Data Mining For CRM: One Way to reach the goals

Following the lecture by Dr. Ilona Jagielska, this week reading materials reveal several data mining techniques, such as neural network, decision tree and CART methodology, and association rules. The three reading materials illustrate how those techniques work, the strengths, weaknesses, and also describe a study case in predicting whether a certain visa cardholder is a good or bad customer.

After reading all those materials, I think there are several important issues to be considered:

First is about the techniques used for mining the customer data. In my opinion, there is no technique that can be claimed as the best method for all problem domains. All techniques have their own characteristic, strength and weaknesses. For that reason, it is important to choose the proper technique and then build a model. In order to do that, the problem should be addressed properly. A model has the same nature as technique; it might only work best on a certain problem. The best model can only be defined according to certain cost or risk acceptable by ones who implement it.

Second, besides technique and model, one thing that is crucial is data. There is a jargon about it, “garbage in garbage out”. No matter how good the model is, if the quality of data is poor, it would not give the expected result (but data with the good quality not merely guarantee the good result. There is a possibility that the model is not suitable). That’s one reason why we should prepare the data.

Nevertheless, those techniques are helping companies in many things related to building and retaining a good relationship with their customers. They help companies to get to know their customers, to predict what their new customers would be like, to scale up their sales by cross selling and up selling since they know the customers now, etc. Companies can also apply the data mining technique to build a tool that would support customers. For example, related to the problem of choice overload, one data mining technique such as clustering analysis can be used to group the customers based on their similarity and then suggesting the choice based on customers on the same cluster. It can be both beneficial to companies and customers. Then it would bring companies a step closer to the goal………