Sample Projects and Implementations

 The main objective of the project was to find common insurance policy baskets, taking into account the order of purchasing policies. The additional objectives included: assessment of the influence that paying out compensation had on customers’ decision to stay, as well as assessment of the cross-selling rate among customers who buy a given insurance type.
Demographic data along with data on the cooperation with two million customers were prepared and recorded in SQL Server 2008 R2 table. Data concerning several millions of insurance policies were aggregated, numbered and recorded in a related subordinate table.
Thus prepared data served to find (using Microsoft Sequence Clustering algorithm which combines segmentation with sequential analysis) the most common sequences of purchasing policies.
 Secondly, models analyzing insurance policies of the customers who decided to take out a tourist policy were created (using Microsoft AssociationRules and Microsoft Sequence Clustering algorithms). This resulted in finding:
1. A set of strong rules (associations between tourist policies and other policies)
2. A list of popular baskets containing minimum one tourist policy
3. A list of frequently found purchase sequences beginning with a tourist policy purchase
The third stage of the project involved creating (with the use of Microsoft DecisionTrees and Microsoft Sequence Clustering algorithms) models assessing the effect that the compensation payout had on continuing business with a particular customer.  The models helped to:

  1. Determine the factors most influencing a customer’s decision to either stay or leave, and the percentage estimation of their influence.
  2. Identify and compare customers’ features and the order of their policy purchases, which allowed to profile both the customers deciding to continue business and those deciding to discontinue.
  3. Identify the most common policy purchase sequences resulting in a customer’s decision to discontinue business.
  1. The most common policy purchasing sequences were determined.
  2. Lists of policies purchased in the first and in the last place were compiled, along with the probability of their choice.
  3. The percentage of customers taking out only travel insurance and the percentage of customers later purchasing other insurance was determined.
The influence that paying out compensation has on a customer’s decision to stay and continue business with the Insurer was assessed.