Abstract

Data silos, arising from fragmented and heterogeneous data sources, pose significant challenges to statistical modeling by increasing bias, variance, and uncertainty in estimation and prediction. This paper investigates the role of artificial intelligence (AI)–based data integration techniques in reducing data silos and improving the quality of statistical analysis. We propose an AI-driven framework that leverages machine learning methods for automated data harmonization, entity resolution, and feature representation across disparate datasets. The impact of silo reduction is assessed through its effects on parameter estimation, model stability, and predictive performance in both simulated and real-world data settings. Results demonstrate that integrated data representations obtained via AI lead to statistically significant improvements in model accuracy and robustness compared with siloed analyses. These findings highlight how AI can act as an enabling technology for principled statistical modeling in complex, multi-source data environments.