Detecting and reporting suspicious activities - AI-assisted compliance management for social e-commerce platforms

Abstract The rapid growth of social e-commerce has transformed retail by enabling transactions to be initiated and negotiated directly through social media platforms. While this model expands market reach, it also intensifies compliance management challenges for platform firms, particularly the need to identify and report suspicious selling activities at scale. Manual monitoring and investigation are increasingly unable to keep pace with the volume, diversity, and multi-modal nature of platform content, creating a critical operational bottleneck in compliance reporting. This study conceptualises and evaluates a prototype of an AI-assisted compliance management system for social e-commerce platforms, guided by the Theory of Constraints. Using an exploratory, case-study approach, we analyse publicly available Instagram posts and identify common textual and visual characteristics associated with suspicious and reportable selling activity. Building on these insights, we develop a proof-of-concept multi-modal deep neural network that jointly learns from image and text features to support automated triage of potential compliance risks. Empirical evaluation shows that the proposed multi-modal model outperforms single-modality baselines, achieving higher overall detection performance (AUC = 0.837; accuracy = 84.1%). The findings demonstrate how AI can elevate a key compliance management constraint by filtering and prioritising high-risk content for human review, while preserving managerial judgment for interpretive and reporting decisions. The study contributes to research on platform governance and compliance management by framing compliance reporting as a constrained operational workflow and illustrating how AI-assisted, human-in-the-loop systems can enhance scalability, accountability, and regulatory responsiveness in social e-commerce ecosystems.