The Data Governance Confession Wall

Reference data needs a named owner and a freshness rule. Record how often each reference table should change, alert when it misses that cycle, and make ownership part of the handover whenever someone changes roles.

Write down the definition of an employee in the business glossary, including the boundary cases (contractors, leave, part-time), and name one system as the authoritative source for headcount reporting.

Agree data contracts between producers and consumers, announce schema changes before they ship, and design pipelines to fail loudly when an expected field disappears rather than carrying on with empty values.

Treat consent as a critical data element with a named owner: define it once, flag it in the catalog, and require campaign lists to be built from a certified view rather than from a spreadsheet export.

Classify data before it becomes a source, not after. An AI assistant inherits every permission and every mislabelled file in the corpus you point it at, so classification and access review belong before the index is built.

Define critical metrics once in a governed business glossary, assign an accountable owner, and require executive reporting to use the same certified source.

Separate credentials and network access by environment, require destructive-query safeguards, and test recovery procedures before an emergency exposes the gaps.

Classify sensitive data at creation, apply least-privilege access automatically, and audit shared repositories continuously instead of relying on manual cleanup.