How about OLAP? It’s really just more reporting.
Operational systems alone are very expensive
They require a long time to implement.
They don’t really give you a measurable ROI.
THEY ARE ABSOLUTELY NECESSARY TO DO BUSINESS IN TODAY’s WORLD
Business Intelligence systems require a second wave of investment
They improve your knowledge about your business
They can help you get to ROI
Predictive analytics require a relatively small investment
They can get you to profitability fast
Traditional predictive analytics get you to the first green hump, which is directly limited by the number of Analysts and Statisticians
Only automated and embedded Analytics like KXEN get you the steep incremental ROI
Data Mining is an iterative process that receives data in a relational or flat file form as its input and generates valuable information as output (i.e., the underlying patterns and relationships of the data).
It is important to understand/define business objectives and to keep business knowledgeable people involved at all stages of analysis.
The data must be prepared for analysis during the Data Pre-Processing step, This step can take any number of forms, such as sampling, statistical analysis, correlation analysis, and/or cluster and feature analysis. In addition, some common data warehousing steps are performed. The data is checked for consistency, consolidated, and organized. At this point business rules can also be applied to the data. Any operations, such as the calculation of derived parameters or aggregation of data in fields or discretization of variables, are completed as well The data should also be available for exploration at all times during the Knowledge Exploration process, since data mining is a highly iterative process.
The next three steps are the ones most often associated with data mining: the construction, validation, and verification of the model.
During model construction, the selected techniques and tools are applied to construct the data mining model (e.g., determine rules for identifying patterns, build models for customer scoring).
Model validation and verification are collectively referred to as model evaluation. They are in fact two separate activities required for testing/evaluating the models.
The model is validated based on pre-determined levels of accuracy and precision. Precision, consistency of the results, is determined by using a test data set that was not used for model construction. Accuracy, goodness of the answer, is determined by yet another test set. It should be noted that, in general, there is always a trade off between the accuracy of the results and the complexity of models, namely, as the model becomes more complex, the accuracy of the model generally increases.
Then, the model is verified against current business knowledge and the preliminary data analysis results, in order to confirm that the business objectives are met.
Finally, the data mining insight is deployed. It can be deployed by rolling out a “scoring model” or applying new insights to business (for example, offering specific cross-selling suggestions to a customer)
1) Data Quality - Data is required. Specifically clean, good quality. Volume of data is often less important than the data relevancy to the business question under examination. Often the largest amount of time is spent on data cleansing and calculating the relevancy of the data. Do not rely on billing or production data only. To do CRM you need personal customer information for example.
2) Do not start a data mining project without the proper skills and consultant experience. This would most often set the expectation too high and will confuse and even demotivate the client. A strong data miner is mandatory. S/He can supervise a team of several beginners that can be statistician at the beginning, Since Data Preparation (Data set or mart creation, data cleansing and consolidation) is mandatory, a couple of database (SQL - data) knowledgeable person is required.
3) Clearly Defined Business Objectives - data mining needs guidance in order to focus the analysis. For example, is the purpose to understand what customer segments exist? Find out who is loyal? Find out who would be likely attrite?. The objective must be documented. The integration of the knowledge at the end of the discovery phase must be thought at the beginning of the phase. If knowledge is discovered, who will be impacted, what process needs to be modified….
4) A data Mining project must be driven by the Business. It can be driven by the marketing department but should not be driven by the IT department. One will focus on small important but tactical problems whereas under IT umbrellas the outcome might not serve the business purposes. The Sponsor must be above the IT and Marketing head. The acceptance from these 2 groups’ heads is however mandatory.
5) Reasonable and planned ROI. Do not pass the message of gold nuggets inside the dusty mine, you will put yourself in trouble. The economic model (like in a business case) must be set as early as possible. The ROI must be planned with some assumption for the prediction rate. One must be careful with ROI. Do not sell gold before having seen the smallest nugget.
6)Analysis Approach/Iterative - The business objectives and the data characteristics are taken into account when developing the model (or analysis approach). Certain techniques lend themselves better to specific types of data (categorical, continuous etc.) and to certain types of questions (ie., predictive, induction). For example very different approaches can be taken to identify fraudulent transactions 1) can look for fraud based on known fraudulent behavior/transactions 2) can look for fraud by looking for transactions that are atypical. Data mining model development is an iterative process, the business reviewers must be available for frequent and short reviews to drive the data mining team work with the DM manager.