Copilot turns hour-long governance chores into a single prompt. This article covers three common bulk actions; flagging PII, applying naming conventions, and generating descriptions across hundreds of objects; and shows how to review and approve the proposed changes before they land in the model.
Who this is for: Data stewards, governance leads, and architects onboarding an existing model into SqlDBM where descriptions, tags, or naming need to be filled in across many objects at once.
How bulk actions work in Copilot
When you ask Copilot to change many objects at once, it does not apply the changes directly. Instead, Copilot:
- Scans the project to find all objects that match your prompt.
- Generates the proposed change for each matched object.
- Shows you a review screen with every proposed change listed side by side with the current state.
- Waits for you to accept or reject the whole batch (or individual changes) before anything lands in the model.
This proposal-and-approve pattern is the same for every bulk action. You never commit changes you haven’t reviewed.
Use case 1: flagging PII across the project
PII identification is the most common bulk ask. Copilot reads field meaning semantically, so it catches PII under different names:
Prompt: “Identify all columns in this project that contain personally identifiable information and apply the PII flag.”
Copilot will surface customer_name, user_email, ssn, person_id, phone_number, home_address, and any variation on those, including columns where the naming is non-obvious (e.g., contact_details_1). Review each proposal before accepting; you know your data better than the AI does, and some flags will be context-dependent.
Use case 2: applying a naming convention retroactively
If a project was imported or built before naming conventions were defined, Copilot can rename objects to match a standard in a single prompt:
Prompt: “Rename every fact table to begin with FCT_ and every dimension to begin with DIM_, following our naming convention.”
Copilot proposes a rename for each matching table. The review screen shows the current name alongside the proposed name so you can spot any edge cases before approving.
Tip: For lasting effect, configure your naming convention as a project-level setting (see “Settings and Conventions”) before renaming. That way, new tables created after the rename will follow the same convention automatically.
Use case 3: generating descriptions across many objects
Missing descriptions are one of the largest single blockers to making a model usable by downstream consumers. Copilot can draft descriptions based on table and column names, relationships, and any governance context already in the project:
Prompt: “Generate a non-technical description for every table in the orders schema based on its columns and relationships.”
Descriptions are drafts. Review them. The AI will get the general sense right, but business-specific nuance often needs adjusting. Still, drafting from scratch versus editing a draft is a large time saving.
Reviewing and approving bulk changes
Every bulk action opens a review screen before committing. On that screen you can:
- Accept all proposed changes at once.
- Reject all proposed changes and refine your prompt.
- Accept individual changes and reject others. Useful when the AI gets most of a batch right but a few cases are ambiguous.
If you reject the batch, the prompt context is retained so you can refine without starting over. Common refinements: “Exclude tables in the staging schema.” “Don’t rename tables that already follow the convention.” “Skip columns where the PII flag is already set.”
What to avoid
- Running bulk actions on a production-connected project without a sandbox first. If you’re new to bulk changes, test on a copy of the project before running them on a model that’s wired into downstream systems.
- Over-scoping the prompt. “Fix all the descriptions” is too open-ended. “Generate descriptions for every table in the customer schema that doesn’t already have one” gives Copilot a specific target.
- Accepting without reviewing. The review screen exists for a reason. Even for straightforward prompts, at least spot-check a handful of proposed changes before approving.
Related articles
- Getting started with Copilot
- Configuring Copilot settings and pre-prompts
- Settings and Conventions
- Database Documentation