in another series of
interviews with managers from carmakers, 1“ tier suppliers, the IDB trade association,
and microelectronics companies- a broad range of in-car network ‘customers’. After
this second round of in-depth discussions, one of the attributes was dropped from the
initial list. These interviews were also used to identify the range of trade-off options for
each of the product attributes. It is important in this step to be very clear about the
implicit assumptions underlying the psychology of choice theory embedded in the
selection of product attribute options. Findings from consumer behavior research
indicate choice processes can be summarized as a two-stage process. In the first
conjunctive stage, the consumer eliminates options with one or more unacceptable
attribute levels. In the second compensatory stage, the options that remain are traded
off on the multiple attributes (Lussier and Olshavsky, 1979). Following conventional
conjoint study design, attribute levels were specifically chosen in our experimental
design such that there were no unacceptable levels. Table 1 shows the final set of
criteria and the available trade-off options.
Table 1: Product Attributes and Trade-off Options for the Market Research
13
Product Attribute / Research Criteria Trade-off Options
What is the required bandwidth? 20 Mbps 50 Mbps 100 Mbps
What is the preferred physical | Optical fiber | Unshielded Shielded
transmission medium? Twisted Pair | Twisted Pair
What is an acceptable node cost for the $5 $10 $15
network transceiver IC?
Is it important that an industry Yes No N/A
association endorses the technology?
The next step in this process is to design a survey questionnaire that can be sent to
(potential) customers for evaluation of the tradeoff options for all product attributes.
Their evaluation of these tradeoffs will allow us to empirically estimate customers’
underlying utility functions for each product attribute. The various options for all four
product attributes result in a total of 54 separate product concept combinations”, which
would result in quite a lengthy survey if we presented subjects with all 54 concepts.
Response rates tend to decrease with increasing questionnaire length, and more
importantly, research indicates that long questionnaires may induce response biases
(Lenk et al, 1996). To minimize the number of concepts presented to subjects, an
orthogonal array experimental design (Addelman, 1962) was employed to select a small
fraction of these 54 possible alternatives. Nine product concepts were selected that
were sufficient to estimate all four attribute-level main effects on an uncorrelated basis’.
It has been typical in conjoint studies to estimate only the main effects and assume away
interaction effects, and we follow this convention. In certain cases, interaction effects
may be important and in those cases the design must be adjusted to measure
interactions.
> The 54 possible product concepts are a result from multiplying the number of trade-off options per
individual product attribute (54 = 3* 3 *3 * 2).
14
A pilot survey questionnaire was tested on a sample of ten experienced industry
managers to determine whether the language and the instructions were clear. No
additions or deletions occurred in the list of criteria as a result of this process, but some
of the language describing the criteria was modified. It is important to clearly
communicate the definition of each product attribute on the survey in order to minimize
ambiguity in the respondent's mind concerning the trade-offs they are making. The
nine different product concepts were finalized as shown in Table 2.
Table 2: Nine different product concepts to choose from
Product | Supported | Supported Transportation | Costof | Technology is
Concept | Bandwidth Medium the endorsed as a
No. Network industry
Node standard
1 100+ Mbps | Shielded Twisted Pair (STP) $15 Yes
2 100+ Mbps Optical Fiber (POF) $5 No
3 50 Mbps Unshielded Twisted Pair $15 No
(UTP)
4 50 Mbps Optical Fiber (POF) $10 Yes
5 50 Mbps Shielded Twisted Pair (STP) $5 Yes
6 20 Mbps Optical Fiber (POF) $15 Yes
7 20 Mbps Unshielded Twisted Pair $5 Yes
(UTP)
8 100+ Mbps Unshielded Twisted Pair $10 Yes
(UTP)
9 20 Mbps Shielded Twisted Pair (STP) $10 No
3.3.2. Administering the Survey and Data Collection
After the survey questionnaire has been designed and pilot tested, the next step is to
send the survey to a large and representative sample of (potential) customers. Survey
research using unqualified leads typically yields a 2-3% response rate. We hoped to
° The presence of interattribute correlation does not violate any assumptions of conjoint analysis.
