Car Sales.sav
Download
Report
Transcript Car Sales.sav
MKT 700
Business Intelligence and
Decision Models
Week 7:
Segmentation and Cluster Analysis
What have we seen so far?
DB Infrastructure
Relational databases, data integrity and
data queries
Data Preparation
Data cleaning and transformation
CLV (Customer Expected NPV)
RFM (Classification)
Where are we going from
now?
Customer Segmentation
Customers’ Profile
Decision tree
Customers’ Propensity to buy
Cluster analysis
Logistic regression
Campaign Metrics and Testing
Outline for Today
Clustering:
Clustering and Segmentation
B2C and B2B
Clustering theory
Lab
SPSS help function (“Show Me”)
Demo with Car Sales.sav
Demo with DMData.sav (Lab)
Customers are not all the
same
Different responses to marketing efforts
Different treatments
Product usage, product attributes,
communication, marketing channels
Packages, prices, copy strategy,
communication and sales channels
Basic marketing rules about segmentation
Consumer Segmentation
Family life cycle (stage in life)
Lifestyle (values)
Product usage/loyalty
Preferred communication channel
Buying behaviour
Data Sources for
Segmentation
Internal
Transaction
Surveys & Customer Service
External (Data overlays)
Lists
Census
Taxfiler
Geocoding
Geo-Segmentation in CDA
Birds of a feather f___k together…
Environics (Prizm)
Generation5 (Mosaic)
http://www.generation5.ca
Manifold:
http://www.environicsanalytics.ca/
http://www.manifolddatamining.com/html/lifestyle/lifes
tyle171.htm
Pitney-Bowes (Mapinfo)
http://www.utahbluemedia.com/pbbi/psyte/psyteCanad
a.html
B2B Segmentation
Firm size (employees, sales)
Industry (SIC, NAICS)
Buying process
Value in finished product
Usage (Production/Maintenance)
Order size and Frequency
Expectations
Clustering & Segmentation
Segmentation Marketing process
Clustering Classification process
(Topology or Taxonomy)
Clustering ≠ Segmentation
A cluster is not a segment
But a segment is a cluster
Clustering
Measuring distances (differences or
dissimilarities between subjects)
Measuring proximities (similarity
between subjects)
BI Modeling Techniques
No Target (No dependent variable,
unsupervised learning)
• RFM
• Cluster Analysis (Unsupervised learning)
Target (Dependent variable,
supervised learning)
• Regression Analysis
• Decision Trees
• Neural Net Analysis
Measuring distances
(two dimensions, x and y)
A
B
C
16
Measuring distances
(two dimensions)
dac2 = (dx2 + dy2)
A
B
C
dac2 = (di)2
dac = [(di)2]1/2
17
Measuring distances
(two dimensions)
D(b,a)
A
B
D(a,c)
D(b,c)
C
18
Cluster Analysis Techniques
Hierarchical Clustering
Metric, small datasets
Distances between US cities
ATL
CHI
DEN
HOU
LA
MIA
NY
SF
SEA
DC
0
587
1212
701
1936
604
748
2139
2182
543
Chicago
587
0
920
940
1745
1188
713
1858
1737
597
Denver
1212
920
0
879
831
1726
1631
949
1021
1494
701
940
879
0
1374
968
1420
1645
1891
1220
1936
1745
831
1374
0
2339
2451
347
959
2300
Miami
604
1188
1726
968
2339
0
1092
2594
2734
923
New_York
748
713
1631
1420
2451
1092
0
2571
2408
205
2139
2182
543
1858
1737
597
949
1021
1494
1645
1891
1220
347
959
2300
2594
2734
923
2571
2408
205
0
678
2442
678
0
2329
2442
2329
0
Atlanta
Houston
Los_Angeles
San_Francisco
Seattle
Washington_DC
SPSS Hierarchical Clusters
Dendogram
SPSS Multidimensional Scaling
(Euclidean Distance)
1
1
2
3
4
5
6
7
8
9
10
Atlanta
.9575
Chicago .5090
Denver -.6416
Houston .2151
Los_Ange
Miami
1.5101
New_York
San_Fran
Seattle
-1.7875
Washingt
2
-.1905
.4541
.0337
-.7631
-1.6036 -.5197
-.7752
1.4284
.6914
-1.8925 -.1500
.7723
1.3051
.4469
Euclidean distance mapping
Cluster Analysis Techniques
Hierarchical Clustering
K-mean Clustering
Metric variables, small datasets
Metric, large datasets
Two-Step Clustering
Metric/non-metric, large datasets,
optimal clustering
Cluster Analysis Techniques
See Chapter 23, SPSS Base Statistics for description of methods
Two-Step Cluster Tutorials
SPSS, Direct Marketing, Chapter 3 and 9
Help Case Studies Direct Marketing
Cluster Analysis
File to be used: dmdata.sav
SPSS, Base Statistics, Chapter 24
Analyze Classifiy Two-Step Cluster
File to be used: Car_Sales.sav
Help: “Show me”
Two Video Demos
http://spss.co.in/video.aspx?id=62
Car Sales
http://www.youtube.com/watch?v=
DpucueFsigA
File not available, but similar to
dmdata.sav. Good demo
Two-Step Clustering
Available
Measurement Scale
Direct Marketing (Simplistic)
Analyse Classify (More Advanced)
Continuous Euclidian Distance
Nominal Log-Likelihood
Repeat for stability
Explore Viewer model
Top line from Chapter 10-1
Customer Segmentation
Two kinds of segmentation: learning segmentation divides your
customers by some criteria and sends them all the same message. See
which group responds best and use that info. Dynamic segmentation:
send each group different offers and text to get a better response.
An ideal segment: large, well defined, can be motivated, and measured.
Justifies a person’s time.
Segmentation requires insight, analytics, and anecdotes. Segmentation
action plan involves a road map, budget, goals, and tests.
Segments and status levels (gold, silver) are not the same.
Response rates can be improved by segmentation and RFM.
Nielsen PRIZM segment codes can be profitable in deciding whom to
promote to.
Top line from Chapter 10-2
Customer Segmentation
Demographic data can be appended to any database that has postal
addresses.
Dynamic segmentation with direct mail is a winner.
Segmentation does not work for e-mail promotions if you are mailing
very frequently. There is not enough time to create dynamic content. The
tradeoff: with e-mails the lift from segmentation is not as great as the lift
from frequent e-mails. Result: many mass e-mail marketers do not use
dynamic segmentation.
Good e-mails with lots of links are more complicated to create than good
print copy.
There are LTV charts that show the lift from dynamic segmentation. In
general, e-mail marketing staff members do not have the budget or the
analytic capabilities to get the resources for database marketing.