Our client is India’s largest foodservice company and wanted to identify customers who could be retained using a Churn Prediction model developed using AI and ML techniques. This decision was prompted by the increasing churn rate during the coronavirus lockdown. The goal is to develop a model for predicting customer churn and utilize the insights from the churn prediction model to assist marketing and CRM teams in creating an incentive plan to encourage potential churners to stay with the client.
Overview
Solution
Data discovery exercise
- Understand existing databases and IT infrastructure (data collection, pre-processing, Exploratory Data Analysis (EDA))
Data integration
- Framework to create an integrated data lake using Reinforcement Learning, deduplication and ensemble learning.
- Data cleaning ,stitching and profiling using Feature engineering, Gradient boost machines (GBMs) and anomaly detection.
Stratified sample considered for modelling purpose
- Developed secondary variables based on inputs from cross-functional teams
Predictive modelling technique based on classification algorithms such as Logistic Regression, XG Boost, Random Forest, Support Vector Machines (SVM), Naïve Bayes, AdaBoost etc. have been used to identify customers with high propensity to churn.
Output
- XGBoost model has been finalized for customer churn prediction based on consistent results in validation and higher accuracy over other algorithms. Other evaluations metrics like precision, recall, F1 score, and ROC-AUC were used.
- Client in process to develop target based marketing offers based on customer segment and churn score.
