Rafal Lukawiecki : Architectual use of Business Intelligence in Application Design

Introducing the idea as BI as a development platform

Data Warehousing and OLAP as application development platform.

Innovative uses of BI for non BI applications

1. Using data warehouse and OLAP in application design
DW for storage?
- Relational DW is an excellent place for all non-transactional data
- Easy queries (star join with optimization)
- MDM for holistic information design
- No need for OLAP

Facts are facts, once written, read-only
Change?= another fact all woth a date stamp (historical versioning)

Use Dimensions - Modern navigation in structures.
Veru compact and beautiful

OLAP - Deductive database (mini)
Use OLAP for logic to de-couple it from the application
Time-partition give support for changing requirements (e.g. calculations) over time. The logic from the pas can always be called.

Embed Predictive KPIs (using DMX and MDX - I have never heard of these terms, some readng to do..)


2. Using data mining in architecture of intelligent applications


What does data minin do
- Finds patterns
- explores your data
- preforms predicions

Examples of intelligent applications.
Business process validation - early detection of failure based on past failures and successes
Adaptive user interface - based on past behaviour
Input validation - based on accepted data, not on fixed rules

This is also known as Predictive programming

What is so special?
Application behaviour evolves and follows its data mining model - influenced by events caused by your appplication. This is AI by the backdoor... :-)

Data mining API's. There are several, but ADO.NET is the most popular.
Need to learn DMX (Data Mining Extensions)

Rafal now had a demo of how to use his for input validation. The app is available on the web (did not get the address..).

The validation logic was driven from the data mining.

What to do:
1. Prepare history (events) for mining
2. Create and rain DM model on your data
3. Test and validate the model
4. The model predicts outcomes
5. Make application logic depend in predicted outcomes
6. Update and re-validate your model periodically

My takeaway: Can be interesting in some cases, but I personnally need much more knowledge in BI before I can evaluate this as an alternative for anything.

Rafal predicts that this will be a standard part of the toolbox used in any application in the future.