Review Center: Why AI Drafts Need Human Approval Before They Become Official
Karu's Review Center is where extracted invoices, recipe links, duplicate merges, and unclear units wait for a human yes before they touch live margin data.
AI can read messy files faster than a tired owner at closing time. It should not get to write business truth without a checkpoint. That checkpoint is the Review Center: drafts in, approved records out.
Drafts protect the business from silent errors
When extraction creates a draft, the owner can inspect supplier, unit, pack size, price, and confidence before anything flows into recipes or margins.
That is the Karu pattern: AI reduces effort, humans approve business-critical truth.
Review is not just for invoices
The same queue handles product-recipe links, duplicate document detection, unclear units, and supplier price updates that need a second look.
One approval surface beats hunting through tabs, spreadsheets, and chat threads.
Approval creates auditability
When a price or recipe structure becomes official, the business should know what changed and why. Review states make that visible.
Staff still see only published recipe versions. Draft work stays on the manager side.
Operator checklist
Treat every AI extraction as a draft until approved.
Review duplicate documents before creating official margin data.
Resolve unclear units and recipe links in the same queue.
Publish recipe versions to staff only after approval.
Sources
Related guides
Operations
The Karu Inbox: One Intake Flow for Invoices, Recipes, Menus, and Sales
How a single document inbox turns PDFs, photos, spreadsheets, and pasted text into structured drafts ready for review.
AI
Invoice Extraction with AI: Turning Messy Supplier Documents into Usable Costs
How multimodal AI can reduce the pain of invoice data entry while keeping humans in control.
AI
AI Recipe Costing Needs Confidence Badges
Why estimated prices and AI-extracted data should be clearly marked without scaring the user.