Professional Documents
Culture Documents
A primer
Introduction & overview
The paradigm of BI systems
Platforms
Appendix
Review questions
- 1-
Our approach to BI
Cost Flexibility & speed Quality & satisfaction
M anager
cost service and products
2. Activity timeliness
2. Productivity 2. Technology
3. Resource flexiblity
Rep
3. Usage / workload dependability
C u s to m e r
1. Customer access / 1. Expectation conformity
1. Response time
Identification /
acquisition unit
cost
2. Response timeliness
3. Vendor flexiblity
2.
of the service
Service dependability
2. Customer use cost
mapping
3. Customer satisfaction
Modeling
BI systems modelling
1. Technology
SIRE
ES
1. Expectation conformity
response time
1.
costHIGO
Execution unit 2. Technology
timeliness
of work / work
environment
W o rk e r
2. Technology
2. Preparation effort 3. Activity &
dependability
technology
3. Employee satisfaction
flexibility
Implementation Level
Software
Engineering
Interface (SEI)
- 2-
Business Intelligence:
the role within Enterprise Systems
Management support
Management Information Systems
[Planning & Management Control + Business Intelligence ]
Operations support - 3-
Acronyms
- 4-
Characteristics of Analytic & Management Information
Information is
Periodical
Output of computation or
aggregations
Reflects objectives or actual data
- 5-
The 4-layer paradigm of BI /MIS systems
Presentation / reporting
Decision support Mining & other
engine
engines (DSS) application engines
(EIS, reporting)
DATA MART
DATA WAREHOUSE
Loading
Tranformation DATA ENTRY
Extraction
- 6-
The 4-layer paradigm of BI /MIS systems
- 7-
Jones case study
CONTEXT REQUIREMENTS
The Supermarket Chain Management want to
Jones includes 300 shops analyze sales
in 3 regions with 60k items Facts : Sales
on sale
Measures: amount,
A POS (Point Of Sale) system quantity, number of tickets
supports all activities of
each shop : Analysis dimensions
item receiving, Date
storing, Item
scrapping, Shop
selling Time span : 24 months
Specifically, POS terminals rolling
record sales transactions
and issue receipts
- 8-
Jones case study: Level 1
Ticket # 2002a23b11
Store #0021MI
Item Master Data
# Item
Item Des Price Qty Amount
# Store
#190 Pen 3560 2 7.12
Description
#69 Mat 550 10 5.50
Price
#90 Lib 32000 1 32.00
Qunatity mesuere
TOTALE 44.62
Stock on hand
Payment Fidelity P.
Stock at the beginning of the day
Date 120109
Average forecasted dayly sale
Receipt Heading
Receipt detail
# Store
# Ticket
# Ticket
# Item
Amount
Amount
Payment
Qty
Date
- 9-
Level 2: extraction-transformation-loading
Extraction includes
Select source data
Check and clean source data (data
Report/ Mining &
cleaning o data cleansing)
DSS
dashboard other Staging of extracted data (as needed)
Log of extractions
Extraction can be
DATA MART Automatic: a batch procedure that
runs periodically (e.g. daily, weekly,
DATA WAREHOUSE
monthly)
Interactive: integrates and fixes
automatic data
LOADING
DATA ENTRY
TRANSFORMATION
EXTRACTION
ETL can use intermediate databases
Staging Area : where extracted data
are temporarily parked (e.g. Data of
each individual shop)
Operational Data Store (ODS): where
TRANSACTIONS DATABASES
granular data are stored and
reconciled for future use (e.g. receipt
data)
- 10-
Level 3: data warehouse
extraction of a warehouse.
TRANSACTIONS DATABASES Warehouse e Mart typically
adopt different data schemas
- 11-
Level 3: data warehouse
Key table 1
Key 1 Key table
Attribute 1
Fact table Key
Key 1 Attribute 1
Attribute 2
Key 2 Attribute 2
Attribute .
Key Attribute .
Measure 1
Key table 2 Measure 2
Key 2 Measure .
Attribute 1
Attribute 2 The warehouse is typically implemented
Attribute . by relational database, whose schema
reflects the corresponding DFM
(Dimensional Fact Model).
