Most Australian bookkeeping teams already run receipts through Dext to pull the numbers off a photo and push them into Xero or QuickBooks. So where does Claude fit? The honest answer is that they solve different halves of the same problem. Dext is built to capture structured data from a document. Claude is built to reason about that data, write the words around it, and handle the messy exceptions that a fixed extraction tool leaves on the desk. Understanding the split helps you build a receipt pipeline that costs less to run and needs far less manual cleanup at the end of the month.
What Dext is genuinely good at
Dext (formerly Receipt Bank) is a purpose-built data-capture engine. You forward an invoice or snap a receipt, and it reads the supplier, date, GST, and total, then syncs a transaction to your accounting ledger. For high-volume, repetitive documents it is fast and dependable, and it keeps a compliant image trail for ATO substantiation. Its strengths are narrow on purpose:
Consistent field extraction from receipts and supplier invoices at volume
Direct sync into Xero, QuickBooks Online, and MYOB with supplier rules
A stored document image for every transaction, which matters for GST substantiation
Publisher rules that remember how a given supplier should be coded
If your workflow is thousands of near-identical receipts a month, a dedicated capture tool earns its keep. It is not trying to be clever, and that predictability is exactly the point. The trouble starts with the documents that do not fit the template it expects.
Where Claude fits the same pipeline
Claude is not a form scanner. It is a reasoning tool that reads a document the way a person would, which makes it strong exactly where fixed extraction struggles. A crumpled cafe receipt with a handwritten tip, a supplier statement that bundles ten invoices into one PDF, an overseas invoice with no GST line, a subscription charge that needs splitting across two cost centres: these are the cases that get kicked back for a human to sort. Claude can handle the judgement in those cases and, importantly, explain what it did and why.
In practice, Australian firms use Claude alongside a capture tool for tasks like:
Reading a multi-invoice supplier statement and separating it into line items
Flagging when a receipt has no valid ABN or GST is calculated incorrectly
Drafting the client email that asks for a missing tax invoice
Categorising unusual expenses against the firm's own chart of accounts, with a reason
Turning a month of coded transactions into a plain-English note for the client
The difference is that Claude does not just return a number. It returns a decision and the reasoning behind it, which is what a reviewer actually needs to sign off quickly and what a fixed extraction score can never give you.
Comparing the two on a real month
Picture a Sydney bookkeeping practice processing about 2,000 receipts a month across 40 clients. A capture tool handles the clean 85 per cent without complaint. The remaining 15 per cent, roughly 300 documents, are the exceptions that eat the week: unreadable images, split charges, missing invoices, and odd supplier formats. If a bookkeeper spends an average of four minutes on each exception, that is 20 hours a month of manual sorting. At a blended cost of $75 an hour, the exception handling alone runs to about $1,500 a month, or roughly $18,000 a year, before anyone has written a single client note.
Point Claude at that exception pile and a large share of it clears without a human touch. The tool reads the awkward documents, proposes a coding with a reason, and hands back only the genuinely ambiguous ones for review. You are not replacing the capture tool. You are giving the 15 per cent it cannot handle somewhere sensible to go, which is where the hours and the cost actually sit. Even halving that exception load returns a full working day each month to the team, and that time tends to go straight back into client advisory work that a scanner cannot do.
How to decide what runs where
A useful rule for an Australian firm: use a dedicated capture tool for the predictable bulk, and use Claude for judgement, exceptions, and the writing. The two are complementary, not competing. If you already pay for Dext, you do not need to rip it out. The question is whether your team is still spending real hours on the documents it hands back, because that is the part a reasoning tool removes. Start by measuring one month of exception time honestly, then decide how much of it is worth automating.
The other consideration is data handling. Australian firms working under the Privacy Act want to know where client documents go and how they are retained. That is a fair question to put to any tool in the pipeline, and one worth settling before you scale a workflow across every client rather than after.
If you want help mapping which parts of your receipt pipeline should stay with a capture tool and which are better handled by Claude, book a brainstorm and we will work through it against your actual volumes.
