RFM and Classification Predictive Modelling to Improve Response Prediction Rate

In consumer electronics where sales cycle is about two to three years, and with increased competition and product differentiation faced by suppliers in online distribution channels, it is important to pay attention to targeted marketing in online consumer electronics sales through the use of predictive analytics, as marketing paradigm is becoming increasingly customer-focused and unsolicited marketing is often costly and ineffective due to low response rates. In this study, customer predictive analytical techniques, including the RecencyFrequency Monetary (or RFM) method and classical classification modelling methods – logistic regression, decision tree, neural network and ensemble models – are utilized to improve predictive accuracy. Results from the neural network model shows a significant improvement over RFM model, with positive response rates improving by more than 2x, from 42.9% to 87.2%. However, if stronger explanability power is preferred, decision tree model may be utilized, although predictive accuracy of about 2% is sacrificed. The study discusses predictive modelling useful to improve the performance of positive response rate targeting, alongside the benefits of improved sampling and reduced computing power, especially with significantly large datasets. In real life implementation, it is imperative that companies understand that classification power of the models and marketing campaign targeting are continuous improvement processes. These processes improve with every iteration from its baseline towards its objective threshold level set by the companies’ management. False positive transactions should be investigated, with the effect of incorporating the findings to the improvement of models going forward.