An insurance company in China is a Chinese-funded professional property insurance company. With the group strength and innovative ideas, it has rapidly achieved great development within a few years. After the rapid expansion and growth in the early stage, the company promotes the informationized marketing management continuously, so as to improve the working efficiency. Especially for the car insurance telesales business that is faced with a large number of customers and numerous telemarketing teams, it will be greatly helpful to the business development about how to increase the sales success rate by using analysis and prediction. In order to handle a large number of customer messages and achieve the prediction goals, the insurance company establishes a telemarketing response prediction model based on SPSS Modeler. It calculates the possibility of the listed customers responding to telemarketing based on the calling history. The insurance companies may choose the potential telemarketing target customers based on the response rate, which may increase the telemarketing success rate.
This test is expected to obtain the customer's existing information and finally confirm the underwriting by studying the telemarketing history. Moreover, it aims to analyze the relationship between customer characteristics and actual telemarketing response. Therefore, SPSS Modeler is used to construct a rating model to sell products to different customer groups in a differentiated way.
Since the original data of the customer lists are incomplete and scattered, it uses SPSS Modeler data processing function to process and integrate all lists. The scattered customer lists are put together to sort the telemarketing records in the specified period, which will match with the corresponding customer list.
Figure 1 Data Sorting and Matching
The sample can be selected from the matched data. Generally the response rate of the telesales is quite low. If all samples are used for analysis and modeling, the model will be affected by many non-responsive customers. Consequently, it will be difficult to successfully summarize the characteristics of the responding customers and identify these customers. Therefore, for such research, we should adjust the proportion of the modeling samples, and select all the responding customers and some non-responsive customers as the modeling samples, based on which the training sets and test sets are separated.
Figure 2 Training Sets and Test Sets
SPSS Modeler provides a variety of algorithms to construct response determination models, such as Decision Tree, Logistic Regression, Discriminate Analysis, Neural Network, Support Vector Machine, and etc., which are used to construct models for various categories.
- The overall accuracy of the model on the test set;
- The telesales repetition rate of the model on the test set;
- The response hit rate of the model on the test set;
- The top 10% and 30% improvement rate of the model on the test set.
Evaluate the model and select the model suitable to the specific business.
Figure 3 Model Construction
Modeling can predict the importance of variables. The analysis and study of these variables can provide foundation for explaining the business model, and provide instructions for the sales personnel in practice.
The Automated Modeler can compare and integrate multiple model algorithms. Through integration, it can avoid the limitations of single model, so as to improve the overall accuracy.
Figure 4 Model Evaluation
The main output result of the model is to score and evaluate the tendency of positive response among the new customers when answering the marketing calls. After evaluating the scores for all customers in the source list, it will assign the selected ones to the specified telesales personnel based on the actual situation of the telesales team.
The predictive response model built based on SPSS Modeler can optimize the telemarketing business model, control costs, and increase marginal benefits.