What is a Document Automation?

The generation side assembles a document from structured data and a template — a contract populated with the right party names and terms, an invoice built from a line-item list — ensuring consistency and eliminating copy-paste errors. The extraction side runs the opposite direction: pulling structured fields out of unstructured documents like scanned invoices or PDFs, traditionally with OCR plus rules, increasingly with LLMs that can handle messier layouts and infer fields that don't match a fixed template. Accuracy is the whole game here, especially for anything financial or legal — an automation that silently mis-extracts a dollar amount or a contract term is worse than no automation, because it creates false confidence. Mature systems flag low-confidence extractions for human review rather than passing them straight through. Document automation increasingly chains generation and extraction together: an agent extracts terms from an incoming vendor contract, checks them against a policy, and generates a redlined response, closing a loop that used to require a human at every step.