I built an invoice-intake workflow in n8n that goes beyond simple AI extraction.
It takes a PDF invoice and extracts vendor, invoice number, dates, totals, PO number, payment terms, and line items, validates required fields, checks arithmetic consistency, routes questionable invoices into a separate Review queue, stores clean invoices separately, blocks duplicate inserts using vendor + invoice number
The design principle is simple: AI extracts, deterministic logic validates.
I packaged the workflow with setup instructions, test invoices, troubleshooting, and internal n8n Data Table storage.
I built it because I wanted something that would not silently trust every AI extraction result.
Product page: Invoice Intelligence for n8n
Happy to answer questions about the workflow structure or validation logic.