PROBLEM OVERVIEW
A large Automotive Insurance provider wanted to understand claims data in order to improve customer satisfaction and reduce operational expenses.
The objective was to improve the productivity, accuracy, and consistency of the claim handling process and minimizing the risk of fraud by profiling the customers who have claimed in the past and then defining business rules using variables that are important to understand an insurance claim. Thus, resolving the legitimate claims in just one phone call.
|
Data & Methodology
-
Sample size= 1.5 million transactions / 150K customers
-
Statistical Techniques: Decision Trees, Logistic Regression and Neural Networks
-
Tools Used: RapidMiner 5.0 and SAS
|
 Download this page (Predicting and Profiling Claims) in PDF
|
Analytical Solution
How data was analysed? Data was collected from various sources including historical claim data from the enterprise data warehouse and CRM system and processed through various analytical techniques to pick the best fit model that addressed the problem.
What was the specific analytics objective? Understand the profile of the claimants using advanced analytics techniques such as decision trees, neural networks and logistic regression. Create predictive models and pick the best fit model to predict an equation that can be used to qualify a claimant. Many thousands of iterations were run to create the 24 models that identify and quantify relationships among various factors, and thereby accurately predicting the likelihood for a customer to apply for a claim.
Managerial Applications:
-
Quick Claims: A quick decision on the claim was possible by just testing the scenarios and asking key questions to fit the model, helping in streamlining the claim process and reducing claim costs by 15%.
-
Fraud Management: Initial screening on the fraudulent claims was possible just by using the Neural Networks Model, thus reducing risk of a fraud.
Benefits: By using the analytical techniques tailored and best suited for the business, operational costs were reduced and fraudulent claims were identified thus saving money for the company.
Decision Tree explaining the profile of the claimant
|
