Supporting organizational decisions on How to improve customer repurchase using multi-instance counterfactual explanations

Improving customer repurchase intention constitutes a key activity for maintaining sustainable business performance. Returning customers provide many economic and other benefits to businesses. In contrast, attracting new customers is a process that is associated with high costs. This work proposes a novel counterfactual explanations methodology that utilizes textual data from electronic word of mouth to recommend business changes that can improve customers' repurchase behavior. Counterfactual explanation methods gained considerable attention because their logic aligns with human reasoning and the fact that they can recommend low-cost actions on how to turn an unfavorable outcome into a favorable. Most counterfactual explanation methods however recommend actions that can change the outcome of individual instances (i.e. one customer) rather than a group of instances. Therefore, this work proposes a multi-instance counterfactual explanation method that recommends optimum changes to an organization's practices/policies that increase repurchase intention for many customers or customer segments.

The proposed methodology utilizes topic modeling to extract customer opinions from online reviews' text and use topics as features to train a binary classifier that predicts customer revisit intention. Multi-instance counterfactual explanations are computed for all or different groups of non-revisiting customers, recommending optimum business changes that can increase revisit intention. The proposed methodology is empirically evaluated through a case study on the restaurant revisit problem and compared against a prominent alternative from the literature. The results show that the method has better performance to the alternative method and produces recommendations that are actionable and abide by the customer-repurchase literature.