OUTPUT DECISION GUIDE
Start with the output you need. Both methods can use the same document, but they solve different jobs.
Use the matrix before uploading. It separates output shape, review work, access, and cost without promising guaranteed accuracy.
| Decision | PDF-to-Excel conversion | AI field extraction |
|---|---|---|
| Primary job | Keep most detected rows and columns | Return only named fields or records |
| Best starting input | A document with visible table structure | Invoices, forms, statements, or irregular layouts |
| Typical output | A workbook that follows detected tables and sheets | Normalized rows with chosen column labels |
| Review focus | Rows, columns, merged cells, and repeated headers | Field mapping, missing values, and ambiguous labels |
| Guest access | One PDF per day, up to 3 pages and 5 MB | A limited preview after a qualifying conversion result |
| Full use | Start with the free converter | Full extraction requires an account and credits |
This illustration uses the public synthetic commercial invoice. It explains output shape, not measured accuracy.
Preserve the detected line-item table for spreadsheet review.
Request selected labels when the downstream system does not need the complete table.
EXISTING EXCEL COLUMN WORKFLOW
FOVATA does not import an Excel template on this screen. Use the header names and business rules from your current workbook to define fields in the same order, then export a new workbook and review it before copying or importing data downstream.
Choose only the columns the next process actually needs.
State the source meaning, missing-value rule, and format in ordinary language.
The exported single-value sheet follows the configured field order.
Compare every row with its source before it enters another system.
No JSON download or hidden template file is required. These are the exact kinds of natural-language instructions a user can enter and save in the signed-in workflow.
Record the PDF page that supports this row so a reviewer can return to the source.
Use the supplier name printed as the invoice issuer; do not substitute the buyer or ship-to party.
Copy the invoice identifier exactly, preserving letters, digits, spaces, and punctuation.
Return the purchase-order reference when printed; leave it blank when the document does not provide one.
Use the printed invoice date and normalize it as YYYY-MM-DD only when the date is unambiguous.
Use the explicit payment due date, not a date inferred from payment terms.
Return the printed ISO currency code or symbol context; flag an ambiguous currency for review.
Return the final invoice total in the stated currency, excluding subtotal and tax-only amounts.
Select a synthetic invoice page to see how its declared reference values fit the same destination columns. This is not live or historical model output.
| Source page | Page 1 |
|---|---|
| Vendor | Northwind Test Supply |
| Invoice no. | NW-260718 |
| PO no. | PO-4108 |
| Invoice date | 2026-07-18 |
| Due date | 2026-08-17 |
| Currency | USD |
| Total | 1480.00 |
A matching column name is not proof that a value is correct. Missing, ambiguous, or differently formatted source values remain a human-review decision.
Source quality still matters. Blurred scans, cropped tables, handwriting, and unusual reading order can degrade either result.
Review is still required. Keep the source beside the output when financial, customs, tax, or operational decisions depend on it.
Costs follow the workflow. The guest converter has published limits. Full AI extraction uses credits after sign-in.
Accuracy is not guaranteed. Review the versioned small-sample benchmark, then test your own document type and quality range.
Use conversion for a reviewable workbook. Choose extraction when named fields and normalized records are the actual deliverable.
