3. 3
Introduction
3
Who We Are: Facts and Figures
About 150 years old with Market Cap of USD$50 bn
1700 branches, 5600 ATMs in more than 70 countries.
Presence predominantly in Emerging Markets
About 75,000 worldwide staff
Technology Development in Singapore, Malaysia, U.K. and India
Singapore has about 6000 staff
Leaders in Foreign Exchange Currencies (FX) in Emerging Market
Consumer Banking Wholesale Banking
Financial Markets
Corporate Banking
4. Business Decision Making
Timely Information (Operational Decisions, running daily business)
Ad-Hoc Analysis (Explorative, what went well and what did not go well)
Predictive Analytic Tools (What if scenarios)
Interactive Visualization (Trends and Patterns)
Areas of Information and Intelligence
Online Transaction Systems
Data Warehouse
Data Marts
Operational Data Stores
End User Computing
4
What lies beneath the Business Analytics?
Data is the Company’s most valuable asset and keeping it in right data structure is
equally important to derive the information and intelligence.
5. Data Structures/Models influence
Business Rules Implementation (Cardinality, Optionality and Business Representation)
Information Retrieved and Interpreted (Metadata, Hierarchies and Metrics)
Data Integrity and Quality
Data Stewardship & Governance
Operational Efficiencies in Data Management
5
Why Data Modeling is Important ?
Data Models explain the Business Rules and Selecting Right Data Modeling
Strategy is key to the success of Data Management
6. Multiple persistent data layers create data obsolescence with silos and redundant processes.
There are approximately three times more offline systems than on-line systems
6
Challenges in Information Management Landscape
Account
(Africa)
Local Reg
Reporting
Cost Reporting
MS Access
MM APPLE
SQL Svr
Trades
Mango
SQL Svr
Capital
AP
eProc
T&E
AM
GL
RMI
PMI
ECMI
BS MI
Risk Frontier
Accounting & Costs
TP Systems
Core and Loan TP
Systems
SIP
Commodities
FX DB
MS Access
DPLSales
BIReportingLayer(XLCube)
Sales & Marketing
Product Control
Sales Cube
Country Finance
Money Market Derivatives
Exotics
Equities
Equity Derivatives
Cash
Fixed Income
Bond Trades
Repos
Bonds
Bloomberg
PnL
IREC
Party Static
DPL Cube
Traders P&L
ODS
Balance
Sheet
Liquidity
Fund
Pricing
Gabriel
(FSA)
FSA
RRT
Tool
Calculation Engine
Performance
Reports
Liquidity MI
Risk Mgt
Reports
EBBS
Core BankingCorp Loans
RDS
DIH
RLS
CENTRY
Auto Loans
Collateral
Auto Loans
Credit Cards
Netting
Cash & Pooling
Collateral
AR/Billing
Pipeline
Collection
System
InterComp
Server
DMS
Hyperion
(Management)
Pipeline
Management
Reporting
Financial
Reporting
WB Cube
CB Cube
Magic Packs
Insight Packs
NF KPI Non Financials
PDW
People
Wise
HR BFS
(Access)
Static
Non PSGL entities
Management Allocations
Country Finance
Country Finance
IDS
Local Reg
Hyperion
(Statutory)
Centrally
managed BU’s
T1-T4
Journals
IFRS Cube
Reg, Tax,
WB, Others
Segmental
Cube
CMM Data
Collection
Templates
DCS
Budgets and
Forecast
GBS Cube
(Budgetting)
Financial Markets
Countries not in CDW
1
2
3
4
5
6
7
8
10
1111
1212
9
1313 1414
1515
1516
Manual Journal
Entries by
Country Finance
Group Finance
(Adjustments)
ALM
PSGL Cube
PAS 2
Management
Adjustments
Payroll
Non PSGL Info
PAS
UK MI
RFRAME
(UK)
MFU
MFU & Mapping
Tables
Mapping Tables
Templates
Adjustments
Downgrade
Provisions
Cheques
Private Banking
Collateral
CollateralAggregator
Local Reg
Reporting
CRM
Sales Pipeline
FDW
FDW
FDW
RSA
RSA
RSA
Other Other Other
CP
Spring
watch
ALM
ALM Reports
RSA
FDW
PAS
No prescriptive data modeling strategies available in OLAP environment
7. 7
Approach to Modeling Strategy Selection
Types of Persistent Layers & Model
Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration - Business
Area Specific v/s Enterprise Model
Vendor Models v/s In-house Models
Selection of Hierarchical Structures
Technology Considerations
Business Coverage
Business Entities
Operational V/s Analytics
Consumption Type (Raw,
Standardised, Derived)
Metadata Requirements
Timeliness
History
Business Requirement Modeling Aspect
Selection of Right Modeling Strategy Depends on Business Drivers and Other Aspects
Hybrid
(3+2NF)
