Data lineage

The history of processing of a data element, which may include point-to-point data flows and the data actions performed upon the data element.

Source: AIEOG AI Lexicon (Feb 2026), NIST CSWP 01162020

Data lineage is the documented history of where data came from, how it moved through your systems, and what happened to it along the way. It traces the path of each data element from its original source through every transformation, aggregation, and storage step until it reaches its final destination, which in the context of AI might be a model’s training dataset or a real-time input feed.

Think of data lineage as a chain of custody for information. Just as a chain of custody documents who handled physical evidence and when, data lineage documents which systems handled data and what operations were performed on it.

For AI systems, data lineage is critical because the quality and integrity of model outputs depend directly on the quality and integrity of the data flowing into the model. If you cannot trace where your training data came from or how your input data was transformed, you cannot fully assess or validate the model’s behavior.

Data lineage is a foundational requirement for AI governance, model risk management, and regulatory compliance in financial services.

• Model validation. Validators need to understand where training data came from and how it was processed to assess whether the model was built on appropriate, representative data.
• Audit and examination. Examiners expect institutions to demonstrate that they understand the data flowing into their models and can trace it back to authoritative sources.
• Incident investigation. When a model produces unexpected results, data lineage allows investigators to trace the issue back to its source, whether that is a data quality problem, a transformation error, or a source system change.
• Regulatory reporting. Accurate regulatory reports depend on accurate data. Lineage documentation provides the traceability needed to defend report accuracy.
• Data privacy compliance. Understanding where data flows helps institutions comply with privacy regulations by identifying where personal information is used and how it is protected.

1. Document lineage for all AI training data. Maintain records of where training data originated, how it was selected, cleaned, transformed, and prepared for model development.
2. Map real-time data flows. For AI systems that process data in real time, document the data pipeline from source systems through the model and into downstream processes.
3. Establish data quality checkpoints. Place quality validation steps along the data lineage path to catch errors before they reach the model.
4. Maintain lineage documentation as a living artifact. Update lineage documentation when source systems change, data pipelines are modified, or new data sources are added.
5. Include lineage in vendor assessments. For third-party data and models, understand and document the lineage of data used by the vendor.
6. Use lineage for impact analysis. When upstream data sources change, use lineage documentation to assess which models and systems are affected.

Data quality/validity, Training data, Documentation, Structured data, Unstructured data, AI lifecycle