Data Mining

mySAP CRM
Detailed View
CRM Analytics:
Data Mining
(Release BW 3.2)
Content
 Introduction
 Predict with:
 Decision Trees
 Scoring
 Clustering
 ABC Classification
 Business Scenarios
 Conclusion
 SAP AG 2003, Predict, Ramine Eskandari / 2

CRM Analytics Architecture

Analytical Methods

What is Data Mining?
Data mining helps you analyze and understand customer

behavior:
 Data mining is an analytical approach that looks for hidden data

patterns and relationships in large databases
 Data mining not only provides insights by analyzing past data, but it is
also capable of predicting future trends and behaviors
 Data mining allows organizations to make the critical jump from

retrospective analysis to prospective decision-making

Why is Data Mining Important Today?
Changes in Technology:
 Increased usage of the Internet
 Appearance of data warehouses
 Increase in computing power
 Better modeling approaches

Changes in Customer Behavior:
 Better informed
 More demanding
 Increased willingness to switch to competitors
 Evolution of needs: more complex, harder to satisfy

Changes in Competition:
 Evolution of strategies:
 Mass marketing vs. One-to-One marketing
 Increased competition
 Fast-paced environment
 Emergence of niche players

Data Mining Methods
Supervised Learning (Predictive):
 Decision Trees
 Scoring
(Regression)

Data Mining Methods
Customers
Clustering (Prediction) Decision Trees

Train
Scoring (Regression)
td
ic
re
P
Exploratory Data New Data
Customer Age Income Household Size Probability of Buying Customer Age Income Household Size Probability of Buying
Sandy Jones 30 low 1 Yes Rob Eagle 45 high 2 Yes
Peter Maier 18 high 2 No Eva Smith 30 high 3 No
Sandra Müller 45 medium 2 Yes John White 20 medium 4 Yes
Susanne Wolf 70 low 1 Yes Tom Wolf 22 high 2 Yes
Oliver Paul 34 medium 3 No Oliver Mann 56 low 1 No
... ... ... ... ... ... ... ... ... ...

Data Mining Methods
Unsupervised Learning (Informative):
 Clusterin
g
A B C
 ABC
Classification
 Association
Analysis
 Scoring (Weighted Score

Tables)

Using and Storing Data Mining Results
SAP EP SAP BW SAP CRM

(Enterprise Portal)
ODS Objects Master Data
InfoCubes Hierarchies Business
Partner
Target
Group
Online Prediction Data Mining

PM
ML
PMML
Intelligence Connector*
3rd Party Tool
*Used for Real-Time Analytics

Data Mining - PMML
Predictive Model Markup Language (PMML) is an XML

markup language that provides a standard way to define
and share data mining models.
 Export data mining models definition rather than just their results
 Use data mining models defined with third-party tools
 Share data mining models easily between different applications and

systems
 Enable operational processes to use data mining models obtained

from multiple sources

Data Mining – Process Overview
Step 1. Create Query Step 4. Transfer Results
BW CRM
Step 2. Define Model Step 3. Run Model

Content
 Introduction
 Predict with:
 Decision Trees
 Scoring
 Clustering
 Conclusion

Decision Trees
Benefits
 Identify the different factors driving the behavior of your

customers
 Break down your customer base into segments that share the
same behavior
New Developments
 Share data mining models between different systems using

PMML
N
E
W

Decision Trees - Definition
A decision tree is a classification approach that uses the

values of input variables to predict the value of a
categorical variable.
 Decision trees allow the user to:

 Generate readily-understandable rules with which to predict customer
behavior
 Evaluate the different values of possible outcomes and the probability to
reach them
 Produce a graphical representation of how different factors affect
an outcome
 Create a segmentation scheme based on the decision tree results

Decision Trees – Process Overview
Customer Income Age Occupation … Buying

Behavior
Rosa Kanter $ 10 000 48 Accountant … Yes
Peter $ 3000 22 Consultant … No

Drucker Existing
Customers
Tom Peters $ 8000 36 Accountant … Yes (Historical Data)
… … … … … …
Kenichi $ 6500 34 Consultant … ?

