Anda di halaman 1dari 77

Segmentation and Targeting

Market Definition
Segmentation Research and Methods
Behavior-Based Segmentation
Market Segmentation
Market segmentation is the
subdividing of a market into
distinct subsets of customers.
Members are different between
segments but similar within.
Segmentation Marketing
Differentiating your product and
marketing efforts to meet the
needs of different segments, that
is, applying the marketing
concept to market segmentation.
Primary Characteristics
of Segments
Basescharacteristics that tell us why
segments differ (e.g. needs, preferences,
decision processes).
Descriptorscharacteristics that help us find
and reach segments.
(Business markets) (Consumer markets)
Industry Age/Income
Size Education
Location Profession
Organizational Life styles
structure Media habits
A Two-Stage Approach
in Business Markets
First stage/rough cut
Firm size
Second-stage/fine cut
Different customer needs, wants, values within macro-segment
Relevant Segmentation Descriptor
Variable A: Climatic Region
1. Snow Belt
2. Moderate Belt
3. Sun Belt
Fraction of
Likelihood of Purchasing Solar Water Heater
0 100%
Segment 1
Segment 2
Segment 3
Likelihood of Purchasing Solar Water Heater
Irrelevant Segmentation Descriptor
Fraction of
0 100%
Variable B: Education
1. Low
Moderate Education
3. High
Segment 1
Segment 2
Segment 3
Variables to Segment
and Describe Markets
Consumer Industrial
Segmentation Needs, wants benefits, Needs, wants benefits, solutions to
Bases solutions to problems, problems, usage situation, usage rate,
usage situation, usage rate. size*, industrial*.
Descriptors Age, income, marital status, Industry, size, location, current
Demographics family type & size, supplier(s), technology utilization,
gender, social class, etc. etc.
Psychographics Lifestyle, values, & Personality characteristics of
personality characteristics. decision makers.
Behavior Use occasions, usage level, Use occasions, usage level,
complementary & complementary & substitute
substitute products used, products used, brand loyalty, order
brand loyalty, etc. size, applications, etc.
Decision Making Individual or group Formalization of purchasing
(family) choice, low or high procedures, size & characteristics
involvement purchase, of decision making group, use of
attitudes and knowledge outside consultants, purchasing
about product class, price criteria, (de) centralizing buying,
sensitivity, etc. price sensitivity, switching costs, etc.
Media Patterns Level of use, types of Level of use, types of media used,
media used, times of use, time of use, patronage at trade shows,
etc. receptivity of sales people, etc.
Segmentation in Action
We segment our customers by letter volume, by
postage volume, by the type of equipment they use.
Then we segment on whether they buy or lease
Based on this knowledge, we target our marketing
messages, fine tune our sales tactics, learn which
benefits appeal to which customers and zero in on key
decision makers at a company.
Kathleen Synnot, VP, Worldwide Marketing
Mailing Systems Division, Pitney Bowes, Inc.
[quoted in Marketing Masters (Walden and Lawler)]
If youre not thinking segments, youre not
thinking. To think segments means you
have to think about what drives customers,
customer groups, and the choices that are
or might be available to them.
Levitt, Marketing Imagination
STP as Business Strategy
Identify segmentation bases and segment the market.
Develop profiles of resulting segments.
Evaluate attractiveness of each segment.
Select target segments.
Identify possible positioning concepts for each target segment.
Select, develop, and communicate the chosen concept.
to create and claim value
Overview of Methods for STP
Clustering and discriminant
Choice-based segmentation

