When people talk about invoice automation, the subject is usually data extraction: how well are amounts, invoice numbers and line items recognised? In practice, though, automation often fails one step earlier – at the question of what actually counts as a document. Emails rarely bring a clean single invoice. They bring stacks of attachments in which several business transactions are mixed together, often in a single PDF file: the reminder covering several invoices, among them an invoice with an attached subcontractor invoice and its own delivery note.
Sorting such stacks by hand does not remove the manual effort; it merely shifts it to the start of the process. Automation must therefore begin with separation: analysing each page individually, splitting business transactions into separate cases and recognising the hierarchy within each case – without separator sheets, without barcodes, without pre-sorting. Only once that works reliably is it worth looking at extraction quality.
How Pridonu Cube separates documents →Please note: this article reflects the legal position and administrative guidance as at the date stated and is no substitute for tax or legal advice.