Ohmae
New Customers
Michael $ 2000 63 Accountant … ? (New Data)
Porter
… … … … … ?

Historical Historical
New New 92% Accuracy
1. Train 2. Evaluate
Customer … … … … Buying
Behavior
Historical
Kenichi … … … … No
Ohmae
Michael … … … … Yes
Porter New
… … … … … Yes
3. Predict
Predicted

Age Root Node  A tree is “grown” from top

to bottom
>= 35 <35 Test
 A record sets out from the
root node
Buy Income  A test determines which
node the record should go
<=$5000 >$5000 Decision Node to next
 All records end up in a leaf
node
Won‘t Buy Occupation
Accountant Consultant
Leaf Nodes
Buy Won‘t Buy

Age Rule:
If Age is less than 35 and
>= 35 <35
Income is greater than $5000 and
Occupation is consultant, then
Buy Income
 The customer will buy the

<=$5000 >$5000
product
Won‘t Buy Occupation
Accountant Consultant
Won‘t Buy Buy

?
What are the most important factors I can use to predict
my customers’ buying behavior?
Purchase Frequency? Age? Occupation?
Income? Family Size? Level of Education?

Gender?
City? Number of Complaints?
Most Important
1. Age
2. Income
3. Occupation
Least Important 4. Purchase Frequency

Decision Trees – Fraud Detection
How Can I Reduce Customer Fraud? Discovery
Train
Evaluate
Step 1. Train
Decision Trees Predict
Step 2. Test
on Historical Identify
and Evaluate Step 3. Apply Model
Data
Model on New Data Step 4. Identify
Customers with High
Fraud Probability
Application
Transfer Act
Step 5. Transfer Step 6. Adapt

Results to CRM Payment Option to
Type of Customer

Decision Trees - Settings
Choose
Choose
Settings
Settings
View
View Train
Train
Training
Training Model
Model
Results
Results

Decision Trees – Training Phase
Root
Root
Node
Node
Class
Class
Distributio
Distributio
nn
Decisio
Decisio
nn Node Test
Test
Node
Criteria
Criteria
Leaf
Leaf
Node
Node
Rule
Rule
Leading
Leading to
to
Leaf Node
Leaf Node
Preview
Preview
Pane
Pane
Decision Trees – Evaluation Phase
Predicted
Predicted
Class
Class Class
Class in
in
Evaluation
Evaluation
Data
Data
Evaluatio
Evaluatio
nn
Results
Results
Error
Error
Rate
Rate

Decision Trees – Prediction Phase
View
View Results
Results
with
with aa
Hierarchy
Hierarchy Prediction
Prediction
Results
Results
Single
Single
Record
Record
Prediction
Prediction
Creation
Creation of
of
CRM Target
CRM Target
Group
Group

Decision Trees – Prediction Phase
Decision
Decision
Tree
Tree Results
Results
Using
Using
Hierarchy
Hierarchy

Decision Trees – PMML Export
Generate
Generate PMML
PMML
with
with Model
Model
Definition
Definition
Export
Export PMML
PMML to
to
Flat File
Flat File

Decision Trees – Online Prediction
Decision
Decision
Tree
Tree Web
Web
Item
Item

Scoring
Benefits
 Create a single score for your customers by taking into

account different dimensions
 Use regression to predict future customer behavior

N
E
New Developments W
 Additional statistical measures have been added
 Scoring output visualization has been enhanced
 Online prediction feature for single records has been added

Scoring - Definition
Three different techniques:
 Weighted Score
Tables
 Linear
Regression
 Nonlinear
Regression

Scoring - Weighted Score Tables
Customer Income Age Gender City …
Rosa Kanter $ 10 000 48 Female San Diego …
Peter $ 3000 22 Male Los Angeles …

