Zipei Lu

Abstract: When analyzing a sequence of customer interactions, it is important for firms to understand how these interactions align with key objectives, such as generating qualified customer leads, driving conversion events, or reducing churn. We introduce a transformer-based framework that models customer interactions in a sequence similar to how a sentence is modeled as a sequence of words by Large Language Models. We propose a heterogeneous mixture multi-head self-attention mechanism that captures individual heterogeneity in touchpoint effects. The model identifies self-attention patterns that reflect both population-level trends and the unique relationships between touch points within each customer journey. By assigning varying weights to each attention head, the model accounts for the distinctive aspects of the journey of each user. This results in more accurate predictions, enabling precise targeting and outperforming existing approaches such as hidden Markov models, point process models, and LSTMs. Our empirical application in a multichannel marketing context demonstrates how managers can leverage the model’s features to identify high-potential customers for targeting. Extensive simulations further establish the model’s superiority over competing approaches. Beyond multichannel marketing, our transformer-based model also has broad applicability in customer journeys across other domains.

Abstract: Despite the abundance of data in the digital ecosystem, emerging consumer privacy regulations have increasingly constrained firms’ ability to access and utilize such information. Notably, the European Union’s General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA) have imposed significant limitations on data collection and usage. In parallel, private entities such as Apple and Google have announced or enacted additional restrictions on third-party cookies, further fragmenting the data landscape. These developments have limited firms’ capacity for targeting, personalization, and attribution by making individual-level data more sparse and inaccessible. In this paper, we propose a generative AI-based solution that constructs synthetic customer journeys by integrating aggregate-level marketing mix data. This approach enables firms to simulate and evaluate marketing strategies without relying on personally identifiable information. By substituting real individual-level data with synthetic counterparts, our method addresses the challenge of data fragmentation while ensuring firms’ compliance with prevailing privacy regulations.

Abstract: Online shopping platforms like Amazon often feature a vast assortment of products within a single category. While prior research has primarily examined product assortments through tangible attributes, this paper introduces an attribute-free approach that leverages unstructured product image data for assortment analysis. We propose using the importance-weighted variational lower bound (IWLB) as an approximation of the marginal likelihood for each product image. Our results show that the IWLB effectively captures a product's visual distinctiveness within an assortment: a lower IWLB score indicates a lower likelihood and, consequently, greater visual distinctiveness. Building on this, we derive an entropy-based measure of assortment variety from the IWLB scores. We hypothesize that higher entropy in an assortment correlates with increased consumer-perceived variety and potential choice overload. This paper is the first to examine product assortments through the lens of image distribution and its implications for consumer behavior. Our findings offer actionable insights for retailers managing large assortments, particularly in categories where tangible product features are difficult to extract.