However, correlation among attributes increases the error in estimating preference parameters and
should be kept to a minimum.
15
increase the response rate by using the SME’s database of 500 industry and customer
contacts. In an attempt to increase the response rate further, we decided to conduct the
survey on the World Wide Web to make it easier for subjects to respond and to reduce
the time needed to complete the survey. The survey was hosted on the SME’s web site
and also on the IDB-Forum web site. The IDB-Forum is an industry association that
promotes the global integration of networking into vehicles, consumer electronics, and
automotive electronics. The IDB-Forum contacted its members via phone and
encouraged them to participate in the survey. Data gathering was confidential and
anonymous. Respondents were asked to rank-order the nine product concepts, listed in
Table 2, from one to nine- most (1) to least preferred (9). The choice of measure for
customer preference need not be ordinal. The alternative is to obtain a rating of
preference on a metric scale to obtain an indication of how much a customer prefers one
product concept versus the others. As always, each preference measure has certain
advantages and limitations. After two months of contacting and re-contacting
tespondents directly, 33 useable responses were received- a 6.6% response rate.
3.3.3 Estimating Attribute Preferences and Market Segmentation
The final step in conjoint analysis is to use the data from all completed surveys to
empirically estimate customers’ multi-attribute utility functions. Using respondents’
rank ordering of product concepts, we can estimate part-worth utility functions for all
product attributes using a modified form of analysis of variance specifically designed
for ordinal data. In practice, we estimated the part-worths using multivariate ordinary
16
least squares (OLS) regression‘ with the order-rankings for each product concept as the
dependent variables and the different levels of each attribute as independent variables.
This separate part-worth form is the most general, allowing for separate estimates for
each level so that the data determines the type of relationship for each factor (i.e. linear
or nonlinear). In the basic additive model, the most common in conjoint studies, it is
theorized that the respondent simply “adds up” the values for each attribute (part-worth)
to obtain the overall worth fora product concept. Let p =1, 2, 3, ...n denote the set of
attributes used in the study design. Let yjp denote the level of the pth attribute for the jth
stimulus. The additive part-worth model assumes that the preference s; for the jth
stimulus is given by:
5 = > fy(yio) (1)
P
where f» is a function denoting the part-worth of different levels of yj, for the pth
attribute. Strictly speaking, part-worth functions are evaluated at discrete levels for
each attribute, but interpolation is generally applied between levels of continuous
variables.
After estimating the individual respondent utility functions using the additive part-worth
model, the resulting individual respondent utility functions were subsequently used as
an input into an unweighted pairwise cluster analysis. The cluster analysis was used to
identify groups or clusters of respondents based on product preferences. Two clusters
clearly emerged from the analyses - the “mass-market segment” including high volume
manufacturers of low/medium and high-end cars such as VW, Ford, and GM, and the
“ We later re-estimated these part-worths using Montonic Analysis of Variance (MONAVOVA), and
found no significant differences.
17
“high-end segment”, including lower volume manufacturers of medium and high-end
cars such as Audi, Mercedes-Benz, and Jaguar. One representative individual utility
function was selected for each market segment, and these representative part-worth
utility functions for mass market and high-end segments are now discussed in detail°.
In this case, we did not have a hold-out sample to test the accuracy of the model since
this must be built in to the experimental design to collect data on additional product
concepts. However, it is always preferable to evaluate model goodness of fit not only
on the original stimuli, but also with a set of hold-out stimuli.
Part-worth estimates are on a common scale, and therefore we can compute the relative
importance of each factor directly. In this study, part-worths have been scaled so that
the lowest part-worth is zero within each attribute. The aggregate relative importance
for the pth product attribute is represented by the range of the part-worth values for all
stimulus levels (i.e. the difference between the lowest and highest value) divided by the
sum of the ranges across all factors:
. Range of Utility Values
Relative Importance, = e
n
2 Range of Utility Values,
Pp
where n is the total number of product attributes.
