OCR Document Recognition (Belegerkennung)
OCR (Optical Character Recognition) technology converts images of text (scanned paper, PDFs, smartphone photos) into machine-readable text. In ERP-adjacent applications, OCR drives the bulk of document automation: invoices, delivery notes, supplier contracts, employee expenses, customer orders. Since 2022, classical OCR has been rapidly displaced by Intelligent Document Processing (IDP) — LLM-augmented extraction that outperforms OCR-plus-regex pipelines on both accuracy and robustness.
Classical OCR versus modern IDP
Classical OCR pipeline: image preprocessing (deskew, denoise), character recognition, post-processing with rule-based field extraction. Tools: ABBYY FineReader, Tesseract (open source), Microsoft Read API. Brittle on layout variations, struggles with handwriting, requires per-vendor template tuning. Modern IDP: deep-learning models for OCR plus document understanding via LLM. Tools: Rossum, Hyperscience, ABBYY FlexiCapture, Microsoft AI Builder, Google Document AI, AWS Textract, Azure Document Intelligence, SAP Document Information Extraction. Handles layout variations, multi-language, handwriting, complex tables. Reaches 90-98% accuracy on common document types. Implementation effort: weeks instead of months. By 2026, IDP is the default; pure OCR remains only in edge cases.
ERP-adjacent applications
- Supplier invoices — the largest single use case; 80-95% straight-through processing in mature deployments
- Customer orders via email or fax — structured order creation in ERP from unstructured input
- Delivery notes from suppliers, auto-matched to POs for three-way match
- Expense receipts — employee expense reports with automatic categorisation
- Contracts — extraction of key terms (parties, dates, amounts, payment terms, termination clauses) for contract-management systems
- ID documents — customer onboarding, KYC processes
- Forms — structured fields from paper or PDF questionnaires
- Quality reports from suppliers and inspection bodies
Leading platforms
Specialist IDP: Rossum (Czech-built, strong DACH adoption), Hyperscience (US, enterprise-focused), ABBYY FlexiCapture (German roots, broad coverage), Klippa, Foxit IDP. Cloud hyperscalers: Microsoft AI Builder, Microsoft Document Intelligence, Google Document AI, AWS Textract. ERP-vendor: SAP Document and Reporting Compliance, Microsoft Dynamics 365 Bill Capture, Oracle Document Automation, NetSuite Bill Capture. AP-specific: Esker, Coupa, Basware, Kofax ReadSoft Online. DACH-specific AP IDP: D.velop process manager, JustOn, ELO, easy software. For mid-market AP automation in DACH, Rossum, ABBYY FlexiCapture and vendor-bundled options are the most-commonly evaluated.
Impact of e-invoicing
The shift to structured e-invoicing (XRechnung, ZUGFeRD) in DACH from 2025 fundamentally changes OCR/IDP economics. Structured invoices skip the extraction step entirely — the data is already structured. As B2B e-invoicing adoption grows through 2025-2028 toward mandate compliance, the share of invoices needing OCR/IDP processing shrinks dramatically. IDP investment in 2026 should anticipate this transition: configure pipelines that handle both unstructured (PDF, paper) and structured (XRechnung, ZUGFeRD) inputs through the same workflow, with the IDP layer absorbing the former while the latter bypass it. Vendor pipelines increasingly handle both natively; legacy OCR-only systems require replacement or significant adaptation.
Related Topics
Frequently Asked Questions
Is OCR still worth investing in given AI document understanding?
OCR-only is no longer competitive. Modern AI-augmented IDP from Rossum, Microsoft, Google and others delivers OCR plus understanding in one platform. New investment should target IDP, not pure OCR. Legacy OCR deployments can usually be upgraded with IDP capabilities through the same vendor or replaced when contracts renew.
How much does IDP cost?
For a 50,000-invoice-per-year AP operation: 30,000-100,000 EUR per year licence cost plus 100,000-300,000 EUR implementation. Cost scales sub-linearly with volume; high-volume operations (250,000+ invoices per year) see per-invoice cost drop below 0.20 EUR.
Can IDP handle handwriting and complex multi-page documents?
Modern IDP handles printed text near-perfectly, machine-printed forms with high accuracy, and reasonable handwriting in structured fields. Free-form handwriting and complex multi-page contracts with cross-references remain harder; human review for these document types is still standard. The capability boundary is moving rapidly with each generation of LLM improvement.