AI & automationJuly 20266 min read
Extracting your invoices and contracts: why vision models replaced OCR
For years, automating the reading of a document meant stacking character recognition and brittle rules, which broke the first time a supplier changed its layout. That is no longer the case. Here is what changed, and how to tell if it pays off for you.
The real cost of a document read by hand
An invoice re-keyed by hand is a few minutes per document, multiplied by the volume, plus the copy errors you pay for later. Over hundreds of documents a month, that adds up to whole days of invisible work and a constant risk of error. It is that cost, often underestimated, that good extraction removes.
Why OCR plus rules was not enough
The classic approach read the characters then applied fixed rules: the invoice number here, the total there. It worked as long as the layout did not move. At the first supplier that changed its template, or the first slightly different invoice, everything broke. You spent more time patching the rules than processing the documents.
What vision models change
Vision models do not just read the characters: they read the page the way a human would. They find a total, a due date or a clause even if it has moved, on a layout never seen before. The result: an extraction that holds without maintaining hundreds of rules, and that adapts to varied documents.
Does it pay off for you?
The math is simple: your volume of documents multiplied by the time spent on each. Three triggers make the topic urgent: a large volume handled by hand, electronic invoicing becoming mandatory for companies in 2026 which pushes you to structure your flows, and sovereignty when your documents are sensitive, since the processing can run entirely on your own infrastructure.
Do you have a pile of documents still going through manual entry? Send me a few anonymized ones, and in one free call I will tell you whether automating it is worth it, and what it would save.