Figure 3 shows the aggregate relative importance of each product attribute for the mass-
market and high-end carmakers. It is obvious that the cost of a node on the network is
the most important criteria for both market segments. After agreement on the
5 All individual utility functions in each market segment were very similar. An altemative approach, and
the one typically employed in conjoint analysis research, is to include all of the individual utility
functions in the simulation model and compute market shares by aggregating the individual responses to
the chosen stimuli.
18
importance of node cost, the customer segments diverge in the importance placed on the
Temaining product attributes.
STANDARD
NODE COST TMASS-MARKET
(HIGH-END
MEDIUM
BANDWIDTH
0% 10% 20% 30% 40% 50% 60% 70%
Figure 3: Relative importance of attributes for mass market and high-end segments
Figure 4 shows the node cost utility function for “mass market” and “high-end”
companies- the function is identical for both market segments. A product that can
achieve a node cost of around $5 achieves maximum utility. The other end of the
spectrum at $15 is a “non-starter”.
UTILITY FUNCTION FOR PRODUCT
ATTRIBUTE "NODE COST"
1.0
08
- “iy
0.4
UTILITY
0.2
0.0
aS 10 15
NODE COST (DOLLARS)
Figure 4: Node Cost Utility Function for the mass market and high-end segments
19
The differences between the two segments emerge in the utility functions for the
remaining product attributes. We will discuss the mass-market carmakers utility
functions on the remaining three product attributes, and then discuss the corresponding
high-end carmakers utility functions. Space prevents us from including figures for all of
the utility functions for each segment, but each utility function will be discussed.
For mass-market companies, the supported transmission medium is the second most
important criteria with nearly all respondents preferring electrical over optical medium.
The choice of transmission medium is driven primarily by ease of maintenance and
Electromagnetic Compatibility (EMC) performance. Maintenance is generally
considered easier for copper wire versus optical fiber. Shielded Twisted Pair (STP) and
Unshielded Twisted Pair (UTP) are therefore easier to maintain than Plastic Optical
Fiber (POF). On the other hand, poor EMC performance might interfere with other
electronics such as the airbag or ABS system and also distort the FM radio band. POF
is the preferred solution for minimizing EMC.
The third most important criteria for mass market carmakers is whether the technology
is endorsed by an industry association, but this is much less important than Node Cost
and transmission Medium. Mass-market carmakers prefer in-car network solutions that
are endorsed by an industry association such as the Automotive Multimedia Interface
Collaboration or the IDB Forum.
Lastly, and somewhat surprisingly, the supported bandwidth of a particular solution is
not important to mass-market companies- the utility curve is flat at the value of zero.
The survey gave bandwidth trade-off options in the range from 20 Mbps to 100 Mbps,
20
and the finding that bandwidth is not important suggests that a bandwidth up to 20
Mbps is considered sufficient for mass-market companies.
For high-end carmakers, the maximum bandwidth offered by a particular product is the
second most important criteria. The utility function is shown in Figure 5. These
companies operate in the high-end, luxury segment of the car market and follow a
differentiation strategy. Part of this strategy is to offer their customers “cutting-edge”
technologies in their vehicles and typically these technologies require higher bandwidth.
We can see that utility increases by 40% for products with 50 Mbps versus 20 Mbps.
Above this point, the slope decreases somewhat, which indicates that a further increase
is less important. A doubling of bandwidth from 50 Mbps to 100 Mbps only yields an
additional 20% utility value.
UTILITY FUNCTION FOR PRODUCT
ATTRIBUTE "BANDWIDTH"
1.00
0.80
E 0.60
E 0.40 ge”
0.20 x
0.00
20 50 100
MBPS
Figure 5: Bandwidth Utility Function in the High-End Segment
The supported medium is the third most important criteria for high-end companies, but
is much less important than Node Cost and Bandwidth. In general, high-end companies
have a preference for POF, and this preference can be explained by the perceived
electromagnetic reliability of POF versus copper wire.
