BankXL / Practical guide

Scanned bank statement to Excel: OCR and source checks

Convert a readable scanned bank statement to Excel with local OCR, understand BankXL's fallback limits, and check every recognized row before exporting.

1. Inspect the scan before uploading

Open the PDF and check that every transaction page is present, upright and readable. Keep an untouched copy. Zoom into dates, minus signs and decimal points; if you cannot distinguish them, obtain a clearer scan instead of guessing. For a scanned bank statement to Excel workflow, start with the source image, not an assumption that recognition will repair it. Identify the statement period and printed opening and closing balances. Keep different accounts and currencies apart so later arithmetic compares like with like.

2. Check whether the PDF contains text

A text-based PDF contains characters that extraction software can read. A paper scan can contain only pictures of those characters. A scanned PDF to Excel workflow therefore needs recognition before table extraction. Try selecting and copying an actual transaction description; selecting a logo or watermark is insufficient evidence. Some scans already have a hidden OCR text layer, but that layer may contain errors. Adobe's help explains that recognition adds searchable text to scanned pages and requires review afterward. Searchable therefore means text is available, not that the amounts match the printed image. Compare copied characters with what the page actually shows.

3. Upload and inspect the OCR result

Upload a PDF through BankXL's converter and inspect the preview alongside the original. The server first extracts PDF text with pdfplumber. Image-only or low-text pages can be rendered for local English Tesseract OCR. A scan-like page containing text but no detected transactions can also receive a supplemental OCR attempt. Results using OCR carry a recognition warning. Uploaded statements and extracted content are not sent to third-party conversion or AI services. This is recognition of supported layouts, not a guarantee that a scan will become a complete transaction table.

4. Understand document and fallback limits

One PDF can contain at most 100 pages, with a 15 MiB upload limit and a maximum of 3,000 transactions. Supplemental OCR fallback is capped at six attempts per document; this is separate from the initial OCR path for image-only or low-text pages, so it is not a promise to recognize every page or a six-page limit on all scans. OCR can time out or fail during rendering. Encrypted PDFs and decimal-comma amounts are unsupported. Check the failure list for source pages and reasons rather than assuming an exported file contains the entire statement.

5. Resolve unreadable and missing activity

Blurred, handwritten or tilted scans can fail or yield incorrect characters. Cropped margins can hide signs or digits, and complex layouts can place a date beside the wrong amount. Obtain a clearer, straight source and try again where practical. If a value remains unreadable, leave it unresolved until you can verify it against an authoritative original. BankXL lists failed pages and skipped rows and offers a free failure-list CSV. Failed pages do not consume free pages. A page containing some recovered activity can still omit rows, so page success alone is insufficient evidence of completeness.

6. Use balances to investigate recognition errors

Compare opening balance + signed transaction amounts = closing balance for the complete statement. Check missing minus signs, switched debit and credit directions and decimal points when the equation differs. Also look for duplicated transactions and descriptions wrapped into unrelated rows. Printed running balances can help locate a suspect transition. Enter or correct opening and closing balances when necessary, using the same sign convention as the transactions. A partial free export cannot establish the closing balance of a longer statement. Even a match can hide offsetting errors, so arithmetic supplements the source comparison.

7. Compare every row before exporting

Work from the first source page to the last. Match each date, description, signed amount and available running balance to its image, then check that every printed transaction has a corresponding row. Review repeated headers, page continuations and ambiguous day/month order. Correct verified values in the preview before downloading XLSX or CSV. The workbook includes a Balance check sheet; keep the original PDF and any failure list with your checked export. Download promptly because sources, previews, corrections and exports expire within 24 hours; editing does not extend that window.

Trying a reviewable scan workflow

I built BankXL so recovered statement rows can be inspected and corrected before use. Try a readable scan and judge its preview against the original. If recognition leaves unresolved values or missing activity, obtain better source material before relying on an export. 10 free pages with data per browser credential, plus a separate 10 per account, across PDFs. Preview, corrections, balance checks and XLSX/CSV downloads are included. Only successful pages use the allowance. Shared network limits: 30 free pages in 30 days across anonymous and account use; anonymous use also shares 10 pages per network in 30 days, so clearing cookies does not renew it. Free uploads: 10 attempts per credential/account and 30 per network per hour. Remaining pages need one-time packs: $9 / 20 pages or $29 / 100 pages. No subscription or automatic renewal. Partial exports state their scope and source pages. No sign-up or card is needed for browser use.

Use your free pages See page packs

Written by Zijian Xiao. Examples are synthetic. Review all extracted results against your original PDF.