Drucker
Tom Peters $ 8000 36 Male New York …

City
Kenichi $ 6500 34 Male New York …

Ohmae City
Michael $ 2000 63 Male Boston …

Porter
… … … … … …

+ 1. Revenue 6 Dimensions
2. Income
3. City
Weights
4. Gender
5. Age
6. Family Size
-
Weighted Score
Tables
1 Dimension
Customer Value Index

1st Level
Weights
2/10 3/10 3/10 2/10
Overall
City + Revenue +…+ Income =
Score
Values Weights
Values Weights Values* Weights 2nd Level
New York City 50 0 0 $0-2000 20 Weights
San Diego 20 $15 000 100 $2001-5000 40
Boston 80 $5001-8000 70
*Assume Linear
Los Angeles 10 Relationship for Values $8000+ 100
Between 0-15 000
… …

Scoring - Regression
Regression is a predictive technique that reveals and

measures the value of a given variable (where values are
continuous) in terms of other variables
 Linear Regression
y = ax + b
 Nonlinear Regression
y = f(x)

Customer … … … … Revenue
Rosa Kanter … … … … $ 5000 Historical
Peter … … … … $ 10 000
Drucker
… … … … … … New
1. Train
& Evaluate
Customer … … … … Revenue
Historical
Kenichi … … … … $ 7800
Ohmae
Michael … … … … $ 6000
Porter New
… … … … … …
2. Predict
Predicted

Scoring – Customer Value Score
How Valuable are my Customers? Discovery
Select
Weight
Step 1. Select
Dimensions Score
Step 2. Assign
Affecting
Weights to Step 3. Score your Identify
Customer Value
each Customers Base with
Dimension Created Model Step 4. Sort
Customers Based on
their Value
Transfer Act
Application
Step 5. Transfer Step 6. Send Special
Results to CRM Offers to Most
Valuable Customers

Scoring - Results
Online
Online
Prediction
Prediction
Score
Score
Distribution
Distribution

Scoring – Weighted Score Tables
Weighting
Weighting
Scheme
Scheme

Quality
Quality
Measures
Measures

Clustering
Benefits
 Segment your customer base into homogeneous clusters

based on the dimensions of your choice
N
New Developments E
W
 Share data mining models between different systems using
PMML
 Evaluate your results and interpret them more easily by using

quality measures
 Use additional functions to pre-process your data

Clustering - Definition
Clustering is a data mining technique that creates

groups of records that are:
 Similar to each other within a particular group
 Very different across different groups
 The degree of association between members is measured by all the

dimensions specified in the analysis
 Clustering helps the user explore vast amounts of data and organize
it in a systematic way

Age
High
Income
Low High

+ Family Size
+ +
Clustering
Income Revenue
+ Age
Revenue 30-40s,
… City $40 000
West Coast,
…
Select for Marketing

Campaign
50s,
High Income
East Coast,
2-3 Children
…
Age
…
Income

Clustering – Marketing Campaign
How Can I Improve the Efficiency Discovery

of my Marketing Campaigns?
Select
Execute
Step 1. Select
Dimensions for Evaluate
Step 2. Run
Clustering Identify
Model on Step 3. Analyze and
Selected Data Evaluate Results Step 4. Identify
Valuable Cluster(s)
Transfer Act Application

Step 5. Transfer Step 6. Tailor
Results to CRM Campaigns to each
Segment

Clustering - Settings
Select
Select Type
Type of
of
Transformation
Transformation
View
View Data
Data
Distribution
Distribution
Choose
Choose
Weights
Weights For
For
Input Fields
Input Fields

Clustering - Results
Quality
Quality Name
Name Single
Single
Measures
Measures Clusters
Clusters Record
Record
Prediction
Prediction
Create
Create CRM
CRM
Select
Select Target
Target
Output PMML
PMML Group
Output Group
Type Export
Export
Type
Cluster
Cluster
Size
Size
Influence
Influence

