Data Lineage
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
A record of where data originates, how it moves and transforms through systems, and where it ends up – the audit trail behind datasets and AI models.
What is Data Lineage?
Data lineage is the record of where data comes from, how it moves and transforms as it passes through systems, and where it ends up. For any value in a report or a training set, it answers the question “how did this get here?”
For AI governance the record does double duty. It supports debugging when something looks wrong, and it supplies the evidence trail an AI audit or a regulator asks for.
How Data Lineage Works
Lineage tools capture metadata at each step of a data pipeline: which source a dataset was read from, which transformation produced it, which jobs and reports consume it. Chained together, these records form a graph that can be traced in either direction.
Tracing backward answers “where did this number come from?” Tracing forward answers “if this source changes, what breaks downstream?” Both are routine questions in any organization with more than a handful of pipelines.
Applications of Data Lineage
- Audit trails: showing an auditor or regulator exactly which sources fed a model or a financial report.
- Root-cause analysis: tracing a wrong value in production back to the transformation or source that introduced it.
- Impact analysis: identifying every downstream model and report affected before a data source is changed.
- Regulatory compliance: GDPR requires organizations to know where personal data lives, and the EU AI Act sets data-governance requirements for high-risk AI systems that cover how training data was collected and where it came from.
Example of Data Lineage
An auditor reviewing a bank’s credit-scoring model questions one input feature: “months since last delinquency.” The team traces it backward through the lineage graph – from the model’s training set, through a feature-engineering job, to a nightly export from the loan-servicing database.
The trace shows the export excluded accounts migrated from an acquired bank, so the feature was silently wrong for those customers. Without lineage, finding that would mean weeks of archaeology across teams. With it, the path from suspect feature to root cause is a query.
Related AI terms: Data Versioning · Data Quality · Data Integrity · AI Audit · EU AI Act
Did you like the Data Lineage gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About