MAI were quite excited about the outcome of Angie's research. The MAI management was in the process of reviewing their current point system of premium pricing and they agreed that it needed to be revamped. Angie's results had shown that strong quantitative benefits were theoretically possible if the proposed data mining solution was adopted.
However there were several issues that needed to be resolved before MAI could begin to implement the data mining approach. Firstly the approach needs to be validated; for the various Australian states. The research was based on only one of the Australian states and there were differences in the premium pricing for the various states. Should the approach be validated using real cases or historical data? Also, the data mining framework does not model the effect of competition. Can the approach be implemented if it has only considered the dynamics of MAI in isolation from their competitors? How can competition be factored into the framework? If it is mathematically too difficult to consider the effect of competition, how should MAI proceed?
MAI do not have any data mining expertise and none of the MAI staff were very involved in the Angie's research project. It is therefore difficult to transfer the skills and knowledge acquired during the project to MAI staff to carry out the validation and implementation. MAI realise that data mining is more than just acquiring the software. Data mining expertise is required to decide which algorithm is most suited for a problem and to interpret the results. Should they recruit people with the data mining skills or should they train the current business analysts to do future data mining work?
Implementing the proposed data mining framework will also require significant business process re-engineering. How will staff react to the changes? How can resistance to change be managed? How are they going to integrate data mining into the existing information system infrastructures?
Since the data mining approach is "modular", the pricing manager, Ryan Lee, suggested implementing the data mining approach in phases. They could use the MAI's existing risk groups to replace the clustering stage of component one, and use neural networks to model the price sensitivity of these risk groups. If the neural networks proved to be successful, they could then look at implementing the integer programming for determining the optimal premium to charge for each risk group. The final phase would be to look at implementing the clustering method of risk classification.
Clearly there are many practical considerations that MAI need to resolve before the proposed data mining approach can be adopted. Some of these are related to personnel and change management, while others are more technological considerations. Until these issues have been resolved, the project has only shown the theoretical benefits that could be obtained.