Customer Life Time Value Analysis
Download as PPTX, PDF1 like700 views
This document discusses predicting customer lifetime value (CLV) for an auto insurance company using various machine learning models. It outlines preprocessing the data, feature engineering, building models like linear regression, ridge regression, decision trees, random forests, and evaluating the models using R2. Random forest was found to have the best performance with an R2 of 0.913. The conclusions recommend focusing on male customers, specific coverage offers, and tailored sales strategies based on customer demographics to improve CLV predictions.
Customer Life Time Value Analysis
- 1. CUSTOMER LIFE TIME VALUE (CLV) PREDICTIONS Auto Insurance Company Made By Soumit kar 1
- 2. V Slide 1 Problem Statement and Business Overview Data Pre Processing and Feature Engineering Model Building R2 Evaluation Data Insights Conclusion and Recommendations AGENDA Made By Soumit Kar 1 2
- 3. Objective Company's Approach and Insight We are expected to create an analytical and modelling framework to predict the lifetime value of each customer based on the quantitative and qualitative features. • A major non-life insurance company wants to evaluate customer lifetime value based on each customer’s demographics and policy information including claim details. • The CLV is a profitability metric in terms of a value placed by the company on each customer and can be conceived in two dimensions: the customer`s present Value and potential future Value. Made By Soumit Kar 3 OVER VIEW
- 4. Made By Soumit Kar 4 BUSINESS STRATEGIES
- 5. Made By Soumit Kar 5 PROFIT PROFIT TIME More Efficient acquisition of new customers Better Cross / Up selling More Efficient Customer Retention Recovery of Potential Valuable Customers Faster Termination of Less Valuable Customer Acquisition Retention Termination CUSTOMER LIFE CYCLE
- 6. . Made By soumit Kar 6 PROFIT PROFIT TIME More Efficient acquisition of new customers Better Cross / Up selling More Efficient Customer Retention Recovery of Potential Valuable Customers Faster Termination of Less Valuable Customer Acquisition Retention Termination CLV HELPS TO ANSWER THIS…… Do the new customer has same profile as valuable customer? Do the Customer have high probability of cross- selling? Do the Customer have high potential? Do the Customer have low potential?
- 7. 7 1. Parameters Used ● Business centric parameter ● Customer centric Parameter ● Demographic Parameter 2. Combine sets used for Top 5 and Bottom 5 Customers. 3. Information and navigation button. 4. Histogram for all variables in Tooltips. Components Used in Dashboard
- 10. 10 Policy Age segment= Months Since Policy Inception > median, high else Low CLV segment = CLV > median, High else Low CS=High PAS=High CS = Low PAS= Low CS = High PAS=Low CS = Low PAS= High
- 11. DATA PRE-PROCESSING AND FEATURE ENGINEERING Data Pre- Processing Dataset was given with 9134 data points and 24 variables. There is no missing value Drop columns those add no significance. A new Features are been extracted based on the existing variables. New Features are: Present Value of Customer, Clubbing the Various Levels into one level. Applied one hot encoding for categorical columns See the correlation features. Using the VIF and statistical analysis some of the variables are been removed. Feature Engineering Variables Extraction Made By Soumit Kar . 11
- 12. FEATURE ENGINEERING . Made By Soumit Kar 12 Present Value of Customer = Monthly Premium Amount Months since policy Inception Total Amount Claim * - Bachelor College High School or Below Doctor Master
- 13. DEMO GRAPHS OF THE FINAL DATA FRAME . Made By Soumit Kar 13 Customer Centric Business Centric Customer Demo+ graphs Income Sales Channel Education Employment status Monthly Premium Value Marital Status No. of Policies Policy & Offers Location Vehicle Class Coverage Gender Initial Data Frame with 9134 data points and 24 feature variables Dataset Summary: Total Observations : 9134 Total variables : 24 Split : 70-30 Target Variable : Customer Life Time Value Scaling : Based on the model Evaluation Metric : R2
- 14. 14 Statistical Analysis Proceed with non parametric tests since the dependent variable is not normally distributed Shapiro–Wilk test Mann-Whitney U test Kruskal-Wallis H-test Assumption No Auto correlation = Durbin- Watson Test Normality of Residuals = Jarque Bera test Linearity of residuals = rainbow test Homoscedasticity Test = Goldfeld Test No Multi colinearity = VIF
- 15. MODEL BUILDING Made By Soumit Kar . 15
- 16. 16 Models R^2 Linear Regression 0.25484160264997946 Ridge Regression 0.2148084953609083 Decision Tree 0.840787974934538 Random Forest 0.9139544276201418 Hyperparameter Random Forest 0.9123540914308492 Hyperparameter Adaboost 0.8945253809639656
- 18. CONCLUSION AND RECOMMENDATIONS ● As the Customer who is using Four Door Car and Two Door car more inclined to coverages and even the Loss incurred by them is more. ● Focus on male customers and promote offer 3 and offer 4. ● Even Group 3 Can be Bought in to Group 4 by giving offers to them and even new attracting policies. ● Apply better sales strategy for Doctor’s and master’s students as they are busy ● Concluding that Random Forest Model is stable in order to predict the CLTV. . Made By Soumit Kar 18
- 19. THANK YOU Made By Soumit Kar . 19