A Purchase Order Is Two Tables: Building Reviewable Worksheets in Skavio Flow
A purchase order looks like one document. Operationally, it is at least two records: a header and a set of repeated line items. Skavio Flow is a live web and API product for turning invoices, orders, scans and text into reviewed XLSX, CSV and JSON using custom fields and saved company workflows. Its useful role in procurement is specific: create a structured draft that a person can inspect, correct and approve before any downstream handoff.
Start with the shape of the record: header plus repeated lines
Purchase-order data becomes difficult to review when the header and the line items are flattened into one block. The header describes the transaction: supplier or customer identity, purchase-order reference, document date, currency and totals. The line items carry the repeatable commercial detail: description, quantity, unit, unit price, line total and any order or delivery reference that the document actually shows. Keeping these as two linked tables makes exceptions easier to see. A reviewer can compare a header total with the sum of the lines, or trace one unusual quantity back to its source page without losing the rest of the record. Flow supports custom extraction fields and saved company workflows, so a team can define consistent field names for these header and line-item concepts and reuse them across new documents. The reusable part is the field structure, not the values: each job still draws its values from the source document.
- Header fields to define: supplier or customer identity, purchase-order reference, supplier reference, document date, currency and header totals.
- Line-item fields to define: description, quantity, unit, unit price, line total and any reference shown on the source document.
- Review fields to preserve: source evidence, warnings, missing values and ambiguous labels.
- Use saved workflows to keep field names stable between batches; do not treat a saved template as pre-filled data.
Separate reading the document from extracting the structure
Flow's processing path separates document reading from structured extraction. Embedded PDF text is read locally, scanned documents use OCR, and structured extraction runs as a separate step. That distinction is useful during review because a digital PDF and a scanned order can fail in different ways: embedded text may be incomplete or badly ordered, while a scan may depend on image quality and clear page boundaries. The outputs are XLSX document and line-item worksheets, CSV, JSON, source evidence and review warnings. Those outputs are designed to support inspection, not replace it. Source evidence helps a reviewer ask where a value came from. Review warnings help identify what still needs attention. Neither proves that every field is correct, and neither turns an extracted draft into an approved procurement record.
- Embedded-text PDFs and scanned PDFs enter the same review workflow, but they should not be reviewed with the same assumptions.
- XLSX is useful for a human review workbook; CSV and JSON are better suited to structured handoffs.
- Source evidence should be kept with the draft so a reviewer can trace a value back to the document.
- Treat review warnings as items to resolve, not as decoration around an already approved row.
Review the exceptions that can change a purchasing decision
A practical review pass should focus on fields that can alter what a colleague does next. Compare supplier or customer identity, purchase-order and supplier references, document dates and currency labels against the source. Check whether the header total is consistent with the visible line totals, and whether each line's quantity, unit, unit price and line total tell a coherent story. Pay particular attention to split tables, changed units, unclear decimal separators and descriptions that look plausible but do not match the original wording. Ambiguous tax labels should remain visible exceptions rather than being turned into a tax conclusion. If a scan is unreadable or a reference is missing, the safe outcome is rejection or escalation, not a silently filled cell. Flow can present evidence and warnings; the approval decision remains human.
- Compare header totals with the sum of line totals before accepting an export.
- Check purchase-order references, supplier references, dates and currency labels against source evidence.
- Review quantities, units and unit prices as a linked set, not as isolated cells.
- Escalate unreadable scans, split tables and ambiguous labels instead of accepting a guessed value.
- Keep any tax-related wording as a visible exception; the worksheet is not tax advice.
Correct the draft, regenerate the export, then make the handoff
Correction belongs inside the review process. On 2026-10-01, an internal acceptance used two synthetic PDF invoices: extraction and export completed, and a scalar correction regenerated results without a second job. That is an internal workflow observation, not a purchase-order benchmark, customer testimonial or evidence of field-level performance for scanned orders. Its limited lesson is narrow: a scalar correction can regenerate results without a second job, so review corrections need not always mean starting a new source extraction. For purchase orders, the same discipline still applies: correct only against visible evidence, record what changed, and approve the result before sharing it. Choose the export format by receiver: XLSX for a review workbook, CSV for a machine-readable table and JSON for a downstream system. Keep source evidence and warnings with the approved draft. Flow's supplied scope does not include automatic ERP posting, so the process should stop at an approved export and a human decision about any later system write.
- Correct against the document and preserve the original evidence alongside the corrected draft.
- When corrections are needed, regenerate exports after review rather than treating the first extraction as final.
- Use XLSX, CSV or JSON according to the receiving workflow, not as a substitute for approval.
- Do not describe an exported worksheet as an accounting approval or an automatic ledger entry.