A Graphical Point Process Framework for Understanding Removal Effects in Multi-Touch Attribution

Abstract

Marketers rely on various online advertising channels to reach customers and are increasingly interested in multi-touch attribution, which evaluates the contribution of each touchpoint to a conversion. However, as the numbers of marketing channels and touchpoints increase, the attribution challenge becomes more intricate because of the complex interplay among different touchpoints within and across channels. Utilizing customer path-to-purchase data, this article addresses this challenge by developing a novel graphical point process framework to investigate the relational structure among various touchpoints. Based on this framework, we propose graphical attribution methods that allocate attribution scores to individual touchpoints or corresponding channels for each customer’s path to purchase. These scores are calculated using a probabilistic definition of removal effects. We evaluate the proposed methods and compare their performance with commonly used attribution models through extensive simulation studies and a real-world attribution application. This paper was accepted by J. George Shanthikumar, data science. Funding: This work was supported by the National Science Foundation [Grant DMS-2210775]. Additionally, the research of L. Xue was partly supported by the National Science Foundation [Grants DMS-1811552 and DMS-1953189]. Supplemental Material: The electronic companion and data files are available at https://doi.org/10.1287/mnsc.2023.00457 .