Customer relationship management has shown its critical role in the success of a company in today’s increasingly competitive service industry. While modern machine learning techniques are widely adopted due to their advantages in working with extensive databases, organizations with little customer information and low-quality data have limited choices when analyzing customer data to retain existing customers. This thesis proposes two approaches to model and forecast existing customer attrition, survival analysis, and regression analysis. The proposed methodologies are demonstrated through customer data from 10 retail energy companies at different data quality. Results from both proposed models show superior performance in terms of Mean Absolute Error, Mean Squared Error and Mean Absolute Percentage Error, compared to a commonly used non-parametric model that forecast attrition rate based on the average of known customers.