LAD / RAD (automatic document reading and recognition)

LAD (Lecture Automatique de Documents, "automatic document reading") and RAD (Reconnaissance Automatique de Documents, "automatic document recognition") are French acronyms for software that classifies incoming documents and extracts data from them without manual keying. Banks, insurers, and public administrations in France have used LAD/RAD systems since the 1990s to process mail, claims, and forms at scale. The terms are largely local equivalents of what English-speaking markets call OCR-based document capture, a subset of what is now broadly named Intelligent Document Processing (IDP).

How it works

A traditional LAD/RAD pipeline runs in two stages, matching the names. RAD comes first: the system looks at an incoming scan or PDF and decides what kind of document it is (invoice, claim form, ID card, contract) using layout templates, keyword zones, or barcodes. LAD comes second: once the document type is known, the system reads specific zones, using OCR for printed text and ICR for handwriting, and pulls out the values that matter (amounts, dates, names, reference numbers).

This template-driven approach worked well when documents were standardized: a fixed insurance form, a structured tax notice, a bank's own account-opening slip. It breaks down when layouts vary, because the software has no real zone to look at. New suppliers, new form versions, or handwritten annotations force teams to build and maintain separate templates for every variant, which is expensive and slow.

Modern IDP replaces the rigid classify-then-extract sequence with models that read a document more like a person would: understanding context, tables, and free text regardless of exact position on the page. Classification and extraction often happen together, guided by meaning rather than coordinates. The result handles unseen layouts and mixed-quality scans that older LAD/RAD engines reject or misread, without a template rebuild for every new document variant.

Why it matters for document processing

Organizations still running legacy LAD/RAD platforms usually carry a hidden cost: a template library that someone has to maintain every time a partner, supplier, or regulator changes a form. That maintenance burden grows with volume and document variety, which is exactly the direction insurance claims, KYC files, and e-invoicing flows have been heading.

Replacing LAD/RAD with model-based IDP does not just modernize vocabulary. It changes the economics: fewer templates to build, faster onboarding of new document types, and extraction that keeps working when a supplier tweaks their invoice layout. For regulated sectors, this also means audit trails and confidence scores that fit modern compliance requirements more naturally than rule-based systems designed decades ago. Teams evaluating a move should compare total cost and accuracy over time, not just per-document processing speed on day one.

Related terms

• OCR (Optical Character Recognition)
• ICR (Intelligent Character Recognition)
• Intelligent Document Processing (IDP)
• OCR vs VLM
• E-invoicing France

FAQ

Are LAD and RAD the same thing?

No. RAD (recognition) handles document classification: identifying the document type before any data is read. LAD (reading) handles the extraction step that follows, pulling specific values from known zones. In practice, vendors and buyers often use the two terms together or interchangeably to describe the whole capture pipeline.

Is LAD/RAD still relevant, or should we move to IDP?

The concepts still describe real needs: classify, then extract. But the underlying technology behind most LAD/RAD platforms is template-based OCR, which struggles with document variation. If your document types change often or arrive with inconsistent layouts, model-based IDP will generally handle them with less ongoing maintenance.