Clustering – Value Distribution
Value
Value
Distributio
Distributio
nn Screen
Screen
Distribution
Distribution
Within
Within
Clusters
Clusters
Clustering – Attribute Charts
Attribute
Attribute
Charts
Charts
Screen
Screen
Cumulated
Cumulated
Percentages
Percentages
Total
Total
Percentages
Percentages

Clustering – Quality Measures
Attribute
Attribute
Value
Value Index
Index
Quality
Quality
Measures
Measures

Clustering – Attribute Value Index
Number Of
Records
Family Size
+
Value Index
+ +
+ 1 -2
2 +2
50 + + 3 0
4 0
+
1 2 3 4
Family Size
Distribution:
Mainly Married Couples
Overall Are In Cluster 1
Cluster 1

Clustering – Online Prediction
Clustering
Clustering
Web
Web Item
Item

Association Analysis
Benefits
 Identify cross-selling opportunities
 Optimize the bundling of your products and maximize revenue
New Developments
 Compound characteristics can be used in defining model
N
E
W

Association Analysis - Definition
Association Analysis uncovers the hidden patterns,

correlations or casual structures among a set of items or
objects.
 It is typically used for Market Basket Analysis (MBA).
 It allows the user to:

 Understand and quantify the relationship between different items (e.g.
products, click stream, etc...)
 Group different items by affinity
 Create readily-understandable rules describing customer buying
behavior
 Organize web pages in order to optimize user accessibility

Bundles
Association
Analysis
Customer
Transactions

When a customer buys a desktop computer,

Rule he will also be likely to buy a printer 80% of the time.
This behavior can be found in 30% of all purchases
Desktop computer Leading Item

Printer Depending Item
80% Confidence
30% Support


he will also be likely to buy a printer 80% of the time
A customer buys a printer 40% of the time
Lift: 80/40=2

he will be 2 times more likely to buy a printer
Importance of Lift
Probability: ,4 X2 Probability: ,8
Importance of Support
Confidence: 100% Confidence: 80%

Support: 10% Support: 30%
Association Analysis – Cross-Selling Opportunities
How Can I Sell More to my Customers? Discovery
Select
Execute
Step 1. Select
Transactional Evaluate
Step 2. Run
Data from
Association Step 3. Analyze and Select
Customers
Analysis Model Evaluate Results
Step 4. Select Rules
with High Confidence
and Support

Step 5. Transfer Step 6. Cross-Sell
Results to CRM Products to
Customers

Association Analysis - Settings
Settings
Settings
Define
Define aa Training
Training
Source
Source for the
for the
Data
Data

Association Analysis - Output Settings
Define
Define Filters
Filters
for
for the
the Data
Data
Output
Output

Association Analysis - Results
Export
Export Rules
Rules
to
to CRM
CRM
Depending
Depending
Leading
Leading Item
Item Item
Item

ABC Classification
Benefits
 Classify your customers into different segments based on their

value to your business
NE
New Developments W
 ABC Classification output visualization has been enhanced
 Classification can be done with grouping variables
 Compound characteristics can be used in defining models

ABC Classification - Definition
ABC Classification is used to classify objects (such as

customers, employees, or products) based on a
particular measure (such as revenue or profit)
 Two different approaches for classification:

 Define intervals for classification criteria
 Define intervals for classified object
 Absolute values or cumulated percentages can be used to create

these intervals

ABC Classification – Gain Chart
% of Revenue
100
 “A” Class:
20% of customers
80
generate 40% of revenue
 “B” Class:
60 Next 30% of customers
40
 “C” Class:
Last 50% of customers
20 A B C
% of Customers
20 40 50 60 80 100

