Invoices and documents · Quebec SMBs
Every document that arrives as a PDF, a photo or an email attachment ends up being read and re-typed by someone. AI can now read these documents reliably: it extracts the fields, checks them against your rules and sends them into your software. Your team only handles the exceptions.

Gabriel Nadon, independent consultant in Montreal. You talk to the person who builds it.
The signs
Document data entry doesn’t appear on any financial statement. It hides in payroll, in billing delays and in errors discovered too late.
How it works
A document goes through four steps: it is captured (email, shared folder, scan), read and structured by AI, checked against rules specific to your business — known supplier, totals that add up, matching order — then entered into your system. Anything that fails the checks lands in a review queue, with the reason.
Before any commitment, we run a sample of your invoices through and measure what comes out correctly.
A total that doesn’t add up, an unknown supplier, an unusual price: the document is held back and flagged.
QuickBooks, Acomba, Sage, your ERP or Excel — through import, API or export, depending on what the software allows.
Which documents, how many per month, from how many suppliers, and where they need to end up.
Around fifty real documents go through the system. We see exactly what is reliable and what isn’t.
Together we write the checks that matter: thresholds, matching, approved suppliers.
Validated data is entered automatically; exceptions land in a queue that’s simple to work through.
We look at what falls into exceptions and improve the rules until the queue becomes marginal.
OCR turns an image into text. AI understands that text: it knows a number is a pre-tax total, a line is a product, a document is a credit note. That’s what makes it possible to read different formats without a template per supplier.
It depends on your documents: scan quality, variety of formats, handwriting. That’s why we start with a test on your real documents, and why rules hold back anything doubtful instead of entering it.
Yes, depending on the access the software offers: API, file import or structured export. We confirm the integration path during the review, before any commitment.
They are read when possible, and sent for review when the reading isn’t reliable enough. The goal isn’t zero human involvement, it’s zero pointless data entry.
Not if it’s governed: we choose providers and settings compatible with your obligations (Quebec’s Law 25), limit access and keep only what’s necessary.
Free review
Tell me which ones and roughly how many. I’ll reply within 24 hours with what can be automated and an honest idea of the effort.