Professional Documents
Culture Documents
1991;8:189-202
189
0000
Designing optimal positioning strategies for target segments has become an area of intense research interest over the past few years. This article examines a variety of strategies that can be
operationalized from conjoint analysis input
data. Paul Green and Abba Krieger discuss strategies for modifying buyer perceptions, ideal-level
preferences and attribute importances that are
attractive for a firm's existing brand(s). They
then consider longer run strategies for modifying
the current brand's attribute levels and develop a
case to illustrate applications of the techniques.
Address correspondence to Paul E. Green, S.S. Kresge Professor of Marketing, Marketing Department, Suite 1400, The Wharton
School, University of Pennsylvania, Steinberg-Dietrich Hall, Philadelphia, PA 19104-6371.
1991 Elsevier Science Publishing Co., Inc.
655 Avenue of the Americas, New York, NY 10010
Introduction
Few marketers would dispute the central roles
that product positioning and market segmentation
play in today's economies, national or global.
The last decade has witnessed a surge of new
methodological and modeling developments for
optimal positioning. Reviews of these developments are provided by Green, Carroll and Goldberg [2], Sen [9], Sudharshan, May and Shocker
[12], Kohli and Krishnamurti [7], Sudharshan,
May and Gruca [11], and Green and Krieger [3].
Conjoint analysis [4] has been central in the
implementation of many optimal positioning
models. The advent of user-friendly personal
computer packages, such as those of Herman [5],
Johnson [6], and, most recently, SPSS [10], has
served both to define the state of conjoint practice and to make the methodology accessible to
wider audiences.
Helpful as they are, however, commercial conjoint packages focus largely on data collection
and part-worth estimation. While all three of the
packages cited above contain buyer choice simulators, none deals with product optimization and
related issues.
The purpose of the current article is to discuss
various extensions of current product positioning
practice. As in the case with commercial packages, the methodology described here collects
conjoint preferences and respondent background
data. Attribute-level perceptions data for each respondent's most preferred brand are also collected. (Perceptions data collection represents a
departure from the commercial packages' input
0737-6782/91/$3.50
190
BIOGRAPHICAL SKETCHES
Paul E. Green is the S.S. Kresge professor of marketing at
the Wharton School, University of Pennsylvania. Dr. Green's research emphasizes quantitative methods and new measurement techniques in market analysis and consumer research. He has been honored for this research by the American Marketing Association, the
American Statistical Association, the American Psychological Association, the Market Research Society, and the Association for Consumer Research.
Dr. Green is the author or coauthor of several books, including the
widely used text Research for Marketing Decisions, now in its fifth
edition. He is also a prolific contributor to marketing and business
journals.
Abba M. Krieger is professor of statistics and operations research at the Wharton School, University of Pennsylvania. He is the
author or coauthor of articles in statistical methodology and the interface between statistical methodology and optimization theory. His
current research interests include theoretical and empirical analyses
of the bootstrap resampling technique and application of statistical
methods and operations research to problems in marketing research.
Target Marketing
Up to this point we have been assuming that Epsilon's product positioning strategy is based on
total market considerations. We now focus on
product design for selected target segments. The
following questions are appropriate:
1. We first assume that Epsilon m a n a g e m e n t
wishes to consider a line extension for a preselected target segment. What should this card's
profile be?
2. We next assume that Epsilon wishes to choose
two target segments and replace its current
card with two new cards, one for each segment. What should their profiles be?
The preceding product positioning and target
marketing actions are only illustrative of the possibilities for which the proposed methodology is
designed. T h e y should be sufficient, however, for
illustrating the versatility of the approach.
The Survey
Epsilon researchers next designed and implemented a national survey of credit card holders,
J PROD INNOV M A N A G
1991 ;8:189-202
191
Cost Data
Obtaining Epsilon's cost data at the attribute
level was a m u c h more difficult job. Our interest
was restricted to variable costs only; even at that,
crude estimates had to suffice. Each of the " n u l l "
or base levels (lowest part-worth levels) was
coded at zero cost, and costs of enhanced levels
were expressed as (negative) departures from this
base. Annual price was translated into annual
gross revenue for the base profile; costs for enhancements were deducted from their revenue
estimates.
The results of these estimation procedures provided Epsilon researchers with an annual contribution to overhead and profit, estimated on a percard level. (The contribution includes estimated
interest charges.)