Perceptual mapping
- later
Segmentation (for Carpet Fibers)
Location of
segment centers.
Typical members:
A: schools
B: light
D: health
Distance between
segments C and D
Water Resistance
.. ..
Perceptions/Ratings for one respondent:
Customer Values
Water Resistance
Segment(s) to serve
Product Positioning
Comp 1
Comp 2
Water Resistance
A Note on Positioning
Positioning involves designing an offering so that the
target segment members perceive it in a distinct and
valued way relative to competitors.
Three ways to position an offering:
1. Unique (Only product/service with XXX)
2. Difference (More than twice the [feature]
3. Similarities (Same functionality as
lower price)
What are you telling your targeted segments?
Behavior-Based Segmentation
Traditional segmentation
(eg, demographic,
Needs-based segmentation
Behavior-based segmentation
(choice models)
Steps in a Segmentation Study
Articulate a strategic rationale for segmentation (ie, why are
we segmenting this market?).
Select a set of needs-based segmentation variables most
useful for achieving the strategic goals.
Select a cluster analysis procedure for aggregating (or
disaggregating customers) into segments.
Group customers into a defined number of different
Choose the segments that will best serve the firms strategy,
given its capabilities and the likely reactions of competitors.
Segmentation: Methods Overview
Factor analysis (to reduce data
before cluster analysis).
Cluster analysis to form segments.
Discriminant analysis to describe
Cluster Analysis for
Segmenting Markets
Define a measure to assess the similarity of customers on
the basis of their needs.
Group customers with similar needs. Recommend: the
Wards minimum variance criterion and, as an option,
the K-Means algorithm for doing this.
Select the number of segments using numeric and
strategic criteria, and your judgment.
Profile the needs of the selected segments (e.g., using
cluster means).
Cluster Analysis Issues
Defining a measure of similarity (or distance) between
Identifying outliers.
Selecting a clustering procedure
Hierarchical clustering (e.g., Single linkage, average linkage, and minimum
variance methods)
Partitioning methods (e.g., K-Means)
Cluster profiling
Univariate analysis
Multiple discriminant analysis
Doing Cluster Analysis
Dimension 2
Dimension 1

Perceptions or ratings data
from one respondent
a =
distance from
member to cluster center
b =
distance from I to
Wards Minimum Variance Agglomerative
Clustering Procedure
First Stage: A = 2 B = 5 C = 9 D =
10 E = 15
Second Stage: AB = 4.5 BD = 12.5
AC = 24.5 BE = 50.0
AD = 32.0 CD = 0.5
AE = 84.5 CE = 18.0
BC = 8.0 DE = 12.5
Third Stage: CDA = 38.0 CDB = 14.0 CDE = 20.66
AB = 5.0
AE = 85.0 BE = 50.5
Fourth Stage: ABCD = 41.0 ABE= 93.17
CDE = 25.18
Fifth Stage: ABCDE = 98.8

Wards Minimum Variance Agglomerative
Clustering Procedure
Discriminant Analysis for
Describing Market Segments
Identify a set of observable variables
that helps you to understand how to
reach and serve the needs of selected
Use discriminant analysis to identify
underlying dimensions (axes) that
maximally differentiate between the
selected clusters.
Two-Group Discriminant Analysis
Need for Data Storage
x = high propensity to buy
o = low propensity to buy
Interpreting Discriminant Analysis
What proportion of the total variance in the descriptor
data is explained by the statistically significant
discriminant axes?
Does the model have good predictability (hit rate)
in each cluster?
Can you identify good descriptors to find differences
between clusters? (Examine correlations between
discriminant axes and each descriptor variable).
PDA Example
PDA Segmentation
Performs Wards method - Code:
proc cluster data=hold.pda method=wards standard
outtree=treedat pseudo;
var Innovator Use_Message Use_Cell Use_PIM
Inf_Passive Inf_Active Remote_Acc Share_Inf Monitor Email
Web M_Media Ergonomic Monthly Price;

proc tree data=treedat;
PDA Segmentation (alternative)
Performs K-means method - Code:

proc fastclus data=hold.pda maxc=4 maxiter=10 random=41 maxiter=50 out=clus;
var Innovator Use_Message Use_Cell Use_PIM Inf_Passive
Inf_ActiveRemote_Acc Share_Inf Monitor Email Web M_Media Ergonomic ;

proc means data =clus;
var Innovator Use_Message Use_Cell Use_PIM Inf_Passive
Inf_Active Remote_Acc Share_Inf Monitor Email Web M_Media

Ergonomic Monthly Price;
by cluster;
The following clusters are quite close
together and can be combined with a small
loss in consumer
grouping information:
i) clusters 7 and 5 at 0.27,
ii) clusters 1 and 6 at 0.28, ii)
fused cluster 7-5 and cluster 2 (0.34).