21
Finally, high-end carmakers do not attribute any additional value in products being
endorsed by an industry association- the utility curve is flat at the value of zero. The
rationale provided in some of the open-ended survey responses was that endorsement by
an industry association might detract from the ‘cutting-edge image’ of the technology
and dilute the differentiation benefits.
3.4 Integrating customer preferences into the system dynamics model
As a result of the market research, the management team gained crucial information
about the product attributes that carmakers considered important and the tradeoff values
among those attributes. Some of these insights about their customers’ utility functions
were altogether new for the management team or even went counter to their initial
expectations. For example, the management team discovered that the market is clearly
segmented and that some product features are not relevant for each market segment. In
addition, one product attribute that management initially felt was quite important for
carmakers was found to be unimportant in the choice preferences of both segments.
With these new insights, the system dynamics model structure was modified to account
for the two distinct customer segments, and the product attribute utility functions for
each segment were incorporated into model.
The utility functions were operationalized into the system dynamics simulation model in
the formulation for Product Attractiveness as shown in Figure 6°. This structure is
® The nonlinear utility functions can be implemented using graphical functions within Vensim, IThink or
Powersim, or using table functions in the Jitia™ Simulation Software.
22
replicated for each of the two customer segments in order to reflect the different
preferences in the “mass-market” and “high-end” segments.
Other Costs
Max Available Bandwidth
+
IR / i.
Bandwidth
Price Attractivness
Attractivness
Unit Price wN
Endorsement as
Industry Standard
+ +
+ Supported Medium
Standard i
Attractivness Medium Attractivness
Product Attractiveness
Figure 6: Product Attractiveness structure in the system dynamics model
Unit Price, Bandwidth, Transmission Medium and Endorsement from an industry
association, are exogenous model parameters that allow the management team to
experiment with different product concepts and pricing strategies to see how these
changes impact the competition for market share over time. An exact replica of this
structure is also operationalized for the competitor and is calibrated to reflect the actual
product attribute levels of their MOST solution at the time of this study. Product
Attractiveness is defined as the additive utilities of each individual product attribute.
The linear additive form is consistent with consumer behavior research on
compensatory choice processes in which acceptable product options are traded off on
multiple attributes (Lussier and Olshavsky, 1979). A number of studies have
demonstrated the ability of the basic additive conjoint model to predict actual behavior
23
(Green and Srinivasan, 1990), and attribute levels were specifically chosen in the
experimental design in this project such that there were no unacceptable levels.
Adopting the additive form has an important implication for the system dynamics policy
analysis. Specifically, it is clear that the operating conditions of the model are bounded
by the acceptable attribute levels. Unacceptable attribute levels would take the model
outside of its operating conditions and the resulting model behavior would not be
reliable.
While the additive model has been the most commonly adopted form in conjoint
studies, conjoint analysis is not limited at all in the types of relationships required
between the dependent and independent variables. Alternative functional forms can be
estimated using a wide range of conjoint methodologies (e.g. self-explicated, adaptive,
or choice-based conjoint models), but the most important consideration is the
experimental design to capture hypothesized decision making process. For example,
Product Attractiveness formulations in SD models are typically multiplicative models of
the decision making process. This formulation captures the conjunctive stage of the
choice process where customers eliminate options with one or more unacceptable
attribute levels (e.g. very high delivery delays or very low product quality). Conjoint
analysis can certainly be used to estimate the multiplicative model using more complex
statistical techniques’, provided the collected respondents’ evaluative data permits such
estimations. In other words, the most important difference from the process described
above is that the experimental design would need to be modified to ensure testing
unacceptable stimuli and extreme attribute levels.
