|
Abstract:
|
Predicting and understanding when customers will buy is crucial for marketing managers who allocate marketing efforts in a timely manner. We propose a new approach to predict what and when customers of non-subscription companies will buy. At non-subscription companies, some customers place orders on a more regular basis, while others do so at sparser intervals, making this a challenging environment for predicting new customer activity using traditional approaches that rely entirely upon transaction data. We augment this transaction data with clickstream data through a Bayesian model to predict what product category customers will purchase and when purchases will take place. Our model clusters customers into transaction and clickstream topics, which depicts profiles of customers and their relative economic potential, and relates these topics to how long customers wait to buy. The profiles of topics and customers are directly interpretable and are of use to marketing managers engaging in promotional activities for customers. Our method discovers relationships between the timing of transactions and clickstream activity that other methods have not been to uncover.
|
