# Workflows — Data Labelling System ## Workflow 1: Label Configuration and Deployment ``` START: New label required or existing label modification │ ├─► Step 1: Label Design │ - Define label name, description, scope, colour, visual marking │ - Define encryption settings (Azure RMS configuration) │ - Define DLP policy association │ - Define auto-labelling conditions (SITs, confidence threshold) │ Output: Label specification document │ ├─► Step 2: Approval │ - DPO reviews label alignment with classification policy │ - CISO reviews encryption and DLP settings │ - Privacy Steering Committee approves │ Output: Approved label specification │ ├─► Step 3: Configuration │ - Create/modify label in Microsoft Purview compliance centre │ - Configure auto-labelling policies │ - Configure DLP policies linked to the label │ - Configure endpoint DLP for the label │ Output: Label configured in Purview │ ├─► Step 4: Testing │ - Test in pilot group (Privacy Engineering team) │ - Verify auto-labelling accuracy on sample documents │ - Verify DLP policy enforcement │ - Verify label inheritance rules │ Output: Test results │ └─► Step 5: Deployment - Publish label policy to all users (phased if major change) - Communicate change to all employees - Monitor adoption metrics for 30 days Output: Label deployed and monitored ``` ## Workflow 2: Label Override and Downgrade Review ``` START: User requests label downgrade or removal │ ├─► Step 1: Justification Capture │ - User enters justification text in label prompt │ - System logs: user, document, current label, new label, justification, timestamp │ ├─► Step 2: Automated Review │ - For Confidential → Internal or Public: flag for Data Owner review │ - For Restricted → any lower: flag for DPO immediate review │ - For auto-applied label removal: flag for Privacy Engineering review │ ├─► Step 3: Review Decision │ - Reviewer assesses: Is the downgrade justified? │ - Is the content actually at a lower sensitivity than the label indicates? │ - Was the auto-label a false positive? │ - Decision: approve, reject, or investigate │ └─► Step 4: Action - Approve: downgrade confirmed; update audit record - Reject: original label reinstated; user notified - Investigate: possible misclassification; escalate to full review ```