How it works
From a pile of documents to a checked ledger
Six steps run automatically in the background. You step in only where the AI is unsure — and it tells you where that is.
1. Upload in bulk
Drop hundreds of PDFs and photos per client company. Duplicates are caught by file content, not file name.
2. Split multi-slip scans
A page with six receipts becomes six documents. Each slip is counted first, then cropped and verified.
3. Classify
The printed title and the layout are read separately. When they disagree, the title wins and the conflict is shown to you.
4. Extract every field
Dates, invoice numbers, parties, line items, totals, payment methods and cheque details — in Chinese and English.
5. Link transactions
Quotations, invoices, delivery notes, cheques and payments are chained by number, amount and date order.
6. Review and export
Work through unclassified and unmatched queues, then export one chronological Excel ledger with the source file on every row.
Where you stay in control
- Low-confidence results are flagged, never silently accepted.
- Disagreements between the printed title and the layout are shown side by side.
- Every extracted value can be edited next to the original scan.
- You decide what goes into the exported ledger.