However, when going from a four-cluster
solution to a three-cluster solution, the
distance to be bridged is much larger
thus, the four-cluster solution is indicated by
the ESS.
In addition, four seems a reasonable number
of segments to handle based on managerial
Four Cluster Solution profile code;
proc tree data = treedata nclusters=4 out=outclus no print;

** create new data set;

data temp;
merge hold.pda outclus;
** profile these segments;

proc means data =temp;
var Innovator Use_Message Use_Cell Use_PIM Inf_Passive Inf_Active Remote_Acc
Share_Inf Monitor Email Web M_MedErgonomic Monthly Price;
by cluster;

PDA profiles
PDA Visual profile
PDA Visual profile
PDA profiles
Cluster 1. Phone users who use Personal Information
Management software, to whom Email and Web
access, as well as Multimedia capabilities are

Cluster 2. People who use messaging services and cell
phones, need remote access to information, appreciate
better monitors, but not for multi-media usage.
PDA profiles..
Cluster 3. Pager users who have a high need for fast
information sharing (receiving as well as sending) and
also remote access. They use neither email
extensively, nor the Web, nor Multi-media, but do
require a handy, non-bulky device.

Cluster 4. Innovators who use cell phones a lot, have
a high need for Email, Web, and Multi-media use.
They also require a sleek device.
Profile based on Demos/behaviour
Name the segments
Cluster 1 - Sales Pros:
Cluster 1 consists mainly of sales professionals: 54% of the cluster members
indicated Sales as their occupation. They use the cell phone heavily, and many
(45%) own a PDA already; practically all have access to a PC. Their work often
takes them away from the office. They mostly read two of the selected
magazines: 30% read BW. From the needs data, we see that they are quite price

Cluster 2 Service Pros:
Cluster 2 is made up primarily of service personnel (39%) and secondarily of
sales personnel (23%). They use cell phones heavily, but only about one fifth
currently use a PDA. They spend much time on the road and in remote locations.
They read PC Magazine, 29%. From the needs data, we see that they are quite
price sensitive.
Name the segments
Cluster 3 Hard Hats:
Cluster 3 is made up predominantly of construction (31%) and
emergency (19%)
workers. They use cell phones, but usually do not own a PDA.
By the nature of their work, they have high information relay needs and
generally work in remote locations.

They exchange information with colleagues in the field (e.g. construction
workers on the site). Many read Field & Stream (31%) and also PC Magazine.
Note also from the needs data, that they are the least price sensitive (willing to pay
highest price plus monthly fee) and also have the lowest income.

This apparent anomaly occurs because these folks are less likely to have
to pay for the device themselves, raising the question of whose preferencestheir
own or their employerswill drive the adoption decision

Name the segments
Cluster 4 Innovators:
Cluster 4 represents early adopters (see needs data),
predominantly professionals (lawyers, consultants,
Every cluster member has access to a PC, 89
percent already own PDAs.
They read many magazines, especially BW
49%, PCMag 32%. Most are highly paid and highly
Who to target
Interpreting Cluster Analysis Results
Select the appropriate number of clusters:
Are the bases variables highly correlated? (Should we reduce the data through
factor analysis before clustering?)
Are the clusters separated well from each other?
Should we combine or separate the clusters?
Can you come up with descriptive names for each cluster (eg, professionals,
techno-savvy, etc.)?
Segment the market independently of your ability to reach the
segments (i.e., separately evaluate segmentation and
discriminant analysis results).
Discrimination based on
proc discrim data=temp outstat=outdisc method=normal pool=yes list
class cluster; priors prop;
vars age education etc ;

** all relevant vars. not used to create segment solutions;
Discrimination based on
This allows us a way
to target and profile future
Discrimination based on
Discrimination based on
The first discriminant function above explains 51% the variation. According to its
coefficients, i.e., the four groups are particularly different with respect to the amount away
from the office.

In addition, the function shares high correlation with the level of education, possession of a
PDA, and income.

The second function explains 32% of the variance and primarily distinguishes the occupation
types construction/emergency from sales/service, and the third function separates Sales and
Service types.
Visualising relationships

Correspondence Analysis
Provides a graphical summary of the interactions in a table
Also known as a perceptual map
But so are many other charts
Can be very useful
E.g. to provide overview of cluster results
However the correct interpretation is less than intuitive, and
this leads many researchers astray
Correspondence analysis plots should be interpreted by
looking at points relative to the origin
Points that are in similar directions are positively associated
Points that are on opposite sides of the origin are negatively
Points that are far from the origin exhibit the strongest associations
Also the results reflect relative associations, not just which
rows are highest or lowest overall
Software for
Correspondence Analysis
Earlier chart was created using a specialised package called BRANDMAP
Can also do correspondence analysis in most major statistical packages
For example, using PROC CORRESP in SAS:

*---Perform Simple Correspondence AnalysisExample 1 in SAS OnlineDoc;
proc corresp all data=Cars outc=Coor;
tables Marital, Origin;

*---Plot the Simple Correspondence Analysis Results---;
%plotit(data=Coor, datatype=corresp)
Cars by Marital Status
Other details
Tandem Segmentation
One general method is to conduct a factor analysis,
followed by a cluster analysis
This approach has been criticised for losing
information and not yielding as much
discrimination as cluster analysis alone
However it can make it easier to design the distance
function, and to interpret the results
Tandem k-means Example
proc factor data=datafile n=6 rotate=varimax round reorder flag=.54 scree out=scores;
var reasons1-reasons15 usage1-usage10;

proc fastclus data=scores maxc=4 seed=109162319 maxiter=50;
var factor1-factor6;

Have used the default unweighted Euclidean distance
function, which is not sensible in every context
Also note that k-means results depend on the initial cluster
centroids (determined here by the seed)
Typically k-means is very prone to local maxima
Run at least 20 times to ensure reasonable maximum
Cluster Analysis Options
There are several choices of how to form clusters in
hierarchical cluster analysis
Single linkage
Average linkage
Density linkage
Wards method
Many others
Wards method (like k-means) tends to form equal sized,
roundish clusters
Average linkage generally forms roundish clusters with
equal variance
Density linkage can identify clusters of different shapes
Density Linkage
Cluster Analysis Issues
Distance definition
Weighted Euclidean distance often works well, if weights are chosen
Cluster shape
Shape of clusters found is determined by method, so choose method
Hierarchical methods usually take more computation time than k-means
However multiple runs are more important for k-means, since it can be
badly affected by local minima
Adjusting for response styles can also be worthwhile
Some people give more positive responses overall than others
Clusters may simply reflect these response styles unless this is adjusted for,
e.g. by standardising responses across attributes for each respondent
PROC FASTCLUS in SAS tries to minimise the root mean
square difference between the data points and their
corresponding cluster means
Iterates until convergence is reached on this criterion
However it often reaches a local minimum
Can be useful to run many times with different seeds and choose the
best set of clusters based on this RMS criterion
See for
more k-means issues
Iteration History from FASTCLUS
Relative Change in Cluster Seeds
Iteration Criterion 1 2 3 4 5

1 0.9645 1.0436 0.7366 0.6440 0.6343 0.5666
2 0.8596 0.3549 0.1727 0.1227 0.1246 0.0731
3 0.8499 0.2091 0.1047 0.1047 0.0656 0.0584
4 0.8454 0.1534 0.0701 0.0785 0.0276 0.0439
5 0.8430 0.1153 0.0640 0.0727 0.0331 0.0276
6 0.8414 0.0878 0.0613 0.0488 0.0253 0.0327
7 0.8402 0.0840 0.0547 0.0522 0.0249 0.0340
8 0.8392 0.0657 0.0396 0.0440 0.0188 0.0286
9 0.8386 0.0429 0.0267 0.0324 0.0149 0.0223
10 0.8383 0.0197 0.0139 0.0170 0.0119 0.0173

Convergence criterion is satisfied.

Criterion Based on Final Seeds = 0.83824
Results from Different Initial Seeds
19th run of 5 segments

Cluster Means


1 -0.17151 0.86945 -0.06349 0.08168 0.14407 1.17640
2 -0.96441 -0.62497 -0.02967 0.67086 -0.44314 0.05906
3 -0.41435 0.09450 0.15077 -1.34799 -0.23659 -0.35995
4 0.39794 -0.00661 0.56672 0.37168 0.39152 -0.40369
5 0.90424 -0.28657 -1.21874 0.01393 -0.17278 -0.00972

