Data quality
Data quality is the degree to which data is fit for the purpose for which it is used: accurate, with integrity, complete, timely and traceable to its source. For risk reporting, financial reporting and AI systems, data quality determines whether decisions rest on a reliable picture. Data quality is measured with indicators and secured through data governance: ownership, standards, controls and oversight.
Source: BCBS 239 (2013); BCBS, Progress in adopting the Principles for effective risk data aggregation and risk reporting (2023); ECB, Guide on effective risk data aggregation and risk reporting (2024); Regulation (EU) 2024/1689 (AI Act) Article 10; Regulation (EU) 2016/679 (GDPR) Article 5(1)(d); COSO Internal Control 2013 (principle 13); IIA Global Internal Audit Standards 2024 (Standard 14.1); DNB, Good practice on assuring data quality during and after the Wtp transition (2024); DNB, Good practice on decision-making on data quality (2024)
Several frameworks use similar characteristics. For risk data aggregation at banks, BCBS 239 (2013) names accuracy and integrity, completeness, timeliness and adaptability. The ECB guide of May 2024 has the management body set requirements for accuracy, integrity, completeness and timeliness, in normal and stress periods, and monitor data quality with indicators. The EU AI Act requires training, validation and testing data for high-risk systems to be relevant, sufficiently representative and, as far as possible, free of errors and complete in view of the intended purpose (Article 10(3)). The GDPR requires personal data to be accurate and, where necessary, kept up to date (Article 5(1)(d)). In COSO Internal Control 2013, principle 13 concerns the use of relevant, quality information.
In the Netherlands, data quality received much attention at pension funds because of the transition to the Future Pensions Act (Wet toekomst pensioenen). In a good practice (December 2024, updated in July 2025), the Dutch Central Bank (DNB) writes that a fund must be able to demonstrate that data quality is assured before, during and after the transition; a good practice is indicative, not a requirement. At banks, data quality remains a persistent issue: of the 31 global systemically important banks the Basel Committee assessed for 2022, only two were fully compliant with all BCBS 239 principles (report of November 2023).
For the management board and the audit committee, the question is whether the figures they steer on are correct and what limitations they contain. BCBS 239 asks that the board be aware of limitations that prevent full risk data aggregation in the reports it receives (para 31). Under the Global Internal Audit Standards, internal audit assesses whether the information it relies on is relevant, reliable and sufficient (Standard 14.1); at significant banks the ECB also expects periodic reviews of the quality of the data used to quantify risks.
Last updated: · Editor: ONE Risk Advisory
Frequently asked questions
Who is responsible for data quality?
In the first place the owner of the process in which the data originates. At significant banks the ECB places responsibility for data quality and data governance with the management body (2024 guide); the second line validates risk data independently and internal audit reviews the whole periodically. For Dutch pension funds, a DNB good practice (April 2024) describes the decision the board takes on data quality before converting accrued entitlements to the new system.
How is data quality measured?
With indicators for each characteristic of the critical data elements, such as the share of missing or incorrect fields, the number of manual corrections and the time a report takes to produce. The Basel Committee recommends a standard set of key performance indicators with which the board can assess the data quality of all material group-level risks (2023). The ECB asks for data quality indicators to be reported periodically to the management body together with an impact analysis.
What does the AI Act require of data quality?
For high-risk systems that train models with data, training, validation and testing data must be relevant, sufficiently representative and, as far as possible, free of errors and complete in view of the intended purpose (Article 10(3)). This comes with data management practices that address origin, assumptions, gaps and bias (paragraph 2). For the use cases in Annex III, these requirements apply from 2 December 2027.