How to extract multi-page tables
Dynamite Docs, 2026-08-30
Map where the table starts and ends
Find the first real header, the final data row, and any sections where the columns change. A document can contain several tables that look similar but describe different periods, accounts, or totals. Do not join them until their meaning and columns match.
Record the page number for every extracted row. Page provenance makes boundary problems much easier to inspect later.
Create one canonical set of columns
Map every repeated header to the same field names. If a narrow page abbreviates "Description" to "Details", decide whether they mean the same thing before merging. Keep genuinely new columns instead of pushing their values into the nearest existing field.
Remove headers and footers by role
Repeated column labels, page numbers, confidentiality notices, and carried-forward labels should not become data rows. Match them by their purpose and position, not by one exact string. OCR may read the same footer slightly differently on every page.
Carry unfinished rows across page breaks
The last line on a page may continue at the top of the next page. Hold that row open when required anchor fields are missing or the sentence clearly continues. Merge the next fragment before starting a new record.
Check carried-forward and brought-forward balances separately. They can be useful validation markers even when they do not belong in the exported transaction table.
Keep section context with every row
A table may reset inside the same PDF for a new account, currency, department, or statement period. Add those section values to each row before appending the next section. Otherwise identical-looking records lose the context that tells them apart.
Validate each page, then the combined table
Compare row counts and page subtotals first. Then validate the complete table using grand totals, running balances, sequence numbers, and date order. This two-stage check tells you whether an error came from page extraction or the join itself.
Process the full document in Dynamite Docs
Upload the PDF once. Dynamite Docs reads all pages, carries the inferred schema through the document, and shows the combined output with source-page references. Review boundary rows and flagged values before exporting to Excel, CSV, or Google Sheets.
Related workflow: Bank statement extraction to Excel and CSV.
Try it yourself. Upload a PDF, scan, or image and let Dynamite Docs infer the schema.