When Tabula is not working, the cause is usually one of four things: it will not start, the PDF is a scan rather than real text, the table selection or extraction mode does not suit the layout, or the output is fine but needs cleaning. Work through them in that order. Tabula is a free open-source tool, so there is no helpdesk, and its own project pages are the place to confirm requirements for your version.
What if it will not start at all?
- Tabula runs on Java, so check that a recent Java runtime is installed and that the version matches what the project pages ask for.
- If a window opens and closes, run it from a terminal so you can read the error message instead of losing it.
- If your browser shows a blank page, check that nothing else is using the same local address and port, then restart it.
- Try a different browser. A privacy extension can interfere with a local page.
What if the PDF loads but no table is found?
A statement saved as a picture, or one scanned on a phone, has no selectable text, so there is nothing to extract. Test it: try to highlight a word in a PDF reader. If you cannot, the file is an image. Ask your bank for a text-based statement, or a CSV or Excel version, instead of fighting the scan.
Which extraction mode fits which layout?
| Layout | Try | Why |
|---|---|---|
| Ruled grid with lines between cells | The mode that follows lines (often called lattice) | Uses the drawn borders to place cells |
| Columns separated only by space | The mode that follows text gaps (often called stream) | Uses gaps in the text |
| Several tables per page | Draw one selection per table | One box over everything merges them |
| Header repeats on every page | Select rows only, not the header | Avoids repeated header rows in the output |
Option names can differ between versions, so read the labels in your own copy.
Worked example: when the numbers look shifted
Say a statement row reads 14 Apr, Grocery store, 1,240.50, 38,760.00, which is the debit followed by the balance. The extractor puts 1,240.50 and 38,760.00 in one cell. Switch the mode, redraw the selection so it sits tightly between the columns, and extract again. The check: take the previous row's balance of 40,000.50 and subtract the debit: 40,000.50 - 1,240.50 = 38,760.00, which equals the balance in the row. If they match, the rows are right. If they do not, the extraction merged, shifted or dropped something.
What if the CSV opens messy in your spreadsheet?
- Remove title lines and page footers above the header.
- Delete empty rows and repeated headers.
- Convert amount columns from text to numbers, removing commas and stray symbols.
- Fix dates to one format and sort by date.
- Compare totals with the statement.
What should you do when only some pages extract properly?
Pages in the same statement can differ: the first page often has an account summary above the table, while later pages carry only rows. Extract page by page, using a different selection box for the first page, and then join the results in a sheet. Keep a column that records the page number, so that if a row looks odd you know exactly where to look in the PDF. Check the row count per page against what you see on screen before merging.
When is a different route easier?
If you are extracting tables only to track spending, you can skip the conversion step. SpendixAI accepts statement files as PDF, CSV or Excel, and you review each row before saving. It cannot promise that every layout will read perfectly, so check the preview against the statement as you would with any tool. You can also add expenses by hand for anything the file missed, singly or several at once.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