192
J PROD 1NNOV M A N A G
1991 ;8:189-202
P . E . G R E E N AND A. M. K R I E G E R
$0
$10
$20
$50
$80
$100
1/2%
1%
9-5pm weekdays
90 days coverage
Scale
Values
0.50.4-
$~""~ *'o
0.3-
0.2-
0.1-
~S0
J1%
None 12%
,~..~ ~ - ,
, , ,
Price
Rebate
~4hr~
Nolle 9-5prn
~ i '~,
Message
Forwarding
,$~k~200k
ays No~ne
l
Purchase
Insurance
Carrier
Insurance
2
1
I
Car
Insurance
0.50.4-
Ye~
0.3-
AHCRG
~"~CRD
0.2-
/el
/,
0.1
,
Baggage
Insurance
Airport
Clubs
,
Acceptance
Medical/
Legal
Yu
Limousine
/,
Emergency
Car Service
193
Attribute
Price
Cash rebate
Message forwarding
Retail purchase insurance
Common carrier insurance
Rental car insurance
Baggage insurance
Airport clubs
Acceptance
Medical/legal
Airport
Limousine
Emergency car serv.
Current mkt. share
Alpha
Beta
Gamma
Delta
Epsilon
$80
none
9-5
weekdays
90 days
$200,000
$30,000
$2,500
rep.
none
AHCRG
yes
yes
20% disc.
yes
20% disc.
28%
$20
none
$20
none
24 hours
daily
none
$50,000
none
$2,500
dep.
$2/visit
AHCRD
yes
yes
20% disc.
yes
20% disc.
45%
$0
0.5%
$10
none
9-5
weekdays
90 days
none
none
none
90 days
$200,000
$30,000
$2,500
rep.
none
AHCRD
no
none
none
8%
none
90 days
none
none
none
none
AHCR
no
yes
20% disc.
yes
20% disc.
9%
none
$5/visit
AHCRD
yes
none
none
10%
(1)
U~ (Js)
s
E U~ (Js)
(2)
S=I
w h e r e the e x p o n e n t a is c h o s e n to be non-negative.
As alpha a p p r o a c h e s infinity the preceding
choice rule mimics the m a x utility rule. Smaller
values o f alpha can mimic a logit rule. If alpha is
set to 1.0, we have the B r a d l e y - T e r r y - L u c e rule.
A c o m p u t e r program, called A L P H , has been
designed to find the value of alpha that, if applied
to e a c h individual's utilities, obtains the closest
overall fit to actual m a r k e t shares for each supplier. Exhibit 1 provides a brief description o f the
program. W h e n applied to the current survey
data, the best fitting value o f alpha was 10.4. This
is the value used in all s u b s e q u e n t analyses.
194
Input
Output
ALPH
PERCEPT
E a c h individual's current
brand identification
Individual's importance
weight file
Individual's part-worths file
Individual's demograhics file
Individual's perception vector
for current brand
Finds s u m m a r y , by brands, of
incidence of each perceived
attribute level and associated
average part-worth
C o m p u t e s sensitivity of
average part-worths as
perceived levels become
increasingly similar to
targeted levels
Can analyze data for the
overall profile, subsets of
attributes, or individual
attribute perception changes
Analysis can be performed at
total-market or by selected
demographic level
Size of demograhic s e g m e n t
Relative incidence of brands
within s e g m e n t
Associated brand utility
changes as perceptions
approach targeted levels
Sensitivity analysis by
individual attributes
S u m m a r y tables of
misperceptions, by attribute
and supplier
IDEAL
N e w matrix of individual
partworths
This n e w file is then
p r o c e s s e d by S I M O P T for
c o m p a r i s o n with original
partworths file
SALIENCE
Individual attribute
importances and
attribute-level desirabilities
Individual importance weight
file
Demographic (background)
file
Current market shares of all
suppliers
E a c h supplier's profile
Value of alpha and
demographic attribute weights
Attribute level cost/return
data
N e w set of attribute
importances
Predicted share/return if n e w
importances are u s e d
N e w part-worth matrix with
modified importance weights
for input to S I M O P T or
conventional simulators
Exhibit 1.
195
(continued)
Program
Input
Output
SIMOPT
SEGUE
3. Change consumers' perceived attribute importances in ways that increase the relative attractiveness of the firm's product.
