Abstract

National Statistical Offices are increasingly tasked with supporting policymakers
who require not merely access to official statistics, but the capacity to derive
cross-domain intelligence from complex and interconnected economic and social
data. Traditional delivery models such as static tables, dashboards, and reports
impose significant cognitive and technical demands on users, thereby limiting the
effective application of statistics in high-stakes decision-making processes. This paper
proposes an artificial intelligence (AI)-enabled decision-support framework wherein
large language models (LLMs) function as an adaptive, natural-language interface
between official statistics and policy users. This approach is grounded in the Bayaan AI
platform, a pioneering initiative developed at the Statistics Centre – Abu Dhabi (SCAD),
which facilitates natural-language interaction with curated indicators, datasets, and
metadata across diverse domains including prices, labour, population, and economy.
The proposed framework is designed around the core principles of confidentiality,
explainability, and institutional trust. To uphold data-protection and confidentiality
mandates, the AI models operate exclusively over approved, access-controlled
statistical layers, rather than raw microdata. Explainability is systematically embedded
by linking every AI-generated insight to its originating tables, classifications, and
methodological documentation. This aligns with the principles of eXplainable AI (XAI),
ensuring that users can trace outputs back to authoritative sources and established
statistical processes. This transparent approach enables policy-makers to explore
complex relationships, compare trends, and generate coherent cross-domain
narratives while preserving the integrity, reproducibility, and credibility of official
statistics.
By leveraging natural language interfaces, the framework empowers users to query
vast statistical repositories and generate visualizations through intuitive dialogue,
lowering the technical barrier to data-driven discovery. The paper demonstrates
how AI can be integrated into national statistical systems not as a disruptive force, but
as a governed and transparent interface that fosters evidence-based decision-making
and reinforces democratic values. This contributes to the discourse on
modernizing statistical delivery and access, offering a model for NSOs to enhance the
utility and impact of their assets in an era of digital transformation.