How AI Can Actually Read Your Receipts: Inside the AI Finance Center
Every NDIS provider organisation eventually drowns in the same pile: fuel receipts, EFTPOS slips, supplier tax invoices, the odd handwritten reimbursement form from a support worker who bought first-aid supplies on the way to a shift. Someone — usually whoever is least busy that week — has to turn that pile into numbers a bookkeeper can actually use. It's slow, it's error-prone, and it's the kind of task that never feels urgent until the BAS is due.
The AI Finance Center is built to do the first pass of that work properly, not just cosmetically. You upload a document — a photo, a PDF, a CSV export — and it comes back with a genuine accounting record: who the supplier is, whether GST applies and why, every line item categorised, a suggested journal entry, and an explicit list of anything that still needs a human to look at it. It's powered by Claude, not a simple text-scraping script, which is the difference between 'here's some text we found on the page' and 'here's a categorised, reconciled record.'
What it actually extracts. For each document: the supplier's legal and trading name, ABN status (never guessed — if it isn't shown, it says so), every line item with quantity, unit price and GST broken out, an expense category (Vehicle & Transport, Office & Admin, Direct Participant Support, and a dozen others), and a check that the line totals actually add up to the document's stated total. If they don't, it tells you by how much and flags it for review rather than quietly rounding the difference away.
The part that matters most: it doesn't guess. A lot of 'AI expense tools' will happily invent an ABN that looks plausible, or assume GST applies because the total happens to divide evenly by eleven. The AI Finance Center is built the other way around — if a detail isn't clearly shown on the document, it comes back labelled "Missing", "Unclear", or "Requires human review" instead of a confident-looking guess. The same discipline applies to NDIS claimability: a coffee bought during a community access shift doesn't automatically get flagged as claimable just because it happened during a shift — that call still needs a service agreement check.
A suggested journal entry, not a final one. Every processed document comes with a proposed double-entry — which account to debit, how much GST is creditable, what to credit against (bank, credit card, or accounts payable). It's there to save a bookkeeper the setup work, not to replace their judgment. Anything with a discrepancy, an unclear GST position, or a possible personal/business split gets an explicit "accountant review required" flag before it goes anywhere near a ledger.
What it doesn't do yet. Bank reconciliation — matching an uploaded receipt against the actual bank transaction that paid for it — is a genuinely separate piece of work, and it isn't built into this first version. The AI Finance Center handles the document side properly first; matching it against a bank feed is a natural next step once that foundation is solid, not something worth faking with a rough guess in the meantime.
The bigger point: this isn't a magic 'AI does your books' pitch. It's a first, careful pass that turns a shoebox of receipts into a structured, traceable record — with every uncertain call surfaced honestly instead of buried, so the accountant's time goes to the decisions that actually need a human, not to re-typing numbers off a photo.