The most expensive typing in your business
Someone opens a PDF. They read a supplier name, an invoice number, a date, a total, and eight line items. They type all of it into another system. They do this forty times a day.
It is boring, it is expensive, and it is the single most error-prone process in most businesses, because human attention degrades exactly when volume is highest. And the errors do not surface until reconciliation, weeks later, when they cost ten times as much to fix.
What we build
A document lands, usually by email. The system reads it, works out what kind of document it is, extracts the fields that matter, checks them against what it already knows, and either files it or flags it.
- Capture: from email, a watched folder, or an upload portal.
- Classify: invoice, credit note, statement, contract, purchase order.
- Extract: fields and line items, with a confidence score on each.
- Validate: against purchase orders, supplier records, expected ranges.
- Route: straight through if clean, to a review queue if not.
- Push: into Xero, MYOB, or whatever holds the record of truth.
Why the validation step matters more than the extraction
Extraction is largely a solved problem now. Modern models read documents well. The hard part, and the part that determines whether the system is trusted, is knowing when the extraction is wrong.
A system that is right ninety-five percent of the time and confident all the time is worse than useless, because you have to check everything anyway. A system that is right ninety-five percent of the time and tells you which five percent to look at saves you ninety-five percent of the work. We build the second kind.
What this looks like in production
The same extraction and validation pattern underpins the screening pipeline in Vantage 360, where the documents are CVs rather than invoices but the shape of the problem is identical: high volume, unstructured input, and a decision that has to be defensible after the fact.