Abstract This paper explores how data-centric artificial intelligence governance frameworks enable responsible organizational value creation within complex institutional environments. Using an empirical case from a European public administration, it examines the implementation of an automated legislative monitoring system designed to detect, classify, and summarize regulatory information. The study highlights the shift from model-centric experimentation to a mature data governance and Machine Learning Operations (MLOps) framework, integrating continuous human oversight and ethical accountability. A qualitative case study, DGOBCAN-AI, was employed, combining technical documentation, process observation, and organizational evaluation. The system evolved from a basic extract–transform–load (ETL) script to a robust MLOps ecosystem, incorporating Airflow for workflow orchestration, MLflow for experiment tracking, and Label Studio for human-in-the-loop annotation. This approach ensured reproducibility, iterative refinement, and continuous validation. Findings indicate that governance shortcomings, rather than algorithmic limitations, were the primary source of underperformance. Key challenges included severe data scarcity and extreme class imbalance (0.27% positive cases). Adopting a data-centric governance approach with iterative annotation, reproducible pipelines, and ongoing human supervision improved transparency, accountability, and reliability. Moreover, a hybrid on-premises/cloud architecture demonstrated both technical feasibility and cost efficiency (≈ €6.6/month). Managers adopting AI should prioritize data quality and traceability over model complexity, embed human oversight throughout the AI life cycle, and design cost-effective infrastructures aligned with ethical and regulatory standards. This research offers a rare empirical link between MLOps infrastructure, human validation loops, and data stewardship, demonstrating that ethical responsibility and sustainable innovation reinforce organizational legitimacy and long-term competitive advantage.
Aguilar et al. (2026) studied this question.