20th run of 5 segments

Cluster Means


1 0.08281 -0.76563 0.48252 -0.51242 -0.55281 0.64635
2 0.39409 0.00337 0.54491 0.38299 0.64039 -0.26904
3 -0.12413 0.30691 -0.36373 -0.85776 -0.31476 -0.94927
4 0.63249 0.42335 -1.27301 0.18563 0.15973 0.77637
5 -1.20912 0.21018 -0.07423 0.75704 -0.26377 0.13729
Howard-Harris Approach
Provides automatic approach to choosing seeds for k-means
Chooses initial seeds by fixed procedure
Takes variable with highest variance, splits the data at the mean, and
calculates centroids of the resulting two groups
Applies k-means with these centroids as initial seeds
This yields a 2 cluster solution
Choose the cluster with the higher within-cluster variance
Choose the variable with the highest variance within that cluster, split
the cluster as above, and repeat to give a 3 cluster solution
Repeat until have reached a set number of clusters
I believe this approach is used by the ESPRI software
package (after variables are standardised by their range)
Another Clustering Method
One alternative approach to identifying clusters is to fit a
finite mixture model
Assume the overall distribution is a mixture of several normal
Typically this model is fit using some variant of the EM algorithm
E.g. weka.clusterers.EM method in WEKA data mining package
See WEKA tutorial for an example using Fishers iris data
Advantages of this method include:
Probability model allows for statistical tests
Handles missing data within model fitting process
Can extend this approach to define clusters based on model
parameters, e.g. regression coefficients
Also known as latent class modeling
Segmentations via Choice Modelling

Choice Models
1. Observe choice:
(Buy/not buy => direct marketers
Brand bought => packaged goods, ABB)
2. Capture related data:
market conditions (price, promotion, etc.)
3. Link
1 to 2 via choice model => model reveals
importance weights of characteristics

Choice Models vs Surveys
With standard survey methods . . .
preference/ importance
choice weights perceptions

predict observe/ask observe/ask
But with choice models . . .
choice weights perceptions

observe infer observe/ask
Behavior-Based Segmentation Model
Stage 1: Screen products using key attributes to identify the
consideration set of suppliers for each type of customer.
Stage 2: Assume that customers (of each type) will choose
suppliers to maximize their utility via a random utility model.
= V

= Utility that customer i has for supplier js product.
= Deterministic component of utility that is a function of
product and supplier attributes.

= An error term that reflects the non-deterministic component
of utility.
Specification of the Deterministic
Component of Utility
= W

where: i = an index to represent customers, j is an index to
represent suppliers, and k is an index to represent attributes.
= is perception of attribute k for supplier j.
= estimated coefficient to represent the impact of
on the utility realized for attribute k of supplier j for customer i.
A Key Result from this Specification:
The Multinomial Logit (MNL) Model
If customers past choices are assumed to reflect the principle
of utility maximization and the error (
) has a specific form
called double exponential, then:


= probability that customer i chooses supplier j.
= estimated value of utility (ie, based on
estimates of b
) obtained from maximum likelihood estimation.
Key idea: Segment on the basis of
probability of choice
1. Loyal to us
2. Loyal to competitor
3. Switchables: loseable/winnable
Applying the MNL Model in
Segmentation Studies
Switchability Segmentation
Current Product-Market by Switchability
Questions: Where should your marketing efforts be
How can you segment the market this way?
Loyal to Us Losable
(business to gain)
Loyal to
Using Choice-Based Segmentation for
Database Marketing
Purchase Purchase
Customer Probability Volume Margin = A B C
1 30% $31.00
0.70 $6.51
2 2% $143.00
0.60 $1.72
3 10% $54.00
0.67 $3.62
4 5% $88.00
0.62 $2.73
5 60% $20.00
0.58 $6.96
6 22% $60.00
0.47 $6.20
7 11% $77.00
0.38 $3.22
8 13% $39.00
0.66 $3.35
9 1% $184.00
0.56 $1.03
10 4% $72.00
0.65 $1.87
Managerial Uses of Segmentation
Select attractive segments for focused effort (Can use
models such as Analytic Hierarchy Process or GE
Planning Matrix).
Develop a marketing plan (4Ps and positioning) to target
selected segments.
In consumer markets, we typically rely on advertising and channel
members to selectively reach targeted segments.
In business markets, we use sales force and direct marketing. You can
use the results from the discriminant analysis to assign new customers
to one of the segments.
Checklist for Segmentation Studies
Is it values, needs, or choice-based? Whose values and needs?
Is it a projectable sample?
Is the study valid? (Does it use multiple methods and multiple
Are the segments stable?
Does the study answer important marketing questions (product design,
positioning, channel selection, sales force strategy, sales forecasting)
Are segmentation results linked to databases?
Is this a one-time study or is it a part of a long-term program?
Concluding Remarks
In summary,
Use needs variables to segment markets.
Select segments taking into account both the attractiveness of
segments and the strengths of the firm.
Use descriptor variables to develop a marketing plan to reach and
serve chosen segments.
Develop mechanisms to implement the segmentation strategy on a
routine basis (one way to do this is through information