’ For example, using nonlinear regression to estimate the part-worths for each attribute level given the
independent attributes are now multiplied to compute overall utility. Initial parameter values must be
given to start the curve fitting process, and care must he taken to avoid local optima.
24
The resulting system dynamics model provided a solid framework to address the SME’s
key strategic questions. The next section discusses a number of simulation experiments
designed to explore a variety of pricing and new product development policy options.
4. Simulation Results - Designing a robust pricing and product
development strategy
This section describes how the system dynamics model was used with the SME’s
management team to address their strategic questions. The management team readily
acknowledged that the SD model represented a quantum improvement in comparison to
a spreadsheet based (financial) model in terms of exploring and evaluating strategic
options. They now had a tool where their assumptions and the preferences of their end-
customers were explicitly captured, and that could be used to explore the dynamic
behavior of the market place. Some of the policy analysis insights contradicted their
initial expectations about the policies that would add value for the firm. For example,
counter to their previous beliefs, simulations demonstrated that decreasing the price of
their existing in-car network solution to target on market segment and introducing a new
product for the other segment can potentially double the value of the business. The
following pages discuss a number of the simulation experiments.
4.1 Base-Case: Business as usual
The base case represents a reference simulation in which the model has been
parameterized to reflect the competitive environment for the in-car network SME as of
May 2001. At the time of this study, the worldwide market for in-car multimedia
25
networking was split between two companies - the SME and one competitor. The SME
currently offers the D2B optical solution and has developed a new solution, branded
D2B Smartwire, which will be launched sometime in 2002. The SME’s D2B
technology has a specific bandwidth and transmission medium. These attributes are
fixed as part of the design of the solution, and any changes result in the development of
anew technology. Regarding the other product attributes, management directly controls
the node cost, or price, and can certainly influence whether or not a trade association
endorses the product. The competitor offers the MOST in-car network solution. The
MOST and D2B product attributes for the base case are given in Table 3. Itis important
to note that product prices are not transparent, and therefore management made an
assumption that the prices are equal to see the impact of the other attributes.
Table 3: Product attributes in the base case
Product Configurations |MOST _D2B
Max. Bandwidth 20Mbps_ 6 Mbps
Supported Medium POF POF
Node cost $10.59 $ 10.59
Endorsed as a standard No No
The time horizon of the model covers the 15 year period from 2001 - 2015 to include
two vehicle lifecycles; one vehicle lifecycle is assumed to be six years on average.
Figure 7 shows the adoption of D2B technology by carmakers including: a) Potential
Automotive Partners, b) Partners in Development, and c) Partners in Production. The
Base Case indicates a peak adoption rate for high-end firms in 2003, and mass-market
firms do not enter the market at all. Gradual introduction for different vehicle models at
one carmaker results in a partial Automotive Partner adoption rate. In the base case, our
SME should expect to have approximately four automotive partners in production and
two partners in the development phase by 2015.
26
Carmakers adopting the D2B Technology
High-end Segment only
S
. \ Potential Automotive Partner
s
N
XN
N
- \ Partner in Production
q
%
g
9
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Time (Year)
Figure 7: Base Case- Adoption of SME (D2B) Technology
The effects on Revenue and Earnings before Interest and Tax (EBIT) are shown in
Figure 8.
Total Revenue and Operating Profit
in the Base Case
40m om
Operating Profit
30M sm @
; + |
i aa poo 77 Revenue sa ?
I
5M
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Time (Year)
Figure 8: Base Case - SME Revenue and EBIT
27
The base case simulation clearly demonstrated to the SME’s management that they
would need to make a strategic move in order to reestablish their position as the leading
in-car network company with carmakers. They had aspirations of capturing more than
50% of the 20 worldwide carmakers as partners, and the Base Case market share of 20%
is along way from that goal.
