How PDF-to-Excel conversion works
Most bank statements contain a hidden text layer even when they look like fixed pages. ArveBot reads that text inside your browser, then sends only the extracted text to an EU-hosted AI service to identify the account period, masked account number, balances and transaction rows. The original PDF is not uploaded on this path. After structuring, you receive an editable table and can export an XLSX workbook with real date and numeric cell types, a UTF-8 CSV, or a tab-separated clipboard table. Rows stay in source order, and descriptions remain editable before any file is created, so corrections flow into every export format. CSV columns are booking date, value date, description or counterparty, signed amount, running balance after the transaction, and account currency.
Reconciliation catches plausible-looking mistakes
A neat spreadsheet is not enough if one amount has the wrong sign or a decimal was read incorrectly. The converter checks that opening balance plus the sum of all signed transactions equals the closing balance within one cent. When the statement provides running balances, each row is checked in sequence. A mismatch highlights the first row that does not produce the bank’s stated balance. Editing an amount reruns the calculation immediately. A green result is shown only when the totals and row checks agree; otherwise the warning remains visible and exported filenames say “needs-review.”
The statement is not stored
Text extraction happens on your device whenever the PDF provides usable text. Only that extracted text reaches ArveBot’s conversion route and Mistral’s EU-served API for structuring. Scanned pages follow a separate path: your browser renders each page as an image and sends those page images for optical character recognition. Requests contain no filename or PDF metadata. Processing is in memory, no document content is written to disk, no document content appears in application logs, and statement data is never used to train ArveBot. Downloads are built locally in your browser.
Scanned and multi-currency statements
A scan has no reliable text layer, so the browser prepares page images for OCR instead. The free tool accepts up to 15 pages and 10 MB. Clearer, straight pages produce better results than photographs with shadows or cropped edges. For multi-currency accounts, the exported amount remains the signed amount in the account currency; original foreign amounts and exchange-rate details stay in the description when shown on the statement. Always review currency, signs and balance warnings before importing the file into accounting software. The converter helps with transcription, but it does not decide bookkeeping treatment or provide tax advice.