Häufig gestellte Fragen
How accurate is OCR document recognition in practice?
Modern AI-assisted systems often achieve extraction accuracies of over 95 percent on documents of good quality, while 80 to 90 percent is more realistic for poorly scanned documents or thermal receipts. Accuracy depends heavily on scan quality, document format and the quality of the configured validation and plausibility rules. With recurring documents from the same supplier, the hit rate often rises noticeably because learning methods adapt to familiar layouts. A validation layer with manual review of uncertain fields nevertheless remains an integral part of a robust process.
What distinguishes template-based OCR from AI- or ML-based document recognition?
Template-based methods expect fixed field positions per supplier template and work reliably as long as the layout does not change, but quickly reach their limits with new or deviating document formats. AI- or ML-based approaches, often referred to as intelligent document processing (IDP), recognise relevant fields largely independently of the layout and can also process documents from unknown suppliers. In practice, many solutions combine both methods to cover standard documents as well as special cases. Which approach fits better depends on the variety of incoming document formats and the desired maintenance effort.
Does OCR document recognition also work for e-invoices such as XRechnung and ZUGFeRD?
With structured electronic invoices such as XRechnung or ZUGFeRD, the invoice data is already available in machine-readable form as XML, so optical recognition is essentially unnecessary and the data can be read in directly. This is generally more accurate and less error-prone than image processing of a scanned document. In Germany, companies have been required since 1 January 2025 to be able to receive e-invoices in the B2B sector, while the obligation to send them takes effect in stages — from 2027 for companies with prior-year turnover above 800,000 euros and from 2028 for all others. However, OCR remains relevant as long as documents arrive as scans or image PDFs, for example from smaller suppliers or from abroad.
What does OCR document recognition mean for GoBD compliance?
Under the GoBD, tax-relevant documents must be retained completely, traceably and unalterably, with the retention period for accounting records such as invoices having been shortened from ten to eight years by the Fourth Bureaucracy Relief Act, in effect since the beginning of 2025. When paper documents are scanned to replace the originals, procedural documentation is mandatory, describing the entire process from incoming mail through capture to audit-proof archiving. If full-text or field data is generated via OCR, this verified information should likewise remain available unchanged throughout the retention period. OCR recognition itself therefore replaces neither proper archiving nor the documented control of the capture process.
Why do documents still need manual review despite high recognition rates?
A high field recognition rate and the so-called touchless processing rate are two different metrics: even if a system reads individual fields correctly 95 percent of the time, the share of fully automatically posted documents can be significantly lower, because deviations trigger a clarification workflow. Fields with low recognition confidence and documents without a clear purchase order or goods receipt reference are flagged for manual review. This concentrates the effort on doubtful cases, while standard invoices that match on supplier, quantity, price and tax rate can largely pass through automatically. The achievable degree of automation therefore depends on document quality, master data and the depth of ERP integration.
How much does OCR document recognition cost?
Cloud solutions for smaller volumes often start at a few dozen euros per month, while enterprise models are sometimes billed by document volume, and implementation can reach the low to mid five-figure range depending on complexity. The pure licence or usage fees usually account for only part of the total cost; a substantial share goes to integration, customising, training and the connection to accounting and archiving systems. Cloud models lower the initial investment but permanently shift costs into ongoing operating expenses. Reliable figures only emerge from the specific document volume, the diversity of suppliers and the existing system landscape.