Guide to Business Intelligence Part 2

Business Intelligence: Implementations and Warehouse Concepts Chapter 2. Intelligenthq

Having examined what business intelligence is and why it is important to business, as well as understanding its key concepts, in the first chapter of this guide to business intelligence over the second chapter of our Guide to Business Intelligence, I will look at implementations and warehouse concepts.

Starting with Business Intelligence implementations, these are focused on trying to find the best way to deliver the requirements of online analytical processing. According to Joerg Reinschmidt and Allison Francoise of IBM there are four different ways of achieving this outcome. These vary in terms of time to implement and robustness, and are:

1. Summary tables on operational machine – these are the least robust but have the shortest time to implement, making them a good interim measure for some businesses, or an approach for businesses just starting out that need to get “quick and dirty” BI on what they are doing. This is actually a fairly common approach among many types of standard software packages.

2. OTLP data moved to separate DB server – this is a bit more robust than the first option, but as a result also takes a bit more time to implement. It also requires space on a separate DB server so may accrue cost. With this approach while data is moved, the database structure is not changed. Joerg Reinschmidt and Allison Francoise call this a “poor man’s data warehouse”. It only offers limited flexibility, and is usually not optimised for query performance.

3. Single Data Mart – this is yet more robust that the previous option, and as a result does take longer to implement. These often start out as a not too complex solution and then grow and develop over time. This option can deliver high performance, and it is possible to get hold of industry specific solutions. For some businesses, nothing else is needed but this approach.

4. Three Tiered Data Warehouse – this is a very robust option, and it takes the longest time to implement compared to the other options. Under this model, the OTLP data is stored in operational databases, and there are “multiple, aggregated and pre-calculated data marts” (Reinschmidt and Francoise) so that the data is presented effectively to the user. Query performance is optimized. This is the best solution but maybe it may not be necessary to go this far for some businesses.

Turning to the data warehouse, this is made up of a number of different components. These components are external sources, operational databases and meta data. There are also a number of processes involved in the operation of a data warehouse, including extraction and/or propagation, transformation (cleansing), data refining, presentation and analysis.

There are a range of different data sources that a business intelligence implementation will usually use. These may include historical data, current databases that are used in daily operations, data from third parties or the Internet, and data from pre-existing data warehouses. All of these different data sources offer relevant and important information to businesses and being able to use them in conjunction with one another is a necessary part of good business intelligence.

During the extraction or propagation process data is collected and drawn into the data warehouse. Once in the data warehouse it has to be cleansed or transformed. This process is particularly important so that the data can be used within the system and can be included in reporting. Transformation includes putting in place business rules for the data, and making sure that the data will actually be useful and meaningful to the end user.

Image Source: IBM Redbooks

When data is being refined it is becoming prepared to be able to be used for reporting. This involves creating data sub sets, calculated fields and virtual fields and summarizing the information that is available. That data must then be aggregated as well. Once it has undergone this process, which might be achieved via a range of different models or technical approaches, the data can then be presented to the user and analysed. This final step allows data to be prepared for people who use the system infrequently but also those that are interested in looking at more complex analyses by creating and generating their own reports based on the data that is available within the system. Of course, reporting is the end goal of the system, so that users are better able to make informed business decisions and seeking input on the types of reports required is an essential step in the development of a good BI implementation.

Guide to Business Intelligence (part 1): An Introduction
Guide to Business Intelligence (part 2): Implementations and Warehouse Concepts
Guide to Business Intelligence (part 3): Project
Guide to Business Intelligence (part 4): Data Sourcing/Movement 
Guide to Business Intelligence (part 5): Solutions Architecture