When a supplier's table splits across PDF pages: a review-first checklist for Skavio Flow

A four-page supplier invoice arrives with the line-item table starting on page 2, continuing under a repeated header on page 3, and closing with a totals block on page 4. The risk is not only the extraction itself. It is accepting a spreadsheet without asking whether the row at the bottom of page 2 and the row at the top of page 3 describe two separate items or one item that ran out of paper. This guide treats each page boundary as a named review checkpoint, using Skavio Flow's documented extraction, evidence and review-warning outputs.

The page break is visible; the ambiguity is the review problem

Embedded PDF text and OCR both deliver characters. Neither delivers the business meaning of a continuation. A supplier layout can repeat the column header on every page, or repeat it only once; it can carry a subtotal forward, restart position numbers, or continue an item description in a second block with no quantity beside it. These are layout conventions rather than extraction errors, which is why they can survive into a spreadsheet unnoticed. The practical response is to decide in advance which of these patterns you will look for, and where in the reviewed workbook you expect each one to appear.

• A repeated column header should be treated as a label, not a new item. Check whether it arrived in the line-item worksheet as a data row.
• A continuation description may appear as an orphan row with an empty quantity, unit or unit price.
• Phrases such as “carried forward” or “transfer” can describe movement between pages, not a new purchase.
• Some supplier layouts restart position numbers on each page, so a position number alone may not be a unique row key.
• Page-level subtotals can look structurally similar to ordinary line items. Keep document-level totals in header fields, not in the line-item sequence.

Define header fields and line-item fields before the PDF is uploaded

Skavio Flow extracts with custom fields and saved company workflows, and its outputs include an XLSX document worksheet and line-item worksheets. That split is a useful discipline for split tables: decide once which values are document-level facts and which are repeatable rows, then reuse the same field set for the next supplier. Consistency across documents gives reviewers the same checkpoints to apply each month.

• Document-level fields: supplier name, invoice or order number, document date, due date, currency, net total, tax total and gross total.
• Line-level fields: position, description, quantity, unit, unit price, line net amount, tax rate and line total.
• Keep reference fields separate: supplier reference, order reference and contract reference are three different things and often look alike.
• Save the field set as a company workflow so a new PDF is checked against the same structure rather than a new one.
• Decide which fields a human must confirm on every document, and apply that rule to single-page and multi-page invoices alike.

Extract, then read warnings and evidence before the totals

Skavio Flow reads embedded PDF text locally, uses Worker Fabric OCR for scanned pages, and applies the configured OpenAI model for structured extraction. The output includes source evidence and review warnings alongside XLSX, CSV and JSON exports. Open the warnings and source evidence first, then the workbook. Review warnings are intended to surface risk; check whether any coincide with the page boundaries in your document. Then check the numbers.

• Confirm the page count before submission; a four-page invoice is a different review task from a one-page invoice.
• Check the live server quote at submission. Current published Flow pricing is 29 shared credits per page with a minimum of 29 credits per document; trial eligibility follows the current signup policy.
• Check the source evidence for the rows around each page boundary before checking any total.
• Nothing in the documented Flow scope says that tables split across page boundaries are stitched automatically. Treat continuity as an explicit review checkpoint.
• Do not treat a successful extraction as approval. Flow's documented review position is that dates, totals and line items need checking.

Reconcile the rows that cross the boundary

This reconciliation cannot be replaced by a summary. Take the last two rows of one page and the first two rows of the next, and compare them against the original PDF line by line. Then widen out: a page-boundary problem may show up as a small arithmetic gap rather than an obvious duplication, because one continuation row was dropped or one repeated header was counted.

• Compare the row count in the line-item worksheet with any stated item count on the document.
• Add the line net amounts and check them against the document net total. A difference of exactly one row is worth investigating.
• Inspect whether the first row on page N+1 repeats the last row on page N.
• Check units and quantities at the boundary, where a page break can separate a quantity from its unit.
• Review the tax lines on the final page. Multiple tax rates should reconcile to the header tax total, not to a page subtotal, subject to your own tax review.
• If a description continues in a second block, decide explicitly whether it is one line item or two, and record that decision.

Correct, regenerate the export, and keep the evidence together

A verified acceptance run on 2026-10-01 extracted and exported two synthetic PDF invoices; a scalar correction then regenerated the results without starting a second extraction job. That matters for split tables because after review the correction might be a single wrong quantity, a duplicated header row, or a totals field that should have been document-level. Fixing a value and regenerating the export is a review step, not a new document pass. The export remains a draft until a person has signed it off.

• Choose XLSX when reviewers need the document and line-item worksheets in one workbook; choose CSV or JSON when the receiving system is machine-readable.
• Keep the original supplier PDF next to the exported file so later questions can be answered from source, not memory.
• Record what was corrected and why, especially at page boundaries, so the same supplier layout has a repeatable review trail.
• Use a two-person rule for anything above an agreed value threshold: one person extracts, another confirms the boundary rows.
• Do not describe an exported workbook as posted to accounting software. Skavio Flow produces reviewed files; the posting decision remains yours.

Open Skavio Flow and map your first split-table supplier PDF →