Your registration forms arrive filled out by hand. Someone has to key them into the CRM, one field at a time — long and tedious work. AI can read handwriting today, but the real question isn't "can it read?", it's "how do we know it read correctly?".
Reading is the easy part
Photographed or scanned, a handwritten form is read well by a vision model today: name, address, email, date of birth, checked boxes. On neat handwriting, the result is excellent.
The problem lies elsewhere. The 1 that looks like a 7, the O mistaken for a 0, an unusual last name, a crossed-out field: faced with those, the AI doesn't stop. It offers an answer, with exactly the same assurance as for everything else. And a silent error in a CRM costs more than a field left blank.
Read it twice, with two different AIs
Our approach: submit the same form to two separate AIs — two models from two different providers, each working on its own — then compare their answers, field by field.
One model getting it wrong is predictable. Two models designed and trained differently getting it wrong the same way, on the same field, is far rarer. The comparison turns a confident answer into a measure of confidence.
They agree: we write. They differ: a human decides
When both readings match on a field, the data goes straight into the CRM.
When they differ — "Tremblay" versus "Trembley" — the form is held for review: we show the image of the field in question alongside the two candidates, and one click settles it. Nobody re-reads forty fields; you arbitrate two.
In practice, most forms go through untouched, and human attention lands exactly where there is doubt.
What this really changes
You aren't replacing manual data entry with an AI you have to take at its word. You're replacing it with a process that can tell you when it isn't sure — and that is precisely what makes it deployable.
The error rate stops being invisible: it becomes measurable. This is a good example of AI-assisted automation: the path stays yours, and the AI settles only one specific question, under supervision.
Do you have paper forms that end up being keyed in by hand? Let's look at what this would give you. Let's talk.
— Frédéric Brabant