An accounts payable team may receive machine readable invoice data, a digital PDF, or a photographed bill. The intake decision is which source should populate the fields before review and approval. Invoice formats and legal obligations differ across the US, Canada and the UK, and can vary by transaction. This guide does not determine tax validity; follow the requirements for your jurisdiction and buyer.
Short answer
If an authorized, verifiable structured record is available, use it as the source of fields and retain its provenance. Use OCR or document extraction when the information is only in a scan, image or visual PDF, or to help a person compare a readable copy. Extracted fields are candidates, not proof that an invoice is genuine, correct or approved. Compare key fields with the original document and the order or receipt before moving on.
Pulse’s intake decision tree
| Input received | Source for the draft | Review before release | Stop when |
|---|---|---|---|
| Authorized structured invoice data | Import fields and retain the source record and identifier | Supplier, buyer, invoice number, dates, amounts, line items and order reference | The record does not match the invoice or a key amount differs |
| Visual PDF, scan or image only | Extract a draft with OCR or a document model | Check critical fields against the image and review totals and applicable tax | A critical field is missing, unreadable or contradictory |
| Structured record plus readable PDF | Use the authorized data for fields; use the PDF as a human aid | Confirm that identifiers and values agree | Versions disagree on identity or value |
| Incomplete or untrusted input | Do not populate automatically | Request a verifiable source and complete document | Evidence is insufficient to decide |
This is an original editorial decision model, not a universal compliance rule. Its purpose is to keep a field read by OCR from silently becoming a claim of validity.
What extraction actually returns
Microsoft’s invoice model analyzes images, scans and PDFs, returning fields and line items in a structured JSON representation. AWS AnalyzeExpense groups summary fields and line items. These capabilities can reduce transcription, but results depend on the source image, layout and mapping. The vendor documentation describes extraction, not an accounts payable approval on your behalf.
Before choosing a tool, inventory the actual input: what the supplier sends, which file holds authoritative data for this process, how origin is checked, which fields your team needs, and who resolves exceptions. A searchable PDF may still be a visual representation rather than a structured invoice record. The European Commission’s structured eInvoicing distinction applies in its EU context; it does not set the rules for a US, Canadian or UK transaction.
A worked decision with fictional inputs
Hypothetical example: a small business receives one invoice as structured data through its agreed channel and another as a scanned PDF. For the first, import the fields and compare the identifier and total with the purchase order. For the second, extraction proposes supplier, date, number, line items and amount; a person checks the critical fields on the scan. If the amount differs, hold the invoice for resolution. Pulse has not tested the vendors in this example.
Keep an audit trail of source → file → fields used → discrepancy → decision → owner. Restrict access to invoices, and obtain approval before sending business documents to an external service. After intake, use the invoice workflow states, the release-or-review decision for extracted fields, and three-way matching when an order and receipt exist.