What you will get from this guide
Scanned statements and photos are the hardest case: there is no selectable text, tables are skewed, and numbers can be misread.
Modern AI/OCR can extract structured transactions reliably, but results depend on scan quality. This guide shows a practical workflow and best practices.
Step-by-step
1. Scan or photograph with enough quality
Aim for 300 DPI scans or a sharp phone photo with good light. Avoid shadows and make sure the page is flat.
2. Upload and let OCR extract transactions
Upload the PDF or image to KontoCSV. AI/OCR detects the table rows and pulls dates, amounts, and descriptions.
3. Verify and export CSV/Excel
Spot-check a few rows and reconcile totals. Then download CSV/Excel (or DATEV/Lexware profiles for accounting imports).
Best practices
- Straighten pages: skewed scans reduce OCR accuracy the most.
- Increase contrast: clearer text and lines lead to better table detection.
- If one page is problematic, re-scan just that page instead of the full statement.
FAQ
Can OCR handle old and low-quality statements?
Often yes, but quality matters. Clear scans with readable text work best. Very blurry or heavily compressed images can cause missing digits or wrong separators.
Why does generic OCR often fail on statements?
Statements are tables with repeating columns and multi-line descriptions. Generic OCR might read text, but it does not reliably reconstruct the row/column structure.
What is a good validation step?
Compare statement totals and balances with the exported CSV sums. Also check a few random rows for date and sign correctness.
German original (more detail)
If you want the full German version with screenshots and extra edge cases, open the original guide here: