Business Analytics I
Alexander Prosser
First Generation BI
SEITE 2
ERP/other operational systems Data warehouse
Aggrega-tion
hierarchy
Graphical representation
First Generation BI
SEITE 3
ERP/other operational systems Data warehouse
Aggrega-tion
hierarchy
Graphical representation
Why aggregation?
In-memory databases:
Disk access – msMain memory access – nsDifference: 106
Second Generation BI
SEITE 4
ERP/other operational systems Data warehouse
Aggrega-tion
hierarchy
Graphical representation
In-Memory
ERP BI
Second Generation BI
SEITE 5
ERP/other operational systems
In-Memory
ERP BI
Graphical representation
Machine learningArtificial intelligence
Run through individual records
BI & AI
Second Generation BI
SEITE 6
“Large main memory”. How large is “large”?
https://www.ibm.com/downloads/cas/VX0AM0EP
Mere example, performance industry standard.
How much is 64 TB?
Average Netflix HD Movie 2 GB => 32k+ moviesNetflix currently offers less than 10k movies or TV shows*
High-quality portrait 1MB => 67m+ photos
… and you can search that content in main memory in a matter of a few seconds
* https://www.businessinsider.com/netflix-movie-catalog-size-has-gone-down-since-2010-2018-2?r=DE&IR=T?r=US&IR=T
Modelling
SEITE 7
The technology may have changed, fundamental case modelling has not.
=> Dimensions and facts => Dimensional Fact Modelling
Let us design a BI system
Modeling
SEITE 8
STEP 1:
What is the fact I want to analyze ?
What are the key figures representing the fact ?
What do the key figures look like ?
Modeling
SEITE 9
()
Operator Nominal Ordinal Interval Rational
Sum No No No
Average No
Minimum No
Maximum No
Modeling
SEITE 10
STEP 2:
What are the axes in my analyses ?
What are their aggregation levels (if any) ?
Modeling
SEITE 11
STEP 3:
Are there any restrictions in aggregation ?
Modeling
SEITE 12
Additivity
Stock_level
Y M W
Storage_location
Plant
*
Material
Material_group
*
=> AVG Σ
Σ
Σ
Modeling
SEITE 13
x
Σ
AVG
min
max
Some dimensions All dimensionsSome aggregation operator Semi-additive Semi-additiveAll aggregation operators Semi -additive Additive
These are logical restrictions.
No technology in the world changes that.
Modeling
SEITE 14
STEP 4:
Do I have parallel hierarchies ?
Modeling
SEITE 15
STEP 5:
Where does the data come from ?
Do I need to reconcile data from different sources ?
Modeling
SEITE 16
Operational IS
DW
Key Integration
Key_1
Key_2
Example:• Accounts receivable• Customer• Transport destination
one object inDW
Modeling
SEITE 17
Operational IS
Field Integration
• Currencies• Measurements• Scope of figures (eg, gross/net)• … ?
• All fields available ?
DW
Filter:
Modeling
SEITE 18
Operational IS
Content Integration
• Material classes the same ?• Account assignment the same ?• Data maintenance discipline/rules the same ?• … ?
DW
Example MM/Procurement:
Modeling
SEITE 19
Dimensional Fact Modelling
Case Study: Mobile phoneprovider helpdesk
SEITE 20
The company looses a large number of customers and wonders why …
Helpdesk calls are to be analysed for customer sentiments towards three topics:- Price- Technical quality - Service quality
This is a combined task for second-generation BI and AI.
Case Study: Data model
SEITE 21
Interaction DateCustomerCustomer group
Interaction type
Download the data …
Case Study: Data flow
SEITE 22
Data sourcefile
Data sourceHana
Data source3rd party
aDSO Corporate memory
Character-isticsKey figures
“Cube”aDSO
Data warehouse core layer
Data source
Case Study: Reporting
SEITE 23
“Cube”aDSO
Reporting.xls
Virtual data mart
Case Study: Voice data analysis
SEITE 24
Voice clips Text filesSpeech to text
Case Study: Voice data analysis
SEITE 25
Text files Clean text filesExtract
& split
R
“Hello, my display is super and your service leaves a lot to be desired”
“1. my display is super” “2. your service leaves a lot to be desired”
Case Study: Voice data analysis
SEITE 26
Annotated text elements
Annotat-ionClean text files
Categories.csv
“1. my display is super” “2. your service leaves a lot to be desired”
Case Study: Voice data analysis
SEITE 27
Annotated text elements Input HANA
Senti-ment
analysis
Sentiment library
“1. my display is super” “2. your service leaves a lot to be desired”