managing knowledge in business intelligence systems dr. jan mrazek
TRANSCRIPT
Managing Knowledgein
Business Intelligence Systems
Dr. Jan Mrazek
Market Conditions
Customer Opens an Account
Customer Transacts
Relationship Mapping
Profitability Calculation Business Performance AnalysisCustomer Segmentation
Customer Relationship Analysis
Cross/UP Sell
Modeling Behavior
Prospects
Model Scoring
Channels and Organization
Our mission is to optimize the business process(CVM, BPM)
Mech
Information Warehouse
IPS NCCS
BI M
eta
data
Rep
osi
tory
Un
iform
BI Tech
nic
al A
rch
itect
ure
Uniform BI Data Architecture
Divisional LeadersPOS Mortgages MBANX Direct
HRInvestment Products
Retail & Commercial
MIND
Exploratory Data MartCustomer based flat file with more than 1,000
variablesSample of 1.5 mil. customers
CKDB
Query Server
Web Server
CVM
Analytical Database
(taking the role of a customer centric
marketing database)
CustomerSegmentation
Exploratory Data Mart
TreatmentSelection
TreatmentAuthoring
Decisionabout offers
Feedback
Assessment(Analysis)
Model Development
Scoring
CRM DatabaseContact Management
Models(PMML)
Sampleset of
variables
Cust. Serv.Profile
Feedbackdata
OCIF
Customering
Householding
DW + DMs CCAPSRaw Data
Primary sources (operational systems)
CRM Front End System
OCIF & Householding System
DW + Profitability System = CVM Base
CVM Core Analytical System
CVM Exploratory System (Advanced Analytics)
Account Profitability
Customer Aggregations
Household Aggregations
VariablesValue Creation
Acc/C
ust/H
H Key
sRaw account level data in monthly
aggregates
Campaign ManagementTransactional
ODS(Holds only “special” transactions)
Detailed transactions
in a daily batch load
ODS System (“Special” transactions)
Event driven filter of transactions right
during the load
Legend:Data Warehousing/Business Intelligence Environment
Monthly run on all custom
ersD
aily re-run for customers w
ith “special” transactions
OCIF SystemOperational Systems
Offer Selection
CVM Architecture
Key objective
At the Bank of Montreal one of our key objectives is to excel in our service to our customers.
To be able to achieve this key objective, we have to learn how to anticipate our customers’ preferences in a timely manner.
Since only a timely understanding can deliver true service excellence, we are focussed on streamlining knowledge discovery processes along an integrated system architecture so, that the time needed from knowledge discovery to knowledge application is minimized.
Overview of the Knowledge Discovery Process
DataAcquisition
DataPreparation
ModelDevelopment
ModelExecution(Scoring)
ScoresDeployment
ResultsAnalysis
Identificationof
Objectives
data data data data
Data Warehouse
•Data preparation•Model development•?Model execution (Scoring)•? Scores deployment•? Results analysis
Knowledge Discovery Executed in a Non-integrated Environment
DB2 UDB EEE
DM technology A DM technology B DM technology C DM technology D
Disadvantages of the Non-integrated Knowledge Discovery Environment
•Data preparation responsibility of analysts/modelers•Not optimal HW/SW for data preparation•Data about all customers need to be moved to place of model execution•Limited capabilities for model execution in the DW environment•Scores not automatically stored in systems with general availability and access•Limited ability to analyze results, quality of models
•That all results in lost of precious time to apply the discovered knowledge
ExploratoryData Mart
data datadata
data
Data Warehouse
•Model development
IM Scoring
model (PMML)
data
scores
•Data preparation •Model execution (Scoring)
Knowledge Discovery Executed in a Highly Integrated Environment
DB2 UDB EEE
(Large sampleof data)
DM technology A DM technology B DM technology C DM technology D
•Model validation and results analysis
•Mass scores deployment
model (PMML)
model (PMML)
model (PMML)
Advantages of the Integrated Knowledge Discovery Environment
•Data preparation executed by DW transformation professionals•Robust DW HW/SW utilized for data preparation•Modelers concentrate on actual model development•Only samples of data moved to modelers’ environments•Models delivered to IM Scoring in PMML format from different data mining technologies•IM Scoring executes models utilizing all robust DW HW/SW processing power•Scores immediately stored in the DW environment where they can be accessed and used by many applications and users•Full ability to analyze results, quality of models
•That all results in:•Reduction of time needed for knowledge discovery and knowledge deployment•Optimal use of HW/SW and professional resources•Improved process quality
Maintaining Model Version Control - DM Metadata
> Model built when, by whom> What tool, algorithm> Variables (links to Metadata repository)> Variables’ transformation rule - link to ETL Metadata> When last time re-balanced, by whom> Since when in production> Who is the owner, contact> QA of PMML translation, who > Treat as slow moving dimension
2001 Best Practices In Data Warehousing Award (TDWI)
2000 Best Data Warehouse Award (RealWare Awards)
2000 ADT 2000 Software Innovator Award for Data Warehousing(Application Development Trends)
1999 DCI Excellence in Business Information Award
Where you can meet me
•August 15 in Anaheim, California on TDWI World Conference Summer 2001 and Best Practices Summit•IBM Webcast on Enhancing CRM with IBM's DB2 Intelligent Miner Scoring http://webevents.broadcast.com/ibm/datamining/home.asp•Adastra Prague: call +420-2-7173 3303 to arrange for a meeting