4.2 Penetration Pricing Strategy Targeting the Mass-Market Segment
The current products, D2B and MOST, have been primarily targeted at high-end
carmakers, and are too expensive to extensively penetrate the mass-market segment. In
fact, neither company has addressed the specific needs of the two distinct segments very
effectively with the current products. During the two years from 1999 - 2001, the SME
developed a new technology that will allow high data-rate transmission of signals via
electrical cables. D2B SmartWire, the name for the new technology, will reduce system
costs and better address the needs of the mass-market segment. However, our market
research indicates that adoption by mass-market carmakers will be quite limited at
current price levels.
Figure 9 shows that in the absence of a competing low price product, introducing D2B
SmartWire as early as 2002 at a price of about $3 can successfully stimulate adoption by
mass-market companies. In contrast to the base case, the penetration strategy results in
a significant number of the mass-market carmakers adopting the D2B technology. The
Adoption Rate peaks between 2005 and 2006 with slightly more than four carmakers
entering the market. After the market becomes mature in 2010, there are on average 1.5
carmakers replacing the technology. Given no competition in the low price network
28
solution, the SME could expect to partner with up to eight carmakers in production in
the Penetration Pricing Strategy.
Carmakers adopting the D2B Technology
Mass-Market Segment only
Potential Automotive Partner
: rl
Partner in Production
in Development
°
2001 2002 2003 2004 2005 ©2006 2007 2008 2009 2010 2011 2012 2013 2014 2015,
Time (Year)
Figure 9: Penetration Pricing- Adoption of D2B Technology (mass-market segment)
There is a notable difference between the Base Case and this Penetration Pricing
Strategy on the effect on the financials of the company- Revenues and EBIT triple. The
impact of this strategy on EBIR is shown in Figure 10.
29
Total Operating Profit
for a Penetration Pricing Strategy Targeting the Mass-Market Segment
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Time (Year)
Figure 10: Penetration Pricing Strategy EBIT
43 New Product Development & Launch Strategy
The SME’s current product offering, D2B Optical, is undifferentiated and does not offer
sufficient bandwidth for high-end firms. At the same time, the solution is too expensive
for mass-market firms. The Penetration Pricing Strategy just discussed enables the firm
to open the mass-market customer segment with an extension of the existing product-
D2B Smartwire. This leaves scope for new product development specifically targeting
the high-end segment. Based on the insights from the market research and conjoint
analysis, we worked with the management team to determine the ‘ideal’ attributes for a
new product targeted for the high-end segment. This process suggested that the new
product should offer a supported bandwidth of at least 50Mbps for a node cost of
approximately $6 per unit. The transmission medium of the solution would be electrical
instead of optical. This combination of product attributes maximizes utility given the
constraints of the SME and the costs of the eventual product.
30
Given the lead times for new product development, we assume that such a product could
be available in 2004. According to the utility functions obtained in market research, this
product would become a clear segment leader with roughly 65% market share in the
high-end segment. The availability of this new, superior product would result in a
sudden increase in the high-end Adoption Rate by about 30%. In the long mun, this
results in a permanent increase in D2B Partners in Development and D2B Partners in
Production. As shown in Figure 11, the New Product Development (NPD) & launch
strategy results in improved revenue and EBIT.
Total Revenue and Operating Profit
Under New Product Development & Launch Strategy in the High-End Market
200m ee 777] em
oc
o7 7
Revenue 4 perating Profit
7
75M : com @
, ee--c- Pe
a } we
rT J eid i
¢ -
i som te an 2
WA
‘-
« x lp H
thf Arrows Indicate the
7
25M Pia financial gain comparative 20M
-
4 to the Penetration Strategy.