In relational schemas:
Fact tables:
Store the value of facts
(measures)
Are identified by multiple keys
(K>= 2)
Key tables
Describe the attributes of
dimensions
- 12-
Jones case study: Level 3
Shop
Shop#
Description
Item
Sales Item#
Shop-class Date# Billing-metric
ZIP-code Item# Item description
Shop# Bar-code#
Time Sales amount Package qty
Date# Sales qty Package-class
Week-day Number of receipts Supplier-brand
Flag work/holyday for local calendar Item-class
Date in muslim calendar
Flag work/holyday for muslim calendar
A simple implementation of the DFM is a STAR schema where key tables are
implemented only for immediate keys
Further analysis / segmentation is obtained by queries on attributes of key tables
- 13-
Jones case study: Level 3
Province-
ZIP
region
Shop
Area
Holiday
Muslim
Sales Date
date
Time
Chinese
Week-day
date
Super-
Class
class
Item
Supplier
- 14-
Level 3: design steps
6 Data extraction
- 15-
Level 3: design steps (a detail)
- 16-
Level 3: data mart
DATA MART
DATA WAREHOUSE
LOADING
TRANSFORMATION DATA ENTRY
EXTRACTION Data mart store frequently accessed
information
- 17-
Level 3: Data Mart
Customer History
Sales Analysis
Accounting
Marketing
From a same
warehouse multiple
Shop
data marts can be
Data Warehouse created
- 18-
Level 3: Hyper-cube display
Pages
Columns
Facts
- 19-
Level 3: Hyper-cube logic
Fact
An hypercube is a matrix of tables
Sales Shop
A Fact (e.g. Sales) is identified in a
multidimensional space whose axes
are Analysis Dimensions (e.g. Shop,
Time, Item)
Quantity = 20
Dimension Amount= 100
- 20-
Level 3: Hyper-cube logic
Shops
MB21000 MB31000 MB41000
Shop Item Month BUDGET
Item
MB21000 0601 Jan 50
0601 0602
MB21000 0601 Feb 55
MB21000 0602 Jan 50
MB21000 0602 Feb 60 Date
MB31000 0601 Jan 65 Jan Feb Mar Apr
MB31000 0601 Feb 45
MB31000 0602 Jan 55
MB31000 0602 Feb 50
MB41000 0601 Jan 60 ITEM
MB41000 0601 Feb 70
MB41000 0602 Jan 65
MB41000 0602 Feb 75
SHOP
- 21-
Level 3: Hyper-cube logic
Dimension Hierarchy
Dimensions are arranged in
Product Type Category
aggregation hierarchies
Svelto (roll-up)
. Washing
powder
Ajax
House Levels of hierarchies are
Dash
called dimensional
Cleaning
Soap
Palmolive
attributes
Tools
Hardware
Nuts &
bolts
- 22-
Level 3: Hyper-cube implementation
Receipt-number
This easies arithmetic
operations and keeps
hyper-cubes light
Shop
Shop
Shop
Item
Item
Item
- 23-
Level 4: presentation
- 24-
Level 4: reporting
Format editing
Semantic Layer
Data warehouse
- 25-
Level 4: reporting semantic layer
- 26-
Level 4: reporting format editing
- 27-
Level 4: reporting information distribution
- 28-
Level 4: decision supporting system
- 29-
Level 4: DSS budgeting example
Data mining applications for research and marketing are designed for
Discover in a data base relations and associations previously
unknown (data mining helps end user extract useful business
information from large databases (Berson 1997)).
Mining software is a key in marketing to calculate predictive
indicators as
Churning,
Fraud risk,
Saving attitude,
Economic potential etc.
- 31-
BI solutions offered by main vendors
- 32-
Business Intelligence Platforms: SAS
Typical data operations Insert one individual Access a vector of records Access one ore multiple a
record or modify one or Roll-up, Drill down, Dice vector of records
multiple records Roll-up, Drill down, Dice
Transaction example Enter a customer order Segment customer in Italy Segment customer in Italy
with a degree of loyalty with a degree of loyalty
>70% by age and region >70% by age and region
- 34-
Review questions
Illustrate the input, process and output of the four layers of BI systems
What is an Hyper-cube ?
- 35-