ROLAP/
MOLAP
Modeling Strategy
Star
8. 8
Modeling Levels – Top Down Approach
Conceptual
Model
TreasuryRisk Finance
Customer Product Employee Geography
Organization
Transactions
Channels
Subject
Area Model
Logical Model
Physical Model
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
9. What are the Applicable Modeling Levels?
Subject Area
Level
Conceptual
Logical
Physical
SemanticODS Core DWStagingOLTP
Certain Persistent Layers do not require all levels of modeling.
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
10. What Normalization Techniques are appropriate?
Subject Area
Level
Conceptual
Logical
Physical
SemanticODS Core DWStagingOLTP
3NF/2NF
(Hybrid)
Schema Type
Normalisation Type depends on the selected persistent layer
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
ROLAP/
MOLAP
Flat3NF
3NF/2NF
(Hybrid)
Most Disputed Modeling
Strategies
11. 11
What Level of Abstraction is appropriate?
Business
Area Specific
Highly
Abstracted
Readily Usable with Business Specific Attribute
Easy to understand
Quick to develop and easy to manage
Not suitable for Enterprise Data Level Integration
Higher Abstraction accommodates more types
General Practice is to use Vendor supplied models
Most Industry Specific Models come with 2000+ entities and 10,000+
Attributes. Highly Flexible
Suitable for Enterprise Level Data Warehouse.
Harder to comprehend for business
Requires extensive modeling effort
Longer and Expensive to implement and affects Time to Market.
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
12. 12
Example of Highly Abstract Models
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
Choose carefully whether abstracted models is the right choice to solve your analytic problem
Require several joins to
explain highly abstracted
concepts such as Locator
Example of Attributes in Enterprise Data Model
Customer Domicile Country Code
Incorporation Country Code
13. 13
Example of Business Area Specific Model
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
In 80% Cases, analytic needs require Business Area Specific problems to be solved
Example of Fixed Income Derivative Model
Abstraction Level is at Product Family/Asset Class
Extensions
anchor on main
entity
14. Financial Markets
14
Levels of Integration Requirements
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
Fixed Income
Flow Products Exotics
Bonds Money Market
Equities
Retail Banking
Commodities
Currencies
Corporate
Banking
Compliance
Department Level
Integration
Division Level
Integration
Determine the right level of data integration to get the right level of abstraction
Enterprise Level
Integration
FinanceRiskHR
15. 15
Vendor Models v/s In-house
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
- Use vendor models as reference and carve out applicable entities and customize it
- Consider usage of frameworks
In-House
A good starting reference point
Requires In-depth understanding
of definitions of generic concepts
Discipline in customizing models
Ambiguity in data mapping to
several thousand entities and
attributes
Vendor Models
Requires high degree of
modeling skills .
Solves business specific or
division level problems
Can be customized to be flexible
enough for that level
Easier to create business
specific model views.
16. 16
Hierarchical Structures
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
Recursive relationships do not work all the time. Consider using flattened hierarchy.
17. 17
Technology Considerations
Types of Persistent Layers &
Model Levels
Levels of Normalisation
Levels of Abstraction
Levels of Integration -
Business Area Models v/s
Enterprise Model
Vendor Models v/s In-house
Models
Selection of Hierarchical
Structures
Technology Considerations
Appliance vendors emerging strong in OLAP environment and many banks investing in
building enterprise level warehouse
Certain Database Appliances designed to work well with
abstracted models but can be expensive
Highly normalized physical models difficult to implement. Most
turn into hybrid of 3NF and 2NF
Constraints can be on number of joins and length of composite
surrogate keys in dimensional models
BI tool adaptability to consumption layer
18. Efficient Information Delivery Model
Consistent data model views along the food chain
Reduction in development costs
Qualitative Data (usefulness of data)
Reduction of redundancies in data processing
Speed to market
Elimination of manual processes
18
Value Proposition in Selecting Right Data Modeling Strategy
19. 19
References
IBM Corporation
Teradata Corporation
Kimball University – Dr. Ralph Kimball
Information Management Magazine – September 2002