ABC Classification – Example
Region
California California
Customer Revenue
… … A
… …
… …
B
… … C
Massachusetts Massachusetts
Customer Revenue
… …
… … A B C
… …
… …
Nevada Nevada
Customer Revenue
… …
… …
… …
… …
… Single Model …
 SAP AG 2003 , Predict, Ramine Eskandari / 68
ABC Classification – Gold Customers
How Much Profit are my Most Valuable Discovery
Customers Generating for my Business?
Select
Step 1. Select Define

Transactional Execute
Step 2. Define
Data from
Intervals for A, Step 3. Run ABC
Customers Evaluate
B, and C Classification Model
Classes
Step 4. Analyze and
Evaluate Results

Step 5. Transfer Step 6. Improve
Results to CRM Customer Service for
“A” Customers

ABC Classification – Settings
Define
Define Model
Model
Parameters
Parameters
Choose
Choose Field
Field
Type
Type

ABC Classification – Settings
Choose
Choose
Classification
Classification
Method
Method
Define
Define
Intervals
Intervals

ABC Classification – Results
Choose
Choose Value
Value
for
for Grouping
Grouping
Attribute
Attribute

ABC Classification – Results
Gain
Gain Chart
Chart

ABC Classification – Web Item
ABC
ABC Web
Web
Item
Item

Content
 Introduction
 Predict with:
 Decision Trees
 Scoring
 Clustering
 Conclusion

1) Customer Retention
Decision Trees ABC Classification Clustering
A B C
At Risk B
A
Loyal
C
Goal: Reduce Customer Churn
Association
Analysis
Transfer Cross-Sell
Knowledge Rules
Marketing Campaign

2) Customer Conversion
Association
Weighted Score Tables ABC Classification Analysis
A B C
…+…+…=… B
Cross-Sell
…+…+…=… A Rules
…+…+…=…
C
Goal: Convert B to A Customers
Transfer
Knowledge
Marketing Campaign

3) Customer Acquisition
Clustering ABC Classification Decision Trees
A B C
Tr
ai
n
A
B
C
Goal: Acquire New “A” Customers
Pr
ed Decision Trees
i
ct Tr
Transfer ai
ne
Knowledge Prospects d
“A”
Marketing Campaign

Content
 Introduction
 Predict with:
 Decision Trees
 Scoring
 Clustering
 Conclusion

Future Plans
Future Developments
 Data Mining methods integrated in Analysis Process Designer
 Ever closer integration of analytical tools with CRM operational

processes
 Additional transformations in the Analysis Process Designer
 Enhanced visualization for data mining methods
 And much more…

Further Information
 CRM Training Course

CR900 and TACRM6 (Partner)
 mySAP CRM Analytics

SAPNET Alias: CRM-Analytics
 Direct contact:
CRM Product Management:
Ramine.Eskandari@sap.com

Copyright 2003 SAP AG. All Rights Reserved
 No part of this publication may be reproduced or transmitted in any form or for any purpose without the express
permission of SAP AG. The information contained herein may be changed without prior notice.
 Some software products marketed by SAP AG and its distributors contain proprietary software components of
other software vendors.
 Microsoft®, WINDOWS®, NT®, EXCEL®, Word®, PowerPoint® and SQL Server® are registered trademarks of
Microsoft Corporation.
 IBM®, DB2®, DB2 Universal Database, OS/2®, Parallel Sysplex®, MVS/ESA, AIX®, S/390®, AS/400®, OS/390®,
OS/400®, iSeries, pSeries, xSeries, zSeries, z/OS, AFP, Intelligent Miner, WebSphere®, Netfinity®, Tivoli®,
Informix and Informix® Dynamic ServerTM are trademarks of IBM Corporation in USA and/or other countries.
 ORACLE® is a registered trademark of ORACLE Corporation.
 UNIX®, X/Open®, OSF/1®, and Motif® are registered trademarks of the Open Group.
 Citrix®, the Citrix logo, ICA®, Program Neighborhood®, MetaFrame®, WinFrame®, VideoFrame®, MultiWin® and
other Citrix product names referenced herein are trademarks of Citrix Systems, Inc.
 HTML, DHTML, XML, XHTML are trademarks or registered trademarks of W3C®, World Wide Web Consortium,
Massachusetts Institute of Technology.
 JAVA® is a registered trademark of Sun Microsystems, Inc.
 JAVASCRIPT® is a registered trademark of Sun Microsystems, Inc., used under license for technology invented
and implemented by Netscape.
 MarketSet and Enterprise Buyer are jointly owned trademarks of SAP AG and Commerce One.
 SAP, SAP Logo, R/2, R/3, mySAP, mySAP.com and other SAP products and services mentioned herein as well as
their respective logos are trademarks or registered trademarks of SAP AG in Germany and in several other
countries all over the world. All other product and service names mentioned are trademarks of their respective
companies.

