What gets extracted
- Header: PO number, customer/bill-to, order date, requested delivery date, ship-to.
- Line items: SKU/part number, description, quantity, unit of measure, unit price, line total.
- Notes and special instructions.
- A 0–100% confidence score on every field and every line.
Why LLM extraction wins
Template OCR matches text at fixed positions and breaks the moment a layout changes. OrderPier reads meaning: it understands that 'P.O.', 'Order #', and 'Purchase Order Number' are the same thing, and that a table of parts is a list of line items — regardless of how the page is arranged.
That's why it handles a customer's brand-new format, a faded scan, or a quantity written three different ways, lowering confidence and flagging for review when something is genuinely unclear instead of guessing.
Frequently asked questions
- How accurate is extraction?
- Modern LLM extraction reads complex documents at 95–99% accuracy and beats template OCR by 10–15 points on scans. Confidence scoring ensures uncertain fields are reviewed rather than trusted blindly.