Key takeaways
- Name an owner for each important dataset before writing any policy.
- Keep the first policy to a page: definitions, access, quality and retention.
- Put controls inside the tools people already use.
- Measure adoption by behavior, not by documents produced.
Many companies launch a governance initiative with a committee, a large policy document and a tool purchase, and six months later nothing has changed in how people work. The document sits in a shared folder while analysts keep exporting spreadsheets. Governance that people follow starts smaller and closer to daily work, and it treats the people who use data as the audience rather than the obstacle.
Start with ownership, not policy
The most useful governance artifact is a short list: the important datasets and the named person accountable for each. Accountable means that person answers questions about definitions, approves access requests and is contacted when quality slips. Choose owners from the business side where possible, since they understand what the data means, and support them with technical stewards who understand how it flows. A dataset with no owner is a dataset nobody will fix.
Write the smallest policy that works
A first policy can fit on one page. Cover only what people actually run into.
- Definitions: where the official meaning of key metrics lives, and how to propose a change.
- Access: who can see what, how to request more, and how requests are reviewed.
- Quality: what checks exist and what happens when one fails.
- Sensitive data: how personal or confidential fields are identified and handled.
- Retention: how long data is kept and who decides when it goes.
Use plain language and examples. If a rule cannot be explained to a new analyst in two minutes, it will not survive contact with a deadline.
Put the rules inside the tools
People follow the path of least resistance. If following the rule means opening a ticket and waiting a week, they will find a workaround. Where possible, build controls into the tools people already use: role-based access in the warehouse or lakehouse, a shared catalog where definitions and owners are visible next to the data, and automated tests that flag problems before a dashboard shows them. The compliant route should also be the easy route.
Build habits, not just structures
Governance is maintained through small recurring behaviors. A short monthly review of open quality issues and access requests keeps it alive. Adding a data-owner question to project kickoffs catches problems early. When a report is wrong, a brief blameless write-up of the cause teaches more than a reprimand. Recognize teams that fix definitions and document changes, so the work is visible and valued.
Measure behavior
Counting documents produced says little. Better signals are whether new datasets arrive with an owner, how quickly access requests are resolved, how often two reports disagree on a shared metric, and whether people cite the catalog rather than asking around. Pick a few and watch the direction over time; they show whether governance is becoming part of how work gets done.
A practical sequence
- List the ten or so datasets that drive the most decisions.
- Assign an owner and a steward for each.
- Publish the one-page policy and the agreed metric definitions.
- Turn on access controls and basic quality tests for those datasets.
- Review monthly, then extend to the next tier of data.
