PDF and invoice data extraction

Problem: manual data entry from PDF invoices (suppliers, amounts, dates, line items) occupied 2 FTEs full-time with a 5% error rate. Solution: automatic extraction pipeline (OCR + Mistral for structured understanding) deployed sovereignly, with human validation for ambiguous cases. Result: 90% of invoices processed automatically, error rate reduced to 0.3%, 1.5 FTE saved.