A Predictive Information System: Harnessing Purchase Intent Signals Buried in Digital Data

Abstract

E-commerce, whereby consumers are able to perform transactions easily online, has undergone rapid growth since its inception. Data mining with AI techniques, such as machine learning, can be a useful tool for market research in this field because it provides valuable information for the design of effective digital marketing strategies. This study examines how predictive information systems based on machine learning can improve the efficiency and competitiveness of e-commerce marketing. Using data from 5,389,731 users in Spain, we implement XGBoost to estimate purchase willingness across seven categories (insurance, hearing aids, NGOs, energy, gambling, telecommunications and finance). The supervised model is trained on historical conversions and behavioural–demographic variables, validated with holdout data, and deployed to score non-converted users. To complement algorithmic profiling with managerial insight, we conducted 63 qualitative interviews to uncover motivations by segment and category. XGBoost accurately predicts purchase likelihood when there is sufficient conversion depth, enabling granular segmentation and targeted promotion with reduced waste. The qualitative findings reveal distinct intrinsic and extrinsic drivers by product and declared sex, informing message framing and offer design. Taken together, the mixed-methods approach supports more efficient resource allocation, higher conversion rates and strengthened competitive positioning through data-driven segmentation and motivation-based personalisation. The contribution is twofold: a replicable, at-scale lead-scoring pipeline and actionable motivational maps for planning. Accordingly, this is an innovative, AI-enabled management method that increases efficiency, improves competitiveness and supports advantages in key processes, with multinational portability.