---25
0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Time (Year)
Figure 11: Financials for NPD & Launch Strategy in the high-end segment
5. Discussion & Conclusions
This paper provides a detailed step-by-step approach for integrating system dynamics
and conjoint analysis to leverage the strengths of both. This process was illustrated
31
through a modeling project in which we developed a small, high-level SD model of the
automotive industry incorporating customer preferences derived through conjoint
analysis to determine the success of competing products. In this case, management’ s
information regarding customer preferences was quite limited and therefore it was
absolutely necessary to do some market research in order to guide formulation of the
system dynamics model. Perhaps this situation is more prevalent in SME’s that have
fewer resources to allocate to market research than large, established firms. Market
research into customer preferences demonstrated that management's initial beliefs about
the set of key attributes customers felt were critical and the relative importance of those
attributes, were quite wrong. The market research had a significant impact on both
model structure and parameterization. If the system dynamics model had been based
only on management's judgments, the resulting policy analysis would have been
erroneous and may have been harmful for the SME. The combination of methods
proved persuasive with the management team, and resulted in a better system dynamics
model and therefore improved insights for the company.
A number of tangible and timely recommendations emerged from the simulation
experiments designed to evaluate different policy options. First, we simulated the
current Base Case strategy of offering an undifferentiated product at a high-price point,
and demonstrated the Base Case strategy offered minimum growth potential. After
testing a variety of other policy options, we produced a new strategy for the SME
automotive supplier that has the potential to double the value of its business. The first
recommendation involved refocusing the existing product, by lowering the price
significantly, to penetrate the large and growing mass-market segment. The second
prong of this strategy focuses on introducing a new product to meet customer
32
preferences in the high-end segment. It is quite clear that we could not have found
robust and reliable quantitative answers to the SME’s critical strategic questions by
using a system dynamics model or a conjoint analysis market simulator in isolation.
Integrating these methods unleashed substantial synergies, and we feel this combination
could be successfully applied to a broad class of problems where behavioral policies of
decision makers include tradeoffs among multiple attributes.
Of course, there are a number of open issues for future research to resolve about the
process of combining these methods. The most important issue is clarifying the
theoretical underpinnings of the choice process to guide our Product Attractiveness
formulations as multiplicative, additive, or otherwise. Researchers in marketing,
psychology and economics, now have a clearer understanding of multi-attribute decision
making, but are still a long way from a consensus view on the best way to portray the
choice process. We need to understand the contexts under which each formulation has
an advantage in predicting choice behavior. As system dynamicists, we have been
focused primarily on reference points and extreme attribute levels to guide our
formulations for nonlinear utility functions. This process helps ensure our models are
robust to extreme conditions tests, but may be sacrificing predictive power in the
acceptable attribute range if we do not appropriately capture the compensatory tradeoff
process among attributes. Another important issue for our community is to understand
how often and under what conditions ‘experts’ judgments regarding customer
preferences are unreliable. It may well be more often than we suspect, and in that case
the importance of integrating an appropriate method for dealing with the multi-attribute
choice problem becomes altogether more important. One last issue deals with the
change in customer preferences over time. New technology is developed, unexpected
33
industries converge, and social forces continuously shape cultural values. Perhaps the
best way to handle these inevitable developments is by updating customer preference
functions over time to keep abreast of shifts in choice preferences.
References
Addelman, S. (1962), Orthogonal Main-Effects Plans for Asymmetrical Factorial
Experiments, Technometrics, 4, pp. 21-46.
Backus, G. M. T. Schwein, S. T. Johnson and R. J. Walker (2001), Comparing
expectations to actual events: the post mortem of a Y2K analysis, System
Dynamics Review, Vol. 17, No. 3, pp. 217-235.
Dolan, R. J. (1999). Analyzing Consumer Preferences. Harvard Business School Note 9-
599-112. Harvard Business School. USA
Doman, A., M. Glucksman, N. Mass and M. Sasportes (1995), The dynamics of
managing a life insurance company, System Dynamics Review, Vol. 11, No. 3,
pp. 219-232.
Ford, A. (1995), Simulating the controllability of feebates, System Dynamics Review,
Vol. 11, No. 1, pp. 3-29.
Forrester, J. W. (1969), Urban Dynamics. Cambridge: MIT Press; Currently available
from Pegasus Communications: Waltham, MA.