The PERCEPT program (Exhibit 1) has been designed to examine consumers' perceptions of
product attribute levels for their current (or most
preferred) brand and to find those attributes that
196
Alpha
Gamma
Epsilon
Mean
.58
.84
.93
.78
.76
.64
.76
.72
.53
.63
.66
.61
.75
.70
.87
.77
.82
.60
.73
.72
Attributes
6
7
.73
.67
.95
.78
.90
.82
.90
.87
10
11
12
Mean
.50
.73
.90
.71
,73
.53
.80
.69
.57
.67
.78
.67
.69
.66
.88
,74
.92
.91
.80
.88
.71
.70
.83
9.8%
10.8
7.8
offer the greatest gains in utility for moving perceptions to the " c o r r e c t " levels of the firm's
brand. Input data consist of each respondent's
multiple choice data representing the level within
each attribute that is perceived to match most
closely the consumer's perception of his/her current brand.
PERCEPT plays two roles. First, it provides
summary descriptions, by brand, of consumer
perceptions versus "reality" levels. In this way,
one can compare brand efficiencies with regard to
how one's advertising messages are getting
across. Alternatively, one can compare attributes
(across brands) in terms of the accuracy with
which their levels are perceived.
Second, PERCEPT can be used to examine the
differential attractiveness of correcting consumer
misperceptions, so that the firm can focus on
those attribute levels that carry the highest payoff
in consumer utility (and, other things being equal,
brand choice). For illustrative purposes, we select three of the five suppliers: Alpha, Gamma
and Epsilon. Table 3 summarizes the results.
We first note that, on average, Epsilon users
show the highest incidence of accurate perceptions; 83% of the time they correctly perceive the
levels of Epsilon's true profile. This compares to
71% for Alpha and 70% for Gamma. In all three
cases, however, the perceived modal profiles corresponded to the true profiles of each supplier.
We also note from Table 3 that, on average, attribute 12 (emergency card service) and attribute
7 (baggage insurance) were most often perceived
correctly, while attribute 3 (message forwarding)
was least often perceived correctly.
In addition, Table 3 indicates that the achievement of perfect agreement between true and perceived attribute levels would increase average
utilities by only 8-11%. We also note that the two
most attractive attributes for Epsilon to work on
are card acceptance and medical/legal network.
These two attributes alone count for almost all of
Epsilon's potential utility increase of 7.8%.
All in all, we conclude that perceptions are reasonably accurate, particularly for Epsilon. Moreover, the gain in average supplier utility for correcting misperceptions is relatively small
(8-11%). Finally, if Epsilon's advertising emphasizes its breadth of acceptance and availability of
medical/legal services, utility gains could be
forthcoming in the future. Finally, it is worth noting that none of the commercial conjoint packages deals with the analysis of perceptions data.
Their simulators all assume that each consumer
correctly perceives each supplier's profile, as defined by the program's user.
PRODUCTDESIGNSTRATEGIESFOR POSITIONING
J PRODINNOVMANAG
1991;8:189-202
I
20
i
60
I
40
Alpha
Gamma
Alpha
Gamma
80
I I~"
197
100
IEpslln
t~"
Alpha
III
Gamma
I
Figure 2. Stacked bar chart showing
market share changes related to various
strategies (total market)
Gamma
Alpha
IV
~"
(~*
Replacement for
denoted ~ *
~**
Addition to ~ ,
denoted (~**
(~,
/~
Alpha
Gamma
~"
Alpha
VI
0
i /~*
20
Gamma
410
~"
60
(~
**
80
Replacement for ~,
denoted~*
100
198
P . E . G R E E N AND A. M. KRIEGER
pursuing the strategic objective. Of course, further study would also consider additional factors
related to payoff estimation and cost outlays.
