Abstract

National statistical systems play a critical role in developing the data that shape policies, monitor economic trends, and support financial stability. Despite the significant investments in developing these statistics and ensuring their quality across multiple dimensions—such as methodological soundness, accuracy, and reliability—many statistical systems underperform in two key areas: accessibility and serviceability. Accessibility pertains to the ease with which users can obtain data across different platforms and technologies, while serviceability refers to the ability to interpret data through comprehensive metadata. This paper investigates how large language models, drastically improve the accessibility and serviceability of global official statistics. By harnessing these advanced models, the global set of official statistics could become more discoverable and authoritative, addressing issues of misinformation. LLMs have the potential to guide users to official, traceable economic and financial statistics, enhancing the overall quality and utility of official data for policy development.