Gardiner, P. C. and A. Ford (1980), Which Policy Run is Best, and who says so?, TIMS
Studies in Management Sciences, 14, pp. 241-257.
Green, P. E. and A. M. Kreiger (1988), Choice Rules and Sensitivity Analysis in
Conjoint Simulators, Journal of the Academy of Marketing Sciences, 16
(Spring), pp. 114-127.
34
Green, P. E., A. M. Krieger and Y. Wind (2001), Thirty years of conjoint analysis:
reflections and prospects, Interfaces, Vol. 31(3) May-June, pp. S56-S73.
Green, P. E and V. Srinivasan (1978), Conjoint Analysis in Consumer Research: Issues
and Outlook, Journal of Consumer Research, 5 (September), pp. 103-123.
Green, P. E and V. Srinivasan (1990), Conjoint Analysis in Marketing: New
Developments With Implications for Research and Practice, Journal of
Marketing, 54, pp. 3-19.
Green, P. E. and Y. Wind (1975). New Way to Measure Consumers’ Judgments,
Harvard Business Review, 53 (July-August), pp. 107-117.
Hines, J., B. Eberlein, G. Richardson, D. Johnson, B. Richmond, and J. Melhuish
(1995), Modeling with Molecules 1.4, LeapTec and Ventana Systems; Available
from http://www.vensim.com/molecule.html.
Homer, J. (1996), Why we iterate: scientific modeling in theory and practice, System
Dynamics Review, Vol. 12, No. 1, pp. 1-19.
Homer, J. (1997), Structure, Data and Compelling conclusions: notes from the field,
System Dynamics Review, Vol. 13, No. 4, pp. 293-309.
Lenk, P. J; W. S. DeSarbo, P. E. Green and M. R. Y oung (1996), Hierarchical Bayes
conjoint analysis: Recovery of partworth heterogeneity from reduced
experimental designs, Marketing Science; Vol. 15, Iss. 2; pp. 173-192.
Luce, R. D. andJ. W. Tukey (1964), Simulataneous conjoint measurement: a new type
of fundamental measurement, Journal of Mathematical Psychology, Vol. 1, pp.
1-27.
Lussier, D. A. and R. W. Olshavsky (1979), Task Complexity and Contingent
Processing in Brand Choice, Journal of Consumer Research, 6 (September), pp.
154-165.
35
Maier, F. H. (1998), New Product diffusion models in innovation management- a
system dynamics perspective, System Dynamics Review, Vol. 14, No. 4, pp. 285-
308.
Mayo, D. D, M. J. Callaghan and W. J. Dalton (2001), Aiming for restructuring success
at London Underground, System Dynamics Review, Vol. 17, No. 3, pp. 261-289.
Nuthman, C. (1994), Using human judgment in system dynamics models of social
systems, System Dynamics Review, Vol. 10, No. 1, pp. 1-27.
Oliva, R. andJ. D. Sterman (2001), Cutting corners and working overtime, Management
Science, Vol. 47(7), pp. 894-914.
Paiche, M. and J. D. Sterman (1993), Boom, Bust, and failures to learn in experimental
markets, Management Science, Vol. 39(12), pp. 1439-1458.
Piatelli, M. L., M. A. Cuneo, N. P. Bianci and G. Soncin (2002), The control of goods
transportation growth by model share re-planniing: the role of a carbon tax,
System Dynamics Review, Vol. 18, No. 1, pp. 47-69.
Reagan-Cirincione, P., S. Schuman, G. P. Richardson, S. A. Dorf (1991), Decision
Modeling: Tools for Strategic Thinking, Interfaces; Vol. 21, Iss. 6; pp. 52-66.
Sterman, J. D. (2000) Business Dynamics: Systems Thinking and Modeling for a
Complex World. McGraw-Hill.
Back to the Top
36
CC BY-NC-SA 4.0