J PROD INNOV M A N A G
1991:8:189-202
199
Current epsilon
Price
Cash Rebate
Message Forwarding
Retail Purchase Insurance
Common Carrier Insurance
Rental Car Insurance
Baggage Insurance
Airport Clubs
Acceptance
Medical/legal
Airport Limousine
Emergency Car
Service
$10
none
9-5 weekdays
90 days
none
none
none
$5/visit
AHCRD
yes
none
none
Replaced or
extended
epsilon
$100
1%
9-5 weekdays
90 days
$200,000
none
$2,500 dep
$2/visit
AHCRD
yes
none
yes
(20% disc)
Current beta
Replacement
beta
$20
none
none
90 days
$200,000
$30,000
$2,500 rep
none
AHCRD
no
none
$20
none
9-5 weekdays
none
$200,000
none
$2,500 dep
$2/visit
AHC
yes
none
yes
(20% disc)
none
Some Caveats
Although SIMOPT is a very flexible algorithm for
finding optimal product positionings, it should be
borne in mind that the mechanism underlying the
model does not consider the gestation period over
which share changes take place. SIMOPT also
takes into consideration existing products' shares
and an opportunity to optimize, conditional on
some attribute levels remaining fixed; however, it
does not explicitly deal with the costs of product
proliferation and the possible dynamic actions
that competitors might take even before new
products are introduced. Thus, in practice, forecast conditions might never be reached, due to a
variety of market factors not under the firm's
control.
Finally, the SIMOPT model assumes that consumers' part-worths remain stable over the planning horizon, that all relevant attributes appear in
the model, that costs are accurately measured at
the attribute level, and that each firm enjoys a
rough parity with respect to advertising and distribution levels. Clearly, these are important assumptions; their recognition underscores the
need to consider SIMOPT (and our approach in
general) as a set of planning models whose outputs need to be checked against other independent information sources and management judgment.
200
P . E . GREEN AND A. M. K R I E G E R
Price
Cash Rebate
Message Forwarding
Retail Purchase Insurance
Common Carrier Insurance
Rental Car Insurance
Baggage Insurance
Airport Clubs
Acceptance
Medical/legal
Airport
Limousine
Emergency Car
Service
Current epsilon
Up-scale market
$10
none
9-5 weekdays
90 days
none
none
none
$5/visit
AHCRD
yes
$100
none
24 hours/day
90 days
$200,000
$30,000
$2,500 rep
$2/visit
AHCRG
yes
yes
(20% disc)
yes
(20% disc)
none
none
Psychographic
segment 1
Psychographic
segment 2
$100
none
24 hours/day
90 days
$200,000
$30,000
$2,500 rep
$2/visit
AHCRD
yes
$80
none
9-5 weekdays
none
$50,000
$30,000
$2,500 rep
$5/visit
AHC
yes
none
yes
(20% disc)
none
none
201
Like any model, SIMOPT no doubt will be modified and extended as further information about its
performance and user reception is obtained. Still,
the model provides an example of how current
conjoint-based simulators can be extended well
beyond their traditional applications to estimating
market shares for a few user-selected profiles.
Concluding Caveats
Throughout, we have also discussed various caveats in the use of the proposed techniques, such
as difficulties in modeling the time trends over
which various predicted outcomes take place, the
possible omission of important product attributes, and various aspects of competitive retaliation.
Many of these limitations stem from the reliability/validity of conjoint input itself. 2 Green and
Srinivasan [4] stress the importance of conjoint
validation experiments and report some recent
findings at the academic and industry levels.
As simulators (and optimizers) routinely yield
market share forecasts, there is always the danger of spurious precision. Our own experience
supports the value of using calibration (e.g., current) market shares, as is done in the SIMOPT
model. The model's forecasts are then anchored
to these base levels, so that the researcher is, in
effect, reporting relative (rather than absolute)
levels.
We also need ways to insure that all important
attributes are measured, that costs can be accurately estimated at the attribute level, that marketing mix variables (e.g., distribution levels) can
be accommodated, and that dynamic competitive
effects can be estimated and included in the
models. Clearly, there is much work left to be
done.
T h e a u t h o r s w o u l d like to a c k n o w l e d g e s u p p o r t o f the Citib a n k F e l l o w s h i p f r o m the Sol C. S n i d e r E n t r e p r e n e u r i a l
C e n t e r a n d the S E I C e n t e r for A d v a n c e d Studies in M a n a g e m e n t , b o t h at t h e W h a r t o n School.
References
1. Green, P. E. Hybrid models for conjoint analysis: An expository review. Journal of Marketing Research 21 : 155-159 (May
1984).
2 To this end, researchers frequently run split-half reliability
tests to check on the stability of shares, returns, etc. over changes in
sample size and composition.
202