ETSI has published a new technical framework designed to make the quality of data used in digital ecosystems and Artificial Intelligence systems measurable and comparable. The document, ETSI TR 104 180, introduces 18 quantitative metrics through which organisations and developers can assess whether a dataset is sufficiently reliable and fit for the purpose for which it is intended to be used.

The metrics are structured around four main areas. The first concerns the intrinsic quality of data and includes, among other factors, completeness, accuracy, reliability, consistency, precision, integrity, redundancy and uniqueness. The second addresses the actual usability of data through parameters such as availability, coverage, provenance, traceability and timeliness. A third area focuses on fairness and potential bias, through indicators relating to labelling quality, measurement bias and representation bias. The framework also includes parameters concerning privacy and the responsible use of data, such as anonymity and confidentiality.

The practical applicability of the framework was tested through a proof of concept carried out on two very different categories of data: industrial Internet of Things sensor data and demographic data. The exercise also demonstrated that not all metrics carry the same weight in every context: reliability, timeliness and integrity may be particularly important for industrial data, whereas representativeness, anonymity, labelling quality and potential bias become especially relevant where datasets describe individuals or social characteristics.

As part of the proof of concept, an open-source Data Quality Validation System was also developed, capable of applying the 18 metrics and generating a score reflecting the quality of a dataset. The model is also intended to enable dataset holders to perform internal assessments and communicate the quality level of the data made available to other operators in a more transparent manner.

The Technical Report is also particularly relevant from the perspective of the AI Act. Article 10 of Regulation (EU) 2024/1689 requires, for the high-risk AI systems falling within its scope, specific data governance practices and provides that the relevant datasets must be sufficiently representative and, to the greatest extent possible, free of errors and complete. It also requires appropriate measures to identify, prevent and mitigate possible biases.

The ETSI Technical Report does not, in itself, constitute a harmonised standard giving rise to a presumption of conformity with the AI Act. It nevertheless represents a particularly significant technical reference because it translates general concepts such as accuracy, completeness, representativeness, traceability and bias into parameters capable of quantitative measurement.

For providers and organisations developing or using AI systems, the new framework may therefore constitute an operational tool for structuring data governance processes, defining documented criteria for dataset acceptance and generating verifiable evidence regarding the quality of data used throughout the system lifecycle. From a compliance perspective, the ability to move from merely descriptive assessments of data quality to documented and repeatable metrics is becoming increasingly important, particularly for AI systems subject to the requirements of the AI Act.