Before you blame the app, run a two-minute test on the PDF itself. Select a few transaction lines, copy them, and paste into a plain text editor. What lands there is roughly what any reading tool gets to work with. If the lines arrive one per row, an app can probably read the file; if they arrive scrambled, you have found the cause.
How do I run the copy-and-paste test?
- Open the PDF in any viewer.
- Drag across five consecutive transaction lines and copy them.
- Paste into a plain text editor, not a word processor.
- Read what you get. One line per transaction with the date, description and amount in order is a good sign.
- If nothing can be selected, the file is an image and needs recognition first.
What do the results tell me?
| What you paste | What it means | What to do |
|---|---|---|
| One tidy line per transaction | Text layer, regular layout | Try importing, then reconcile |
| Amounts on separate lines from descriptions | Columns are stored as separate blocks | Rebuild by hand, or use a spreadsheet export |
| Garbled characters | Odd font encoding | Ask the provider for a CSV or a fresh PDF |
| Nothing pasted | Image, not text | Recognition needed; expect errors |
Say a pasted line reads 14 May Bakery on King Street A$7.80 A$1,208.45. Two money figures follow the description, and nothing in the text says which is the amount and which is the balance. A person knows; a program has to be taught.
Why is a printed layout hard to turn into data?
Because a PDF stores drawing instructions, not records. The statement's neat table is an illusion made of aligned text. When you ask an app to read it, you are asking it to look at the page and work out the table, in the same way you do, using only positions. Variation between banks, long descriptions and page breaks all break the regularity it depends on.
How do I rebuild the table by hand?
- Paste the lines into a sheet, one per row.
- Use Data, Text to Columns to split on a delimiter if the text has one, or on fixed width if it lines up.
- Join continuation lines to the row above.
- Delete headers repeated on each page.
- Convert amounts to numbers by removing symbols and commas, and give debits a minus sign.
Then reconcile. Imagine a statement opening at A$2,640.00, with credits of A$1,950.00 and debits of A$2,078.40. The closing balance should be 2640 + 1950 - 2078.40 = 2,511.60. Sum your rebuilt amounts: if it equals the difference between opening and closing, 1950 - 2078.40 = -128.40, the table is complete.
Is a spreadsheet export really better?
Usually, since the structure is stored explicitly and nothing is inferred. Look for an export option next to the PDF download. If it exists, use it, check the header and date order, and import. If it does not exist, the copy-and-paste route works for a month or two at a time, though it gets tiring for a year of history.
SpendixAI is built around PDF, CSV and Excel statement import, showing rows for review before saving, and does not promise that every layout will parse perfectly; features for unfamiliar layouts process the uploaded document. It is built around India and INR, so it suits readers who also keep Indian accounts, not Australian dollar ones.
Keep this routine
- Run the paste test on any new statement format.
- Prefer a spreadsheet export.
- Check totals before saving.
- Store the original file.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

