In this talk, Olivier Vasseur (Solution Consultant) and Raphaël Godinot (partner sales manager) from ABBYY issue a warning and offer an alternative approach: using generative AI alone (ChatGPT, Claude) to extract information from critical documents is risky — costly in token consumption, prone to hallucinations, and hard to audit. Instead, they present trusted document AI, combining a specialized, deterministic AI engine (the ABBYY Vantage platform) with LLMs used within a tightly controlled scope.
The talk walks through the document processing pipeline (image enhancement, OCR/ICR, classification, multi-technology extraction, control, and traceability), illustrated with a live demo on a rail freight invoice, followed by a series of concrete examples: energy performance certificates, purchase orders, handwritten technical drawings, CVs, payslips (including document fraud detection), energy savings certificates, logistics documents, and automatic splitting of multi-document PDFs.
The session closes with an overview of ABBYY — 40 years of experience, inventor of the first OCR, a leader in Gartner's ranking for Intelligent Document Processing — and a look at a new document format designed to be efficiently ingested by AI systems.
