Purpose This study aims to address the lack of transparency in AI-powered digital marketing tools by integrating Explainable AI (XAI) and Causal AI. It explores how these technologies can enhance understanding of customer conversion dynamics and support more informed, strategic marketing decisions. Design/methodology/approach Using a dataset of 8,000 customer interaction records, the research employs XGBoost for conversion prediction, SHAP for model interpretability, and DoWhy for causal inference. This combination allows for both accurate forecasting and insights into cause-effect relationships driving customer behavior in digital marketing contexts. Findings The study identifies key predictors of customer conversion, including TimeOnSite and ClickThroughRate, and reveals direct causal relationships, such as the influence of AdSpend and ClickThroughRate on conversion likelihood. This dual analysis enables both predictive accuracy and a clearer understanding of underlying marketing drivers. Research limitations/implications This research is based on a static dataset, limiting its scope for dynamic behavior modeling over time. Future studies could incorporate time-series data to analyze customer journey evolution and assess long-term causal impacts. Practical implications Marketers can use these insights to optimize campaign strategies, such as reallocating ad budgets and targeting high-potential customer segments. The framework provides actionable guidance for improving conversion rates through transparent, data-informed decisions. Social implications By enhancing transparency and accountability in AI applications, the study supports ethical AI practices in marketing. It promotes trust and fairness in automated decision-making that affects consumer experiences. Originality/value This work bridges predictive and causal AI methods to offer a novel, interpretable framework for digital marketing optimization. It contributes to both academic research and industry practice by demonstrating how XAI and Causal AI can jointly support transparent, effective campaign design.
Nguyen et al. (Fri,) studied this question.