Data Mining

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Data Mining

Uploaded by

Copyright:

Available Formats

mySAP CRM

 SAP AG 2003, Predict, Ramine Eskandari / 2

 SAP AG 2003, Predict, Ramine Eskandari / 3

 SAP AG 2003, Predict, Ramine Eskandari / 4

Data mining helps you analyze and understand customer

 Data mining is an analytical approach that looks for hidden data

 Data mining allows organizations to make the critical jump from

 SAP AG 2003, Predict, Ramine Eskandari / 5

 Increased usage of the Internet

 Appearance of data warehouses

 Increase in computing power

 Better modeling approaches

 SAP AG 2003, Predict, Ramine Eskandari / 6

Changes in Customer Behavior:

 Increased willingness to switch to competitors

 Evolution of needs: more complex, harder to satisfy

 SAP AG 2003, Predict, Ramine Eskandari / 7

 Emergence of niche players

 SAP AG 2003, Predict, Ramine Eskandari / 8

Supervised Learning (Predictive):

 SAP AG 2003, Predict, Ramine Eskandari / 9

Clustering (Prediction) Decision Trees

 SAP AG 2003, Predict, Ramine Eskandari / 10

Unsupervised Learning (Informative):

 Scoring (Weighted Score

 SAP AG 2003, Predict, Ramine Eskandari / 11

SAP EP SAP BW SAP CRM

Online Prediction Data Mining

3rd Party Tool

*Used for Real-Time Analytics

Predictive Model Markup Language (PMML) is an XML

 Use data mining models defined with third-party tools

 Share data mining models easily between different applications and

 Enable operational processes to use data mining models obtained

 SAP AG 2003, Predict, Ramine Eskandari / 13

Step 1. Create Query Step 4. Transfer Results

Step 2. Define Model Step 3. Run Model

 SAP AG 2003, Predict, Ramine Eskandari / 14

 SAP AG 2003, Predict, Ramine Eskandari / 15

 Identify the different factors driving the behavior of your

 Share data mining models between different systems using

 SAP AG 2003, Predict, Ramine Eskandari / 16

A decision tree is a classification approach that uses the

 Decision trees allow the user to:

 SAP AG 2003, Predict, Ramine Eskandari / 17

Customer Income Age Occupation … Buying

Peter $ 3000 22 Consultant … No

Kenichi $ 6500 34 Consultant … ?

 SAP AG 2003, Predict, Ramine Eskandari / 18

New New 92% Accuracy

 SAP AG 2003, Predict, Ramine Eskandari / 19

Age Root Node  A tree is “grown” from top

Buy Won‘t Buy

 SAP AG 2003, Predict, Ramine Eskandari / 20

 The customer will buy the

Won‘t Buy Occupation

Won‘t Buy Buy

 SAP AG 2003, Predict, Ramine Eskandari / 21

Purchase Frequency? Age? Occupation?

Income? Family Size? Level of Education?

City? Number of Complaints?

 SAP AG 2003, Predict, Ramine Eskandari / 22

How Can I Reduce Customer Fraud? Discovery