Sample: Invoice Intake Batch for an Ecommerce Operator
This worked example walks through a single week of supplier invoices handled by the invoice intake and categorisation service for an ecommerce operator. It shows what was sent in, the categorised table that came back, and the corrections a human reviewer made before the operator imported anything into their books.
What was sent in
The operator forwarded twelve invoices for the week. They were not tidy:
- A mix of PDF invoices and photos of paper invoices taken on a phone.
- Three invoices from the same packaging supplier, whose name was spelled slightly differently on each.
- Two invoices with handwritten delivery notes in the margin.
- One invoice that had been emailed twice, a few days apart.
This is a normal week’s pile, not a clean test set. The point of the service is to take it as it arrives.
What came back
The service returned a single categorised table, one row per invoice, with consistent columns:
- Supplier, invoice number, and date — read from each document.
- Net, tax, and total — separated out so the totals reconcile.
- Expense category — assigned per invoice using the operator’s own category list.
- Status — clean, held, or needs-confirm.
The duplicate was not silently merged. It was flagged and held with a note, so the operator could decide rather than have a payment go out twice.
What the reviewer changed
A human reviewer checked the read against the source images before the table was delivered:
- The emailed duplicate was caught and held.
- A tax amount on one photographed invoice had been misread; the reviewer corrected it against the image.
- The packaging supplier’s three name spellings were standardised to one, so the supplier groups correctly in the books.
Two reads stayed genuinely ambiguous — a faint figure on a creased photo — so rather than guess, the reviewer marked them needs-confirm and pointed the operator at the exact field.
The deliverable
The operator imported eleven clean rows, looked at one held duplicate, and confirmed two figures in under a minute each. No retyping, no chasing which packaging invoices belonged together.
This example is intentionally small. Real batches vary in volume and in how deeply each invoice is reviewed; both are set at intake.