Customer churn and retention rates are considered as the most important indicators to CRM activities such as modelling customer valuation and budget allocation. Such models, which are trying to predict customer’s survival rate, assume that churn rates are constant both over time and across customers. The current practice, however, have shown that these assumptions are not valid and are currently tested in a systematic manner. The practice shows that churn propensity vary dramatically across customers and over time; yet, most models do not account for such differences. This study takes under consideration a number of factors that may underlie customer retention patterns. Specifically a model of mobile services’ retention is developed that takes into account: duration dependence, cross-cohort effects, customer heterogeneity, and contractual obligation type. Moreover, a methodology is proposed that easily allows practitioners to evaluate the relative importance of each of these effects based on in-sample and out-of-sample performance using limited information. Using a dataset from a major Greek telecommunication company concerning customers’ gross additions for twelve months and the relevant disconnections for the same time period ahead, two models on aggregate data at the monthly level are tested. The proposed modelling approach reveals that the notion of a constant churn rate is rejected. The representation of customer’s retention process requires all the proposed components listed above towards improving the retention and churn rate forecasting performance.