Analyze documents with AI: Data out, no retyping

A first document type including the handover into your system usually starts at around €6,000, and larger projects covering several kinds of paperwork land between €10,000 and €35,000. The number moves with the count of document types, the number of fields per type and whether the processing runs locally, in the EU cloud or as a mix of both. After the first conversation and a look at your documents, you get a fixed price per stage.

That depends on the document, so we measure it: during the trial run we count, on your own paperwork, how often a field is recognized correctly. Clean digital invoices are no trouble. Poor scans and handwritten additions are a different story. For every field we set the confidence level at which a value is imported directly and the point at which a person takes a look.

The AI makes a proposal and stops there. Every value comes with an estimate of how sure the model is, together with the spot in the document it was taken from. Your team sees both side by side and corrects with one click. Only what has been approved moves into your system.

Yes. A scan or a photo carries no text layer to read from, so text recognition runs in front of the analysis. The quality of the original counts for a lot: a sheet scanned straight reads reliably, a skewed phone photo with shadows across it much less so. The trial run shows you, on your own files, where the limit sits.

Only if you want them to. We have built an analysis where the file and the processing stay entirely inside the employee's browser, with no upload at all. Where that isn't enough, we use models on servers in the EU. Which documents take which route is something we settle together beforehand and put in writing.

Analysis in the browser is the simplest case: no transfer takes place, because the file never leaves the machine. On the EU route you need a data processing agreement with the provider, and we tell you which provider that is and what it commits to. What gets stored at the end is your call: the fields on their own, or the extracted text alongside them as evidence.

For everyday volumes at a mid-sized company, quantity is rarely the constraint, because the analysis runs per document and can be parallelized. The workflow around it matters more: do the documents arrive by email, from a folder or straight off the scanner, and who looks at the uncertain cases? We settle that during the concept phase, because the payoff hangs on it.

No. The analysis sits in front of what you already run and hands the reviewed data to accounting, inventory or your archive. Where a system has an API, we use it. If there is none, we talk about a file import or a connection we build ourselves.

With the document type that arrives most often and gets typed out most mindlessly. Usually that means incoming invoices. One type, read cleanly and wired into your system, tends to save more time than three half-finished ones. The next ones follow in whatever order pays off.