Last updated

PDF to Excel

A PDF document becomes a workbook — up to 25 pages on one credit.

VisionParse converts a PDF document into an Excel workbook — the header fields on one sheet, every line item on another, with amounts still stored as numbers.

Digital PDFs and scanned ones both work: a PDF with a text layer is read exactly, and a scan is recognised first. Five corpus documents were run through the production pipeline, written to .xlsx by the same code your browser runs, then reopened with a spreadsheet library and compared cell by cell against committed ground truth — 115 of 115 values in the right cell. A single credit covers up to 25 pages.

Last updated: 2026-08-06 · Measured by VisionParse engineering

  • AI tier · 1 credit a document
  • PDF · PNG · JPG · WEBP
  • Photos auto-compressed · PDF to 25 pages
  • Deleted within 1h

What this tool does, in numbers.

  • ACCURACY

    115 of 115 cells correct across five benchmark documents, verified by reopening each workbook with a spreadsheet library.

  • LIMIT

    One credit covers a document up to 25 pages; longer statements have to be split before sending.

  • SPEED

    A one-page PDF returns in about two seconds; a large multi-page scan takes longer because every page is recognised.

  • ACCESS

    Needs an account with AI credits: every paid plan includes them monthly, or a one-time pack adds them without a subscription.

  • COST

    One AI credit a document, whether it is one page or twenty-five.

  • PRIVACY

    The workbook is assembled in your browser, so the extracted fields are not sent anywhere a second time to become a file.

  • RETENTION

    The uploaded PDF is deleted within an hour, the extracted fields within 30 days, and nothing trains a model.

Where the 115 of 115 cells, five documents comes from.

Every figure on this page traces to a dated run on hardware we name. Nothing here is a vendor estimate.

Measured 2026-08-06

Claim
PDF documents convert into workbooks with every value in the right cell — and the numbers still behave as numbers.
Method
AI extraction tier (gemini-3.5-flash-lite) into a browser-built .xlsx, one credit. Full methodology.
Sample
Three invoices, a delivery note and a goods received note from the benchmark corpus, run through the production extraction pipeline, written to .xlsx by the browser’s own code, reopened with the openpyxl library and compared value by value against the ground truth committed beside each sample.
Measured
2026-08-06 on Contabo Cloud VPS 8 — 8 vCPU AMD EPYC, 24 GB RAM, Ubuntu 24.04.4
Limitations
Five documents is a corpus, not a census, and all five are business documents in English. Bank statements, price lists and non-English layouts are unmeasured for spreadsheet output; a document beyond 25 pages must be split.
Contact
VisionParse@senithu.lk

What this tool gets wrong

HELD UP

A digital PDF is read exactly, and a scanned one is recognised first. PDFs carrying a text layer are read from the layer itself rather than re-recognised, which is why a digital invoice comes back exact. A scan goes through recognition first, and the same arithmetic checks then run over whatever was read.

KNOWN LIMIT

Twenty-five pages a credit, and long statements need splitting. The page ceiling is ours and is stated up front rather than discovered halfway through a 200-page statement. A document past it is refused with the number, not truncated quietly — a spreadsheet missing pages nobody mentioned is worse than an honest refusal.

KNOWN LIMIT

A large scanned PDF is not fast. An 11 MB scanned invoice measured 48.5 seconds end to end, most of it recognising the pages before the fields are read. Digital PDFs of the same length return in seconds. It is worth knowing which kind you are sending before you wait.

Three steps, no account.

  1. Drop the PDF

    Digital or scanned, up to 25 pages on one credit. Files larger than the browser can post go straight to our API instead.

  2. Say what it is

    Invoice, receipt, purchase order, delivery note or goods received note — the choice decides which fields come back.

  3. Download the workbook

    The document’s fields on one sheet, its line items on another, and the arithmetic checked before you see it.

What this tool accepts.

PropertyFree tool
EngineAI extraction tier, 1 credit per document
Measured result115 of 115 cells across five documents (2026-08-06)
OutputExcel .xlsx with a Document sheet and a Line items sheet, plus JSON
PagesUp to 25 a credit; longer documents must be split
Large scans11 MB scanned invoice measured at 48.5 seconds end to end
Input formatsPDF, PNG, JPG, WEBP — up to 35 MB a document
AccessAccount with AI credits — any paid plan, or a one-time pack
File retentionOriginals deleted within 1 hour; results within 30 days

When to use the API instead.

This page converts one file now. To convert files from your own software, the same engine runs over HTTP from Starter upward.

# same engine, from your code
curl -X POST https://api.visionparse.app/v1/ocr \
  -H "x-api-key: $VISIONPARSE_KEY" \
  -F "file=@image.jpg"

→ 202 { "jobId": "c0df8e2c…" }

Use the API when

You convert more than a few files, convert on a schedule, or need results inside another system. Batch, webhooks and bounding boxes are API-only.

Frequently asked

Do the numbers stay numbers?

Yes, and it is measured: every quantity and amount in the benchmark came back typed as a number, so totals sum the moment the workbook opens. Values silently stored as text are the usual failing of converted spreadsheets.

Does it work on scanned PDFs as well as digital ones?

Both. A PDF with a text layer is read from that layer, which is why digital invoices come back exact. A scan is recognised first and then read — slower, and measured: an 11 MB scanned invoice took 48.5 seconds end to end.

How many pages can one credit cover?

Twenty-five. A longer document is refused with that number rather than quietly truncated, because a spreadsheet missing pages nobody told you about is worse than a clear refusal. Split it and send the parts.

Can I automate this?

Yes — the same extraction is available over the API with a key from your dashboard, and the JSON it returns is what the workbook is built from. Batch upload in the browser is on the roadmap and is not built yet.

What if my PDF is a bank statement or a price list?

It may well work, but we have not measured it, so we will not claim it. The five benchmark documents are invoices, a delivery note and a goods received note. Send a few real statements and we will measure them properly before you rely on it.

Measured by VisionParse engineering. Method and raw data are published.Updated Benchmark · Method · Data handling