AI workflows

Combine human approvals with machine speed. CataZenta workflows route products through stages — enrichment, legal, brand, channel — while AI handles repetitive steps inside each stage.

Pattern

Stage	Human	AI
Intake	Assign family	Suggest attribute mapping from import
Enrichment	Approve copy	Draft descriptions per locale
Media	Pick hero shot	Propose alt text from DAM tags
Channel gate	Sign off publish	Flag completeness blockers
Syndication	Monitor job	Summarize per-SKU failures

Why workflows matter for AI

Without gates, AI output goes straight to Amazon — a compliance risk. With workflows:

• Draft stays draft until approved
• Tasks assign to the right role (brand vs legal vs ops)
• History shows who accepted AI suggestions

→ Workflows guide · MCP: list_workflows, list_tasks, update_task_status

Example workflow

Launch collection to Shopify + Amazon

1. Import ERP rows → extraction proposes attributes
2. Bulk enrichment for en_US and de_DE
3. DAM link + completeness to 100% for each connection
4. Parallel approval tasks (brand + legal)
5. Export jobs per channel
6. MCP agent polls get_export_job_items and opens tasks for failures

→ AI workflow examples