What exactly does OCR do?
OCR, optical character recognition, converts an image of text into machine-readable text. It makes a scan searchable and lets you copy text. What OCR does not know is what that text means: which amount is the total, which line is an item, and whether the VAT number belongs to the supplier. One example is ABBYY FineReader PDF, desktop software to convert PDFs and scans and make them searchable. For archiving paper that is enough; for processing invoices it is not.
Template OCR and rules: fine as long as the layout stays the same
Template OCR reads fixed zones of a document: the invoice number top right, the total at the bottom. That works until a supplier changes its template. Microsoft describes the difference clearly: a template model needs a model per variation, a neural model uses a single model for all variations. Docparser is an example of the rules approach: you set up parsing rules per layout. With ten fixed suppliers that is fine; with two hundred different senders, maintenance becomes the job.
What does AI document processing (IDP) add?
IDP does four things OCR does not. It recognises what kind of document it is, extracts the fields regardless of layout, validates the extracted data, and indicates per field how confident it is. Anything below a threshold goes to a person for review, which Microsoft and UiPath call human-in-the-loop. The result is not a text file but structured data your system can take in.
Which IDP solutions are there?
Rossum calls itself AI-first and works without templates, with default hosting in Frankfurt. Parseur combines an AI engine with templates for fixed layouts, on servers in the EU. ABBYY Vantage is a platform with pre-trained models for more than 150 use cases; according to ABBYY its CMR skill is still a preview and does not support Dutch. UiPath offers document processing within its automation platform, with both templates and AI models. Doxis AI.dp, until March 2026 Klippa DocHorizon from Groningen, reads invoices, CMRs and delivery notes among others without templates.
Where does Dottle fit?
Dottle by Bonsai reads unstructured documents and emails, from packing lists and CMRs to invoices and orders, turns them into validated data and writes it straight into your system, such as Exact, SAP, Business Central or AFAS. No fixed format is needed: nobody has to set up a new sender in advance. Doubtful cases go to a review screen where an employee confirms or corrects. At HDG, one of the largest fresh-produce inspectors in the Netherlands, it processes more than 200,000 packing lists a year, with 90% less manual work. You pay per document, from 18.5 cents, and it is live within four weeks. If you only need searchable PDFs, an OCR tool is enough. If you have a few fixed layouts and low volume, Parseur or Docparser are cheaper.
How do you choose?
Five questions decide it. How many different senders and layouts do you have? How many documents per month? Does the data need to land in your ERP or TMS, or is an export enough? Where is the data stored, and does it need to stay in the EU? And who reviews the doubtful cases? Test every solution with a stack of real documents from your own suppliers, including the bad scans.