Skip to main content

Approvals

The approval queue is the governance gate for Agentic PIM. Specialist agents propose catalog changes; humans approve, reject, or refine before any write goes live.

This is the primary “humans approve every write” mechanism on the platform — distinct from (and complementary to) PIM workflow tasks for legal or brand review.

What lands in the queue

Agent runs emit approval proposals when they recommend catalog changes, including:

Approval typeExample
attribute_updateProposed SEO title or bullet change
category_changeTaxonomy agent suggests new category
content_generationGenerated description awaiting review
translationLocale copy proposal
bulk_actionBatch attribute updates across a scope
status_changeLifecycle transition recommendation

Each proposal includes a proposed diff — what would change, on which product or scope, and which agent produced it.

Review workflow

  1. Open Agent Control Center → Approvals
  2. Filter by agent, status, or product
  3. Open a proposal to inspect the diff
  4. Approve — applies the catalog write through governed MCP/API paths
  5. Reject — discards the proposal; no write occurs
  6. Retry apply — if apply failed after approval, retry from the detail view

Bulk approve and bulk reject are supported for high-volume review sessions.

Approvals vs PIM workflows

Agent approval queuePIM workflows & tasks
SourceSpecialist agent proposalsHuman-defined workflow stages
Typical useAI-suggested attribute/content fixesLegal, brand, channel sign-off
Write timingOn approveOn task completion
DocsThis pageWorkflows & approvals

Both can apply to the same product — agent proposals for AI fixes, workflows for organizational gates.

Auto-apply (limited)

Some agent types (e.g. duplicate detection) may be configured for auto-apply when tenant policy allows. Default posture is review first. Configure under Agents → Configuration.

MCP and approvals

External MCP clients can propose changes via tools, but in-platform agent approvals provide the structured queue, diff view, bulk actions, and audit trail for specialist agent output. Sensitive operations should still pass workflow gates where required.