Let the model read the invoice, not approve it: an n8n pattern for AP automation

This article outlines a robust pattern for automating accounts payable (AP) workflows using n8n, emphasizing a clear separation between AI extraction and business logic. The author argues that while AI models are excellent at parsing PDF invoices into JSON, they should not be entrusted with approval decisions. Instead, the workflow uses a structured approach: an AI model extracts data, while custom code performs rigorous validation checks—such as arithmetic verification, duplicate detection, and bank detail matching. By routing invoices based on these programmatic checks, the system ensures that only valid, low-risk invoices are auto-approved, while others are flagged for human intervention in Slack. This method mitigates risks like business email compromise and ensures auditability. The author provides practical tips on normalization, handling purchase orders, and implementing separation of duties, ultimately offering a template for developers to build safer, more reliable financial automation.