Increasingly, on-demand nature of customer interactions put pressure on companies to build real-time campaign management systems. Instead of having managers to decide on the campaign rules, such as, when, how and whom to offer, creating intelligent campaign management systems that can automate such decisions is essential. In addition, regulations or company policies usually restrict the number of accesses to the customers. Efficient learning of customer behaviour through dynamic campaign participation observations becomes a crucial feature that may ultimately define customer satisfaction and retention. This paper builds on the recent successes of deep learning techniques and proposes a classification model to predict customer responses for campaigns. Classic deep neural networks are good at learning hidden relations within data (i.e., patterns) but with limited capability for memorization. One solution to increase memorization is to use manually craft features, as in Wide & Deep networks, which are originally proposed for Google Play App. recommendations. We advocate using decision trees as an easier way of mining high-level relationships for enhancing Wide & Deep networks. Such an approach has the added benefit of beating manually created rules, which, most of the time, use incomplete data and have biases. A set of comprehensive experiments on campaign participation data from a leading GSM provider shows that automatically crafted features make a significant increase in the accuracy and outperform Deep and Wide & Deep models with manually crafted features.

In many tasks, classifiers play a fundamental role in the way an agent behaves. Most rational agents collect sensor data from the environment, classify it, and act based on that classification. Recently, deep neural networks (DNNs) have become the dominant approach to develop classifiers due to their excellent performance. When training and evaluating the performance of DNNs, it is normally assumed that the cost of all misclassification errors are equal. However, this is unlikely to be true in practice. Incorrect classification predictions can cause an agent to take inappropriate actions. The costs of these actions can be asymmetric, vary from agent-to-agent, and depend on context. In this paper, we discuss the importance of considering risk and uncertainty quantification together to reduce agents' cost of making misclassifications using deep classifiers.