Untangling budget allocation in E-commerce search advertising

Abstract

For retailers on e-commerce platforms, search advertising serves as a core channel for acquiring traffic and boosting sales. Budget allocation, as one of the fundamental decisions in search advertising, directly relates to retailers' advertising costs and effectiveness, rendering it critically important. However, facing increasingly intense market competition and limited marketing budgets, optimizing budget allocation has become a strategic challenge that retailers need to address. Against this backdrop, this thesis takes "budget allocation" as its central theme and conducts an in-depth investigation into the budget allocation problem for retailers on e-commerce platforms from three complementary dimensions: micro-foundation, intertemporal dimension, and inter-product dimension.
The first study addresses the micro-foundation of budget allocation by directly challenging the fundamental yet contested assumption that "higher bidding price leads to better performance" in search advertising. Employing a two-way fixed-effects model and innovatively leveraging deep learning (ResNet-50) for visual analysis to quantify a product's "relative competitiveness" in terms of word-of-mouth and price, this study rigorously examines the true relationship between bidding price and advertising performance (i.e., Click-Through Rate and Conversion Rate). The findings reveal a significant inverted U-shaped relationship, rather than a monotonic increase. Crucially, the study uncovers that product competitiveness plays a key "flattening" moderating role: the advertising performance of more competitive products is less sensitive to changes in bidding price. This discovery not only subverts the theoretical cornerstone of numerous bid optimization models but also provides an optimal "pricing" basis for the effective spending of a "unit budget," establishing the logical starting point for all budget allocation strategies.
The second study explores intertemporal budget allocation, filling a gap in the literature by investigating the spillover effects of advertising from a retailer's decision-making perspective. Through a dynamic panel data model, this study is the first to systematically reveal a "temporal paradox" in the spillover effect of advertising investment on organic traffic: current advertising spend significantly boosts current organic traffic but markedly suppresses it in the subsequent period. Based on this key finding, the study develops an advertising budget optimization model formulated as a Markov Decision Process (MDP) that integrates this dynamic spillover effect and inventory constraints, which is then solved using a Deep Q-Network (DQN) algorithm. Simulation results demonstrate that, compared to traditional models that ignore spillover effects, the proposed integrated optimization model can increase total profit by up to 18.98%, clearly quantifying the substantial economic value of incorporating traffic source interactions into the strategic decision-making framework.
The third study investigates inter-product budget allocation, deeply integrating advertising and product strategies. It focuses on the strategic regulatory role of advertising in the core e-commerce scenario of "introducing new variants within a product category." This study utilizes computer vision to objectively quantify the "visual similarity" between new and existing variants and examines its impact on the sales of both. The findings show that similarity has an inverted U-shaped effect on the sales of both new and existing variants. Most importantly, the study reveals that advertising investment plays a nuanced "strategic cross-moderating" role. "Offensive" advertising for the new variant effectively mitigates its sales decline at high levels of similarity, but at the cost of exacerbating the sales loss for the existing variant. Conversely, "defensive" advertising for the existing variant, while suppressing the new variant's growth, solidifies the market position of the existing variant itself. This finding provides a clear "tactical playbook" for achieving synergistic optimization of product and budget allocation strategies when introducing new variants within a category.
In summary, through a deep investigation of budget allocation across these three dimensions, this thesis not only deepens the theoretical understanding of budget allocation in e-commerce search advertising but also offers invaluable managerial insights for retailers to optimize advertising budgets, reduce costs, increase efficiency, and enhance overall return on investment.

Date of Award	15 May 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Zhao Cai (Supervisor), Xiangtianrui Kong (Supervisor) & Hing Kai Chan (Supervisor)