What scanning actually has to do
Most people mean two different things by scanning a document. One is producing an image of it. The other, the useful one, is turning the content into data: a date, a description, an amount, a balance, a tax figure, in a shape you can add up.
The second job is harder because financial documents were designed for human eyes. Columns are implied by whitespace, descriptions wrap over three lines, amounts sit in separate debit and credit columns or carry a trailing minus, and dates appear in whichever order the issuing country prefers.
How Karchu reads a document
The file is validated first, so an empty, corrupted or locked file gives you a clear message rather than a silent failure. Then the document type is detected, statement, receipt, invoice, bill, payout report or spreadsheet, with a confidence figure and a prompt to confirm when that confidence is low.
For PDFs the embedded text layer is used when there is one, because it is exact and costs nothing. Only when there is no usable text does the page go through image cleanup and OCR. Images and photos go straight to that path, with orientation corrected first.
Structure detection then works out where the table is, which column holds what, and how rows continue across page breaks. Each extracted row carries its own confidence and a pointer back to the place on the page it came from.
Validation, because reading is not enough
For statements, opening balance plus credits minus debits should land on the closing balance within a small tolerance. When it does not, Karchu says so and shows the expected figure, the extracted figure and the difference, rather than quietly saving numbers that do not add up.
Rows below the confidence threshold, rows with an unreadable date, and rows with an amount that is not a finite number are all sent to review. The rule the whole pipeline follows is that a missing value stays missing.
What you get afterwards
Categorized transactions, category and merchant totals, recurring charges, duplicate and overlap handling across statements, receipts linked to the transactions that paid them, and exports to CSV, Excel, QBO and the common accounting formats.
Everything remains editable. Corrections are recorded with the old value, the new value and who made the change, and a repeated correction can be turned into a plain rule you can read and switch off.
Frequently asked questions
What is a financial document scanner?
It is software that reads a financial document, a bank statement, a receipt, an invoice or a payout report, and turns the text on the page into structured rows and fields you can sort, total and export. Scanning here means reading and extracting, not just producing a picture of the page.
Which documents can Karchu scan?
Bank and credit card statements, receipts, vendor invoices, utility and SaaS bills, payment processor payout reports, accounting exports and spreadsheets. Files can be PDF, JPG, PNG, CSV, XLS, XLSX or plain text.
Does it work with photos of documents?
Yes, within reason. A phone photo, a screenshot or a scan will be read if the text is legible. Rotated pages are corrected and low quality images are still processed, but rows the reader is unsure about are marked for review rather than saved as fact.
What happens to rows it cannot read confidently?
They are flagged and routed to the review screen with the original page shown next to the extracted values. Nothing is invented to fill a gap, an unclear amount stays empty and waits for you.
Is the original document kept?
Yes. The original file is stored unchanged alongside any processed version, so you can always go back to the source. Retention is configurable at the workspace level and processing never deletes the original.
Do I need to connect my bank?
No. Karchu works from files you upload. It never asks for bank credentials and there is no bank connection to authorise.