Statistical and Machine Learning Approaches to Predict the Next Purchase Date: A Review

In the era of the digital revolution, customer expectations have drastically changed. To meet and exceed customer expectations, companies are transforming their marketing strategies by anticipating the date when a particular customer is likely to purchase a specific product. By doing so, companies can predict a customer’s next order and can target the particular customer at the right time. This would ensure that the specific product which is likely to be bought by the customer doesn’t go out of stock on the predicted date of purchase. The customer also benefits by experiencing a personalized touch by seeing fewer and relevant ads. However, predicting the next date of purchase still remains a challenge as the cost of wrong predictions would directly lead to unfavourable outcomes for the company. There has been a significant number of researches with an aim of reducing the difference between the actual date of purchase and the predicted date of purchase. The researchers experimented with numerous statistical techniques to estimate when a particular customer would re-purchase a specific product. However, statistical models were not flexible enough to adapt to dynamic consumer buying behaviour. Researchers wanted to build a robust model and thus, experimented with multiple machine learning algorithms such as Linear Regression, XGBoost model, Neural Network, and Recurrent Neural Network. Among these algorithms, neural network model outperformed others. This paper attempts to review each of the statistical and machine learning models developed by researchers for predicting the next purchase date (NPD) of a customer-product pair.