When narrative disclosures become data
This week, XBRL International’s Revathy Ramanan kicks off a new blog series exploring how Large Language Models (LLMs) can be used to analyse narrative disclosures in structured reports using XBRL data. In this first instalment, she focuses on audit opinion data from Ukrainian Inline XBRL annual reports for 2022, uncovering how structured narrative data can be mined quickly and meaningfully using LLMs.
Narrative disclosures (especially in ESG and audit contexts) are rich but challenging to analyse at scale. Structured data tagging enables targeted, accurate extraction of specific sections, such as audit firm IDs or climate risk commentary, allowing LLMs to surface trends, gaps, and outliers. In contrast, unstructured PDFs make such analysis far more difficult and error-prone.
This blog series demonstrates the way that regulators, analysts, and investors can detect anomalies, benchmark disclosures, and monitor evolving risks with much greater ease while supporting quality, transparency, and data-driven oversight.
Read the post here.
