# Essential data controls for AI > We’ll pilot the AI first, then fix governance once we prove the value. If you’re about to scale AI and these controls aren’t in place, your AI pilot is likely going to fail without governance. Instead, take 4-6 weeks of calendar time with a part-time working group to go through six steps. ## Step 1: Identify the Critical Metrics List every metric that feeds your AI model. Not every metric in your organization. Just the ones the model depends on. Usually 10-20 metrics for a single model. ## Step 2: Assign Named Owners For each metric, assign one person who is accountable for its accuracy, definition, and business meaning. Not a team. A person. Document it in a system of record (not a spreadsheet). ## Step 3: Document the Logic For each metric, write down: • What it measures • How it’s calculated • Where the data comes from • When it was last changed and why Version control this. Treat it like code. ## Step 4: Define the Change Process Create a simple approval workflow: • Who can request a change? • Who approves it? • What’s the impact assessment process? • How is the change communicated? This doesn’t need to be complex. It needs to be clear. ## Step 5: Clarify Access Map out who has read, write, and change permissions for each metric. Remove access that doesn’t need to exist. Log all changes. ## Step 6: Set the Escalation Path Define who breaks ties when metric owners disagree. Name the person. Set the timeline. Document it. Source: [John Wernfeldt on LinkedIn](https://www.linkedin.com/in/john-wernfeldt-82894b58/) — [Data Governance Field Library on Substack](https://datagovernancefieldlibrary.substack.com/)