Not every PDF is the same kind of file, and that is the real reason budgeting apps struggle with them. A statement downloaded as text can often be read, a scan needs optical character recognition, a photo taken on a phone is harder still, and a protected file cannot be read until its owner removes the protection. Knowing which kind you hold tells you what to expect.
Which kind of PDF do I have?
| Kind | How to tell | How hard to read |
|---|---|---|
| Text PDF straight from the bank | You can select and copy words with the cursor | Most readable |
| Scan of a paper statement | Cursor selects the whole page as one block | Needs character recognition |
| Photo of a screen or paper | Skewed edges, shadows, uneven lines | Hardest; errors likely |
| Protected PDF | Asks for a code to open | Unreadable until the owner opens it |
The quick test is to try to highlight a single word. If the word highlights, there is a text layer and a tool can use it. If you highlight a block, you have an image.
Why can't a text PDF simply be read like a spreadsheet?
Because the file stores positions, not rows. A line such as 12 Mar, Corner Newsagent, £6.40, £873.15 is four separate pieces of text that happen to share a height. A reader has to work out that the first is a date, the second a description, and that both numbers are different things, one a payment and one a balance.
Layouts that vary make it worse. One bank prints Paid out and Paid in as two columns, another prints a single amount with a minus sign, a third prints a letter after the number. A reader built to expect one of these has no way of recognising the others.
What goes wrong in practice?
- Balances are mistaken for amounts, so a £6.40 coffee becomes a £873.15 one.
- Dates without a year, such as 12 Mar, are read in the wrong year across a December to January statement.
- Long descriptions that wrap onto a second line become separate rows.
- Page headers and footers get read as transactions, creating phantom lines.
- Credits and debits swap sign when a column heading is missed.
What are the practical workarounds?
- Ask whether your provider offers a CSV or spreadsheet download for the same period; this avoids the problem entirely.
- If only a PDF exists, copy and paste a table into a sheet, then tidy it.
- Add the year to short dates with a formula if needed, for example =DATE(2026, MONTH(DATEVALUE(A2&" 2026")), DAY(DATEVALUE(A2&" 2026"))), which works where the sheet understands day-month text; check the first few results.
- Delete page headers and footers by sorting and spotting rows with no amount.
- Reconcile with the closing balance.
A reconciliation example: a statement opens at £2,030.00, takes in £2,450.00 and pays out £2,311.85. Opening plus movement is 2030 + 2450 - 2311.85 = 2,168.15. Your cleaned table should total to that; if it is £6.40 short, you know one purchase vanished.
What does a sensible tool do instead of guessing silently?
It shows you what it read and lets you fix it before saving. SpendixAI is built around importing PDF, CSV and Excel statements with a review step, and it does not claim that every layout parses perfectly; features for unfamiliar layouts process the uploaded document, so check what that means for you. It is built around India and INR, so it is relevant mainly to readers who also keep Indian accounts and not to pound ones.
Before you blame the app
- Is the PDF text or an image?
- Is it protected?
- Is a CSV available?
- Do the opening balance and the closing balance reconcile with your rebuilt rows?
Useful ideas start with a clearer picture of your own spending.
Explore your reports

