Skavio Flow: prepare supplier invoice PDFs for reviewable extraction
Skavio Flow turns supplier invoices into structured data with custom fields, source evidence and review warnings. But a well-formatted spreadsheet is not proof that an invoice has been read correctly. The useful starting point is a complete, legible PDF—and a clear definition of what a reviewer must confirm before export.
Keep the original and check the whole invoice
Before uploading, preserve the supplier’s original file and inspect a working copy. Adobe’s guidance on recognizing text in scanned documents likewise recommends keeping an original backup before editing. Preparation should preserve the evidence, not merely make the document look tidier. Check completeness before considering extraction fields. A readable first page does not establish that the invoice includes every line item or its final totals. These are preparation checks for the person handling the files, not claims that Flow automatically repairs or reorganizes PDFs.
- Confirm that every expected page is present and that continuation pages appear in the correct order.
- Inspect page edges for cropped invoice numbers, currency labels, table columns and totals.
- If a page is missing or unreadable, request a complete or clearer source rather than reconstructing the missing information.
Distinguish embedded text from scanned pages
Not every PDF needs OCR. Flow reads embedded PDF text locally; scanned documents use Worker Fabric OCR. Structured extraction uses the configured OpenAI model. Local text reading should therefore not be confused with an entirely local processing pipeline. Trying to select text is a useful initial inspection, but check that the selected text matches the visible invoice. Adobe recommends reviewing recognized text for accuracy and completeness: searchable text alone is not evidence of a correct reading. For scans, prioritize legibility over a smaller upload. Google Document AI recommends at least 200 dpi for its service and says 300 dpi or higher generally produces its best results. That is preparation context, not a verified Flow resolution requirement or guarantee. Google also warns that lossy compression can degrade image quality and extraction results. Straighten visibly skewed scans before submission where practical. Tesseract’s documentation explains how substantial skew can impair OCR line segmentation. This is general preparation advice, not a claim that Flow provides a deskew tool. These external references do not imply that Adobe, Google Document AI or Tesseract is integrated into Flow.
Define the fields before extracting them
Flow supports custom extraction fields and saved company workflows. Use that flexibility to define the information needed for the finance handoff, rather than collecting every visible label. A useful starting schema might include supplier name, invoice number, invoice date, due date, currency, subtotal, tax and total amount. Treat line items as a separate structure: description, quantity, unit price and line amount, where those values appear on the source. These are suggested fields, not claimed built-in presets. Keep the distinction between document-level values and repeated rows explicit. Once the field definitions are agreed, a saved company workflow can reuse them for subsequent documents. Reusing a definition does not remove the need to review each invoice.
Review against the source—not just the output
Flow provides source evidence and review warnings alongside structured results. Use them to compare extracted values with the invoice itself. Warnings should guide attention, not be treated as a guarantee that every possible error has been identified. The essential checks remain dates, totals and line items. A plausible value can still belong to the wrong field: an order reference is not necessarily an invoice number, and an invoice date is not a due date. The following are reviewer actions, not claims of automatic reconciliation or accounting validation.
- Check identifiers character by character, including leading zeros and punctuation that matter to the receiving system.
- Confirm dates and currency labels against the source. Leave missing information unresolved rather than silently inventing it.
- Compare line-item sums with the relevant subtotal, then examine tax, discounts and rounding before checking the final amount.
- Investigate discrepancies before accepting the result; a successful extraction is not accounting validation.
Choose the export after the review
Flow’s outputs include XLSX document and line-item worksheets, CSV and JSON. Choose XLSX when reviewers need separate document and line-item worksheets, CSV for a tabular handoff, or JSON for structured software inputs. Before passing the file onward, confirm the destination’s field names, date conventions and treatment of currencies and missing values. An export is a data handoff—not evidence that an invoice has been approved or posted to an accounting system. The practical sequence is simple: preserve the source, define the fields, extract, review, then export.