OCR · Document Digitisation
Upload the PDF or a photo of a purchase invoice, delivery note or order, and the system extracts the supplier, line items, net amounts, taxes and due dates, and creates the document for you.
This is not text recognition returning loose words: it is structured extraction that understands what each number is. That is why it works across invoices from different suppliers without configuring a template for each one, exactly where traditional capture systems fail as soon as twenty new suppliers appear.
It also learns your equivalences: the name a supplier uses for an item is matched to your own reference and comes in already filled next time. Documents are uploaded one at a time or in batches, with the whole month's folder at once.
What makes it usable in production is the review step: the document is proposed and a person validates it with the original alongside. The machine removes the mechanical work and the person keeps the decision.
What it extracts
It is not reading text: it is understanding what each number is
A classic OCR returns the words it sees. That forces you to tell it where to look on each document, with one template per supplier, and the system collapses as soon as a supplier redesigns their invoice or twenty new ones appear.
Here extraction is structured: it identifies each field by what it means, not by where it sits on the page.
- The supplier, their tax ID and fiscal details.
- Document number and date.
- Line items with quantity, price and discount.
- Net amounts, VAT rates and withholdings, with their breakdown.
- Due dates and payment method.
It works for purchase invoices, delivery notes and orders, both as PDFs and as a photo taken on a phone, which is how many receipts arrive from a technician or a rep.
Review before saving
What makes it usable in production
No extraction system is right one hundred per cent of the time, and a system that is right 95% of the time is excellent if it is reviewed and dangerous if it is not. The difference between the two is how the validation step is designed.
Here the document is proposed: it appears on screen with the extracted data and the original alongside for comparison at a glance. You confirm, correct whatever needs correcting and save. Anything that could not be read is flagged rather than slipping through silently, which is the costliest failure mode of tools like this.
And it learns along the way: when you correct an item reference, that equivalence is linked to the supplier and comes in already filled next time. With regular suppliers, review goes from correcting to confirming within a handful of invoices.
What changes afterwards
The books stop running three weeks behind
The obvious saving is admin hours: keying a twenty-line invoice takes time, and some companies receive two hundred a month.
But what really shows is something else. When entering invoices is slow they pile up; when they pile up, the books run late; and when the books run late, the cash position and margins you look at are three weeks old. Every decision gets made on stale data without anyone saying so out loud.
Once the document is in, the flow continues on its own: the due date goes to the payables ledger, the entry to accounting, and the invoices become available for SII reporting where applicable.
Clearing things up
Frequently asked questions
Answers to what we get asked most.
Do I need a template for each supplier?
No. Extraction identifies each field by what it means rather than where it sits on the page, so it works across invoices from different suppliers with nothing to configure. That is exactly where traditional capture systems fail when a supplier redesigns their invoice.
Which documents can it read?
Purchase invoices, delivery notes and orders, both as PDFs and as photos taken on a phone. From each it extracts the supplier and their tax details, number and date, line items with quantity and price, net amounts, VAT rates, withholdings and due dates.
What if it misreads something?
The document is proposed on screen with the original alongside: you confirm or correct before saving, and anything unreadable is flagged rather than slipping through silently. A system that is right 95% of the time is excellent if it is reviewed and dangerous if it is not.
Does it learn from corrections?
Yes. When you correct an item reference, that equivalence is linked to the supplier and comes in already filled next time. With regular suppliers, review goes from correcting to confirming within a handful of invoices.
Artificial intelligence and OCR
AI reads your invoices and creates the document, and an assistant answers in plain language about the software and your data, respecting each user's permissions.
AI Assistant
An assistant one click away on any screen, answering questions about how the system works by consulting the official documentation, in the user's language.
View moduleAI over your data
Extend the assistant to query your company's real data (sales, customers, stock, work reports, clock-ins, outstanding payments) by asking in plain language.
View moduleSmart FAQs
Semantic search over the FAQ knowledge base: the assistant understands what the user means, not just the exact words they typed.