[Skip Global Navigation]

Customers

Customers Home

Winterthur Insurance

Challenge

Winterthur Insurance has more than 1 million customers in Spain—and more than 130,000 cancel their car insurance policies each year. Given the loss of revenue and the cost of underwriting new customers, customer churn was a very expensive problem. Winterthur needed to predict which customers would leave—and why.

Solution

After inviting several data mining vendors to produce predictive models from a large data set, Winterthur then tested models on real, unseen data. The objective evaluation led Winterthur to choose PASW Modeler, with its combination of powerful algorithms and easy-to-use data visualization methods, as its modeling tool of choice.

Interested in Winterthur Insurance? Download the PDF

Winterthur’s initial dataset contained revealing details on its car insurance policy holders, with records containing 250 descriptive fields per case. Using PASW Modeler’s data visualization methods, and combining business knowledge of Winterthur’s situation, the first step was to reduce this data set to a core set of relevant fields.

To identify the 30 most significant fields, Winterthur used four approaches: sensitivity analysis, rule induction, statistics, and common sense. As a combination of neural networks, the fully developed model correctly predicted who would cancel their policies for an impressive 90 percent-plus of the blind test data.

First, PASW Modeler produced a score for each customer, indicating the likelihood of cancellation. Then, Winterthur used each score to order the data. This resulted in 75 percent of all canceling customers appearing within the first 15 percent of the data set.

Armed with the invaluable understanding of which customers were most likely to cancel, Winterthur was able to focus more readily on reducing customer churn and retaining profitable customers.

The fully developed model correctly predicted who would cancel their policies for over 90 percent of the blind test data.