The AMF shares the lessons learned from its latest experiments with automated processing of regulatory data

This new phase of research, conducted in 2023 and 2024 using its Big Data platform ICY, aimed to assess regulators' ability to leverage information extraction technologies to process and analyze the information published by listed companies and regulated entities more effectively. Due to European regulatory developments, the number of documents under the AMF's supervision has increased significantly in recent years, creating additional pressure that these technologies can help alleviate.

Taxonomy and SFDR reporting experiments

The experiments conducted by the AMF and presented in this study focused on two types of documents required by European sustainable finance regulations:

• the first Taxonomy reports, published in 2022, which detailed the eligibility of economic activities based on the sustainability criteria defined by the European Union in its green taxonomy.
• the SFDR annexes, introduced in 2023, which specify the environmental and social characteristics of financial products.

These documents, illustrative of regulatory reporting due to the variety of their formats and structures, serve as particularly representative case studies of the challenges related to the automation of regulatory data processing. The insights gained from this analysis can be applied more broadly to other financial reports.

Some key lessons for the development of regulatory standards

The AMF's work has identified several practices that could enhance the usability of regulatory documents and should be considered to guide future developments in standards:

• it is preferable to systematically associate images with descriptive text or tables. Any image containing data intended for machine processing should be accompanied by a text or descriptive table presenting this information in a usable form. This complementarity would help prevent information loss and enhance the ability of systems to automatically extract relevant data.
• it is easier to process a document with a standardized structure. Harmonizing the structure of documents (in terms of layout) and information fields is essential to facilitate their automatic processing. Uniform templates would enhance data comparability and optimize processing by regulators and market participants.
• to ensure optimal data usability, the requirement for a "machine-readable format" often used by legislators must be accompanied by rules. The AMF has observed that using a machine-readable format alone is not sufficient. Indeed, the lack of strict structuring can limit the effectiveness of automated processing, even when documents are published in formats that are theoretically suitable for machine-readability.
• strengthening the technical standardization of documents. Regardless of the format defined by regulations, it is essential to establish precise methods for entering information in these documents (e.g., tags to distinguish titles, specific characters used to check a box). More detailed requirements within the technical standards applicable to regulatory reporting would enhance their readability and interoperability. Moving towards more strictly regulated formats would help reduce discrepancies and difficulties in data extraction.

In this regard, additional research conducted by the AMF, particularly through the analysis of practices in the United States—both in reporting and more broadly on websites in general—provides valuable insights. The widespread adoption of the XHTML format, in compliance with World Wide Web Consortium (W3C) standards, would ensure a more efficient use of regulatory documents. This approach could be implemented even for future regulations that do not require the integration of iXBRL tags, to enhance the readability and accessibility of data.

These insights contribute to the ongoing reflection on the evolution of norms and standards governing regulatory reporting in Europe. The AMF will continue its work to support this dynamic and contribute to discussions on future supervisory frameworks, with a focus on the efficiency of automated processing and the quality of data accessible to regulators.