demeris
From image to text: recognition
It all starts with an image: a scan, a PDF, sometimes a simple photo. Optical character recognition (OCR) converts this image into usable text. This is necessary, but not sufficient: reading characters does not yet reveal where the amount is located, or who the supplier is.
Understanding, not just reading
The next step is to understand the document. The system identifies its type (invoice, delivery note, reminder) and then locates and extracts the relevant fields: supplier, date, number, amounts, VAT rate, IBAN or Swiss QR-bill data.
Unlike a rigid template, modern capture adapts to varied layouts and improves with the documents it processes — essential when every supplier has its own presentation.
Good capture is not measured by what it reads, but by what it knows how to verify.
Verification: the step that really matters
Extracting data is useless if it is wrong. The real work lies in verification: checking the consistency of totals, controlling VAT, reading the structured data of the QR-bill, and matching against a known order or supplier.
When doubt remains, the document is put on hold for human validation rather than being processed as is. This step is what makes the difference between reading fast and reading correctly.
Swiss specifics
In Switzerland, two points help a great deal: the QR-bill provides data that is already structured, and therefore read very reliably; and multilingualism (French, German, Italian) requires capture that is not limited to a single language.
People remain in control
The goal is not to eliminate human oversight, but to focus it where it adds value: exceptions, ambiguous cases, decisions. The rest, being repetitive, is handled automatically.
The best way to get a sense of it is still to test it on your own documents. That is exactly what demeris Capture allows you to do.
