Drop your PDF on bankpdf (or drag it into the drop zone on the home page). bankpdf sends the file to Mistral AI for OCR extraction, automatically identifies your bank via a dedicated adapter, then normalises dates, amounts and descriptions. In under a minute you get an Excel (.xlsx) or CSV file ready to open in Excel, Google Sheets or your accounting software. Each transaction gets an automatic category from 15 values. No account is required to convert the first statement — you can test immediately without signing up or entering a card. To save your statements in a private library and ask questions in natural language via the AI chat, create a free account.
Triple extraction is the analysis method bankpdf uses to maximise PDF reading accuracy. Each statement is sent to Mistral AI through three independent channels: a markdown extraction (which reproduces the table structure), an HTML extraction (which preserves formatting and tables), and a vision analysis by Mistral Large (which reads the PDF as an image, useful when the text structure is absent). The three results are then compared transaction by transaction. The retained value is the one appearing in the majority of extractions — the max-count principle. This approach strongly reduces classic OCR errors: swapped digits, lines split in two, misread decimals. The result is then reconciled against the statement's declared totals to detect any missing lines.
Usually under a minute for a standard monthly statement (1 to 8 pages). Multi-page statements or low-quality images can take up to 2 minutes.
Yes. JPEG, PNG, WebP, HEIC and GIF are accepted in addition to PDF. Useful for old statements received by mail or photographed from your phone. Photo quality directly affects extraction reliability.
80+ banks have a dedicated adapter: Chase, Bank of America, Wells Fargo, Citibank, Capital One, HSBC, Barclays, Lloyds, NatWest, RBC Royal Bank, TD Bank, BNP Paribas, Société Générale, Crédit Agricole, LCL, Boursorama, Qonto, Deutsche Bank, ING, Santander, Revolut, Wise, N26. Other banks are handled in generic mode — most still work fine.
Each bank has its own conventions: date format (DD/MM/YYYY or MM/DD/YYYY), main currency, description prefixes (e.g. Wells Fargo adds 'PURCHASE AUTHORIZED ON' before each merchant name). The adapter knows these conventions and normalises the result for a clean export.
Yes, privacy is a design priority of bankpdf, not an option. Here is the exact path of your data: your PDF is sent to Mistral AI for OCR extraction — Mistral applies a short retention policy (data is not used to train their models). After extraction, the original PDF is stored on private Vercel Blob in European infrastructure: only your account can access it via a signed URL. Parsed transactions are kept in a Neon Postgres database hosted in Europe, scoped strictly to your user ID. bankpdf does not resell any data, includes no advertising network, and does not exploit your statement content for any purpose other than the requested extraction.
Yes. All bankpdf infrastructure is configured in Europe: Vercel is deployed in eu-west (Dublin, Ireland), Neon Postgres in eu-central-1 (Frankfurt, Germany), and Vercel Blob uses Vercel's European data centres. The tool is GDPR-compliant.
The Excel file contains 6 columns: Date (normalised to YYYY-MM-DD), Description (cleaned of technical prefixes), Debit (outgoing amount), Credit (incoming amount), Balance (after the operation), Category (automatically assigned from 15 values).
Excel (.xlsx), CSV (.csv) and OFX (.ofx). OFX is compatible with Quicken, Sage, Money and most accounting software that supports bank imports.
Email us at support@bankpdfapp.com — we usually reply within 24h on business days.