Document parsing extracts structured fields and relationships from files such as invoices, quotations, purchase orders and shipping documents. It converts human-readable content into data that a system can validate, route and reconcile.
What can be extracted?
- Document type, number and date
- Buyer, seller and account identifiers
- Line descriptions, quantities and unit prices
- Currency, tax, freight and total amounts
- Payment and delivery terms
- References to orders, shipments and contracts
How does parsing work?
- Ingest the original file and preserve it.
- Detect layout, text and document type.
- Extract candidate fields and line items.
- Normalize values into the target schema.
- Validate totals, formats and cross-document references.
- Route low-confidence results for review.
Parsing vs. OCR
Optical character recognition converts an image of text into machine-readable characters. Parsing identifies the meaning and structure of those characters. Accurate OCR is useful, but it does not by itself determine which number is the invoice total or which date is the due date.
How should accuracy be evaluated?
Measure field-level precision and recall by document type, supplier, language and layout. A single document-level accuracy score can hide errors in high-risk fields such as bank details, currency or totals.
What controls are needed?
Keep the source file, extracted value, confidence, validation result, reviewer correction and model or rule version. Do not allow an uncertain extracted field to trigger payment, approval or compliance action without the required checks.
