Independent reporting on data, AI and digital regulation in Luxembourg

Explainers · Glossary

Data lineage: definition, and why the ECB and the AI Act ask for it

Data lineage is the record of where data comes from and how it is moved and transformed on its way to a report, a model or a decision.

Key takeaways

  • The ECB's Guide on effective risk data aggregation and risk reporting of May 2024 defines data lineage as information about the movement and transformation of data from capture to end.
  • The ECB expects complete and up-to-date lineage at data attribute level for the key risk indicators and their critical data elements, from data capture through extraction, transformation and loading.
  • For high-risk AI systems, Article 10(2) of the AI Act requires data governance practices covering data collection processes, the origin of data and data-preparation operations such as cleaning and aggregation.

Data lineage is the documented path of a piece of data: where it was captured, which systems it passed through and which transformations changed it before it reached a report, a model or a decision. In Luxembourg, banks under direct ECB supervision are expected to keep complete and up-to-date lineage for their key risk indicators, as set out in the ECB guide of May 2024, and providers of high-risk AI systems will need comparable traceability for their training data under the AI Act.

What supervisors expect

The ECB guide on risk data aggregation and risk reporting, which applies the BCBS 239 principles, asks for lineage at data attribute level, starting from data capture and including extraction, transformation and loading. It names six uses: checking that data quality controls sit in the right place, linking definitions and taxonomies, keeping transformed fields in line with reporting definitions, finding data for ad hoc requests, tracing a data quality incident back to its source in time, and allowing external validation.

Under Article 10(2) of the AI Act, training, validation and testing data for high-risk AI systems must be covered by practices on data collection processes, the origin of the data and preparation steps such as labelling, cleaning and aggregation. Following the Digital Omnibus, these obligations apply from 2 December 2027 for Annex III systems.

Why it matters

A risk figure nobody can trace is an opinion. When a number in a board report looks wrong, lineage is what turns a week of email archaeology into an afternoon.

The good news: the ECB does not ask for lineage on every field. Start with the few indicators that matter most.

What to do now

  1. List the five to ten figures your management body or supervisor relies on most.
  2. For each one, name an owner and draw the path from source system to report, including spreadsheets and manual adjustments.
  3. Record that path next to the business definition in your data dictionary, and update it whenever a system changes.

Questions & answers

What is the difference between data lineage and a data dictionary?

A data dictionary says what a data element means. Data lineage shows where its value comes from and which systems and transformations it passed through. The ECB guide expects both, as part of the same data taxonomies.

Does every data field need lineage?

No. The ECB guide limits the expectation to the key risk indicators and critical data elements within the scope of application, and asks for implementation choices that are fit for purpose.

Sources

  1. Guide on effective risk data aggregation and risk reporting (May 2024) · ECB Banking Supervision
  2. Principles for effective risk data aggregation and risk reporting (BCBS 239) · Bank for International Settlements
  3. Regulation (EU) 2024/1689 (AI Act) · EUR-Lex
  4. Article 10: data and data governance · artificialintelligenceact.eu

Written and fact-checked against primary sources.