Whether you are reconciling a year of business expenses, preparing for an accountant, or building a budget, you eventually need transactions in a spreadsheet, not locked in PDF statements. Here are the methods, from most to least reliable.
Option 1: download CSV instead of PDF
Before converting anything, check your online banking. Most banks offer transaction exports in CSV, OFX, QIF or Excel format, usually under "Export" or "Download transactions". This is the cleanest data you can get, with no conversion errors.
Limits: some banks only offer recent months, and closed accounts may only have PDF statements. See what is a CSV file.
Option 2: table extraction tools
For digital (not scanned) PDF statements, table extraction tools read the text layer and rebuild rows and columns:
- Excel's Get Data → From PDF (Windows, Microsoft 365) imports tables directly.
- Tabula (free, open source) lets you draw a box around the table on each page.
- Dedicated statement converters are built for common bank layouts.
See how to convert PDF to Excel and how to convert PDF to CSV.
Option 3: AI extraction
Multimodal AI assistants can read statement pages and produce a table with date, description, amount and balance columns. They handle messy layouts and scanned statements better than older tools.
Two cautions:
- Accuracy: AI can skip rows, swap digits or misread signs, especially on long statements. Always check totals.
- Privacy: bank statements are sensitive. Use an approved, no-training plan, or a local model. See risks of using AI on confidential PDFs and extracting tables from PDFs with AI.
Option 4: scanned statements
If your statements are scans, run OCR first, then use one of the methods above. Quality improves a lot at 300 DPI. See how to make a PDF searchable with OCR.
Always reconcile
Whatever the method, check the result:
- Opening balance + credits − debits = closing balance for each statement.
- Count the rows against the statement where it shows the number of transactions.
- Check signs. Debits shown in brackets or with "DR" often get misread as positive.
- Check dates, particularly day/month order and year boundaries in December–January statements.
- Look for page breaks, where a row split over two pages can be duplicated or lost.
If the closing balance matches, the extraction is almost certainly complete.
Clean the data
- Split combined columns (some statements put debits and credits in one column).
- Normalise descriptions ("AMZN Mktp" and "Amazon.com") if you plan to categorise.
- Save as CSV with UTF-8 encoding to avoid garbled characters. See fix garbled characters in CSV files.
Keep the original PDFs
Your spreadsheet is a working copy. Keep the original statements as the record, ideally in an organised archive. See building a personal document archive.
Takeaway
Download CSV from your bank when you can. When you cannot, use a table extraction tool or AI, then reconcile against the statement balances before trusting a single number.
