Data governance for reliable AI

AI increases the number of ways data can influence decisions. Data governance must therefore connect policy with the concrete use case, model, retrieval process, output, and action.

Create a use-case data contract

List approved sources, owners, purpose, legal or contractual basis, sensitivity, access, retention, refresh, and prohibited use. Include external content and generated data, not only internal databases.

Define quality in business terms

Completeness or freshness matters only relative to the decision. Establish thresholds for accuracy, timeliness, representativeness, consistency, and traceability based on potential impact.

Monitor the data lifecycle

Track lineage, access, source changes, failed updates, retrieval quality, and downstream corrections. Give data owners a clear route to stop or restrict use when conditions change.

• Use least-privilege access.
• Separate production and evaluation data.
• Document transformations and derived fields.
• Test retrieval and source attribution.
• Apply deletion and retention consistently.