OCR

What is OCR?

OCR (Optical Character Recognition) is a technology for automatically recognising text from images — scanned documents, photographs, PDF files (without a machine-readable text layer). OCR analyses pixels, identifies individual characters, and assembles them into text that can be further processed, searched, copied, or sent to other systems.

Distinctions:

• Classic OCR — character recognition (Tesseract, ABBYY)
• IDP (Intelligent Document Processing) — OCR + AI layout understanding, also understands document structure (tables, headers, footers)
• ICR (Intelligent Character Recognition) — recognition of handwritten text

Modern deep learning-based OCR achieves 98–99% accuracy on good-quality printed text. Problems arise with: skewed scans, stains, illegible stamps, tables without clear lines — this is where AI/RAG combination helps.

In B2B companies, OCR is most useful for:

• Incoming PDF invoices — extracting company registration number, amount, VAT, due date
• Contracts — full-text search across archives
• Expense receipts — reading paper receipts

When it is used

OCR is the entry point to virtually every AI document automation flow. Without OCR, an AI model could not process a PDF invoice or a scanned contract.

See the Document Extraction module and the Files module.

Related terms

• Document Extraction — the end-to-end process of OCR + AI extraction. See /en/glossary/vytazovanie-dokladov-pojem.
• e-Invoice — the alternative where OCR is not needed. See /en/glossary/e-invoice.
• RAG — after OCR, documents are typically indexed into a RAG system. See /en/glossary/rag.

In Modulario

The Document Extraction module in Modulario combines OCR with AI extraction — a PDF invoice arrives by email, the system OCRs it, an AI model extracts all relevant fields, and automatically creates a received invoice record in Accounting.

Modulario uses a layered architecture: for machine-readable PDFs text is extracted directly; for scanned documents or layered PDFs an OCR engine trained on various European character sets with diacritics is used. Extraction accuracy for common invoice fields (company registration number, amount, date) in Modulario is around 98%, with ambiguous cases flagged for manual review.