Multi-Source Attribution in Marketing
Operations and Information Management Department; Faculty Adviser: Raghuram Iyengar
Consumers often make decisions influenced by multiple sources (e.g., multiple ads across different channels and multiple contacts on social media platforms). It’s important for retailers to identify the contribution of each source in the decision process. Despite the obvious managerial implications of the multi-source attribution problem, there is limited past research primarily due to the lack of a suitable methodological framework. In this paper, we propose a general reduced-form regression model that can be easily applied to various kinds of multi-source attribution contexts.
Our model is an extension of the popular proportional hazards model, in the sense that it allows the cause of an event to be multiple sources. This model has several advantages as compared to past modeling efforts in multi-source attribution. First, the model does not speculate on the contribution of each source a-priori, but lets the data automatically determine the contribution of an individual source based on its characteristics. Second, it is easy to use and interpret as compared to structural models for attribution, and more rigorous as compared to previous rule-based attribution methods. Third, it provides unbiased estimates even if only part of the sources impacts the decisions of consumers.
The proposed methodology has implications for retailers in several contexts. In omni-channel advertising, our model can help retailers understand the quantitative contribution of individual channels, so as to optimally allocate their advertising budget across different channels. In social media marketing, our model enables retailers to identify the social influence of one user on another at the dyadic level, which can help retailers improve the cost-effectiveness of their marketing campaigns through micro-marketing strategies such as tailored seeding and tailored targeting.