Guide
Native PDF import, general AI or a statement converter?
A new client arrives with a folder of old statements. Before choosing a tool, check what the bank and the client’s accounting account already provide. This guide is for bookkeepers in the UK, Australia, New Zealand, South Africa and Ireland preparing historical transactions.
Published · Updated
First, see whether you need to extract a PDF at all
A bank feed covering the missing period saves an extraction step. If the bank still offers those transactions as CSV or OFX, download that file and review it for import. You do not need BalancedPDF for either route. Our bank export reference lists documented formats and history limits by bank and account.
Ask the client for the account and dates you actually need. A feed may begin part-way through a month; importing the whole month again can duplicate transactions already present. Keep the statement PDF as your reference even when you obtain a structured download.
Use native extraction when your accounting account offers it
An available native PDF importer can keep the work inside the ledger you already use. Check the help for the organisation’s country and plan, the accepted documents and the review step before relying on it for a client’s history.
Xero’s direct PDF bank statement import is currently limited to organisations in the United States and Canada. Hubdoc’s bank statement extraction is also a US and Canada feature. For the five countries this guide addresses, do not assume a document upload will extract bank transactions. The Xero import walkthrough explains the separate file-import route; the Hubdoc workflow covers preparing missing history outside those regions.
Native extraction still needs a check against the statement. Being inside your accounting software does not establish that every transaction, date and sign was read correctly.
General AI can produce a table; plan the checking work
A general assistant such as ChatGPT or Claude can help turn statement text or a PDF into rows. For a short, legible document that you can verify yourself, that may be sufficient. Specify the columns and the date interpretation, and keep the original document beside the result.
Treat these as failure modes to test for in any generated table: an omitted transaction, an extra transaction that is absent from the PDF, a reversed money-in or money-out sign, or a day and month exchanged. A new conversation can produce a different reading of the same input. These are possibilities to check, not measured error rates for a named product.
Before uploading client information, check the data-use settings and retention terms of the particular plan and workspace you will use, and whether that use meets your client’s requirements. Do not assume that every plan handles uploaded documents the same way.
Apply the same checks to every route
A matching final total is useful evidence, not a guarantee. A missing payment and a missing receipt of equal size leave it unchanged; so can two compensating sign errors. Page counts, dates and individual balance links test different things.
- For a bank account, add the signed transactions to the printed opening balance and compare the result with the printed closing balance. If either balance is missing, record that this total cannot be verified. Credit-card statements may use a different sign convention; check amounts owed accordingly.
- Count actual transaction entries on each PDF page and compare with the output for that page. A description wrapping over three lines is one entry. Check the first and last entry too, including pages that continue a table.
- Inspect dates around month changes and dates with both parts between 1 and 12. For example, 03/04 can mean 3 April or March 4, and a balance equation cannot decide which. Compare with an unambiguous printed date or confirm the bank’s convention.
- Match payments, receipts, refunds and fees to their printed signs or columns. Do not infer direction from the description alone.
- Where running balances are printed, check each successive balance against that row’s signed amount. Trace a break to the original entry before importing. If balances run newest first, follow the chain in chronological order.
When a dedicated converter earns its place
Several months of catch-up work or a client handover make repeated checking expensive. A dedicated review workspace is useful when you need each output row tied to its source, visible exceptions and accounting export files after review.
BalancedPDF links rows from text PDFs to their original lines and checks the statement against its balances. Scans can be converted, but do not have the same line-by-line tracing; inspect the page image. Read our published test results and limits before deciding whether it fits the documents you have.
Choose the route that leaves you with verifiable transactions and the least remaining work. If a bank download or your software’s native workflow already does that, keep using it.
To weigh dedicated converters against each other on published prices and terms, see our comparison of converters for Xero.
Questions
Does a balanced total prove that AI read every row correctly?
No. Opposing errors can cancel, and dates do not enter the balance equation. Compare transaction counts by page, inspect dates and signs, and check running balances where available.
Should I buy a converter if my bank already exports CSV?
Usually there is no extraction job to pay for. Check that the bank file covers the required account and history, then import the supported format and reconcile it in your accounting software.
Sources
Xero PDF availability checked on 10 November 2026. The Hubdoc regional note follows its CA/US extraction guide, previously checked on 9 November 2026. Recheck local availability before choosing a workflow.