Two kinds of PDF, one file extension
A PDF exported by a bank usually contains real text: every character is stored, positioned on the page. Reading it is a matter of recovering the layout, and the characters themselves are exact.
A PDF produced by a scanner or a phone contains images. There are no characters at all, only pixels that look like them. Reading it requires OCR, which is slower, costs more and introduces its own errors.
Tools that treat both the same way get one of the two wrong. Karchu checks first and takes the cheaper, more accurate path whenever it exists.
What happens to a scanned page
Before OCR runs, the page is straightened, orientation is corrected, and contrast is adjusted conservatively. Conservative matters here: aggressive cleanup can erase thin digits and turn an 8 into a 3, which is worse than a slightly grey page.
The processed image is stored separately and the original is kept untouched, so a bad read can be revisited without asking you to upload again.
OCR output carries per character and per field confidence. Where confidence is low, the value is marked rather than accepted.
From page to table
Structure detection finds the transaction table, identifies the date, description, debit, credit and balance columns, and handles the layouts that break naive parsers: single amount columns with signs, separate debit and credit columns, running balances, multi line descriptions, and summary blocks that must not be imported as transactions.
Date and number formats are read according to the document rather than a single assumption, which is what makes a UK statement, an Indian statement and a European one all come out correct.
Then the arithmetic is checked
For statements, the balance chain is verified: opening balance plus credits minus debits should reach the closing balance. A mismatch is reported with the difference and the likely causes, missing rows or duplicated rows, rather than hidden.
This is the step most extraction tools skip, and it is the one that tells you whether the extraction can be trusted at all.
Frequently asked questions
What is a PDF analyzer?
It is a tool that reads the contents of a PDF and returns structured data. For financial PDFs that means the transactions, dates, amounts, balances and totals on the page, in a table you can sort and export, rather than text you have to retype.
Does it work with scanned PDFs?
Yes. A PDF that contains real text is read from that text layer, which is exact. A scanned PDF has no text layer, so the pages are cleaned up and read with OCR instead. Karchu picks the right path per document without you choosing.
How many pages can it handle?
Up to 300 pages per document. Every page is processed, not just the first few, and rows that continue across a page break are joined back together.
What about password protected PDFs?
You can unlock a password protected statement in your browser by entering the password, and the unlocked content is then processed normally. The password is never sent anywhere.
Can it convert the PDF to Excel or CSV?
Yes. Once the transactions are extracted you can export to CSV or Excel, and to QBO and the common accounting formats. The point of extracting first is that the export contains checked data rather than a rough text dump.
Why does it sometimes ask me to review rows?
Because some pages are genuinely ambiguous: a faint scan, a wrapped description, a column that does not line up. Rather than pick a value and present it as fact, Karchu marks the row and shows you the original page beside it.