Document AI Glossary
Key terms in intelligent document processing, document fraud prevention, and compliance, explained by the AITECA team
AITECA Platform and Products
The building blocks and technologies of the AITECA platform
| AITECA |
A modular, sovereign document AI platform developed in France by ITESOFT Innovation. It reads, verifies, and secures the documents in your customer journeys. |
|
Add-on |
A capability you activate on top of the IDP Core or a business module, with no integration break. |
| Altermetry® |
AITECA's proprietary technology that detects altered images and PDFs, such as a modified name, date, or amount. |
| Base M+ |
A shared document fingerprint database, exclusive to ITESOFT Innovation. It spots the same document submitted multiple times to detect duplicate claims, inflated damages, and organized fraud. |
| Private base |
A training base dedicated to your specific documents, deployed privately and managed by our specialists or by your own teams. |
| Zero-shot learning |
Zero-shot learning (ZSL) is an AI method that enables a model to recognize or classify objects, concepts, or tasks it never saw during training. |
|
Pre-trained library |
A catalog of more than 100 ready-to-use document types, organized into 8 families: civil status, income, proof of address, identity, banking, health, business, and administrative. You start with no training phase. |
| Case Module |
A business module that automates the collection, assembly, and completeness of customer and user case files. Case handlers receive a complete, compliant file. |
| Fraud Module |
A business module that assigns a trust score to documents and data, centralizes alerts, and equips teams to investigate suspected fraud. |
| Orchestrator |
An engine that selects the processing strategies and combines the AI technologies and checks best suited to each document, with no dependence on a single technology. |
| IDP Core |
The common foundation of the AITECA platform: omnichannel capture, document typing, extraction, consistency checks, and REST API. |
| Verify |
An add-on for advanced checks: cross-checks, calls to external reference databases, and document anomaly detection. |
| Vision+ |
A visual-understanding add-on that reads complex, degraded, or atypical documents: tables, handwriting, claim photos, quotes. |
Document Processing and AI
Key concepts in intelligent document processing
| API-first |
A product designed from the ground up to integrate via API. Integrating AITECA with your information system takes less than 24 hours. |
|
Omnichannel Capture |
Receiving documents through any channel: mail, email, web portal, mobile app, or branch. |
| Extraction |
Reading and structuring the useful data in a document, such as a name, an amount, or a date, to feed your information system. |
| Lean AI |
An approach that activates, for each need, the lightest model able to deliver the expected performance, to limit energy consumption. |
| Symbolic AI |
AI based on explicit rules, also known as an expert system. It is explainable and uses few resources. |
| ICR |
Intelligent Character Recognition: recognition of handwritten text. |
| OCR | Optical Character Recognition, ou reconnaissance optique de caractères. Il transforme l'image d'un texte dactylographié en texte exploitable. |
| IDP |
Intelligent Document Processing. It automates the capture, reading, and verification of documents by combining multiple AI technologies. |
| ADR (LAD) |
Known in French as lecture automatique de documents. This metric measures the share of fields extracted correctly. |
| LLM |
Large Language Model. At AITECA, LLMs are hosted in-house, with no calls to external APIs. |
| MCP |
Model Context Protocol: a standard for connecting AI agents to tools and data sources. |
| NER |
Named Entity Recognition: identifying names, dates, places, and amounts in a text. |
| NLP | Natural Language Processing |
| Omnichannel journey |
A consistent customer journey, whatever channel is used to submit or complete a case file. |
| RAD |
Known in French as reconnaissance automatique de documents. This metric measures the share of documents identified correctly. |
| Stateless service |
A service that retains no data after processing: your documents are used only for the duration of the processing. |
| Document typing |
Automatic identification of a document's type: paystub, bank account details, tax notice, and so on. |
| Manual keying (video coding) |
Manual entry or correction, by an operator, of data that a system could not read with certainty. |
| VLM |
Vision Language Model: a model able to understand an image and its context. AITECA uses it, for example, to analyze claim photos in 4 seconds. |
| Human-in-the-loop |
A way of designing and running AI and machine learning systems in which people step in at key points: the AI proposes, and a person validates or corrects. |
| Human-over-the-loop |
The AI operates on its own at scale. People no longer handle individual cases; they act upstream (design, rules) or downstream (audit, maintenance). |
Fraud and checks
The vocabulary of document fraud detection and investigation
| 2D-DOC |
A secure barcode printed on many official French supporting documents, such as tax notices and energy bills. Reading it verifies the document's authenticity. |
|
Fraud investigation unit |
A team dedicated to detecting and investigating fraud within an insurer, a health insurer, or a bank. In US insurance, it is often called a Special Investigations Unit (SIU). |
|
Cross-check
|
Comparing the same data across several documents or sources, for example the name on an ID card and on bank account details. |
|
Consistency check |
Machine Readable Zone : zone de lecture automatique située au bas des pièces d'identité et des passeports. |
|
Proactive detection |
Detecting fraud at the start of the journey, before any processing or payment. |
|
AI-generated document |
A fake document produced by generative AI, increasingly hard to spot with the naked eye. |
|
Face matching |
Comparing the face in a selfie with the one on an ID document to validate a customer's identity. |
|
Document tampering |
Modifying an authentic document, for example a retouched amount or date. Not to be confused with a fake document, which is fabricated from scratch. |
|
Document fraud |
Using a fake, tampered, misused, or AI-generated document to obtain a right, a payment, or a service. |
|
Fraud investigation |
Analyzing a suspected fraud case, from alert qualification through to the decision. |
|
Fraud prevention (LAF) |
Known in French as lutte anti-fraude. |
|
Anti-money laundering (AML) |
refers to the comprehensive set of laws, regulations, and procedures designed to prevent criminals from disguising illegally obtained funds ("dirty money") as legitimate income ("clean money") |
|
MRZ |
Machine Readable Zone: the automatic reading zone at the bottom of ID cards and passports. |
|
External reference database |
A third-party database queried to verify information: SEPAmail for bank details, FINESS or RPPS for French healthcare professionals, asset freeze lists. |
|
Trust scoring |
A reliability score calculated on a document or a piece of data, which helps prioritize the cases to review. |
|
Shift-left |
Users correct their errors as they submit their documents. No more back-and-forth or manual follow-ups. |
|
Weak signal |
An isolated clue that, cross-checked with others, reveals fraud. |
Compliance, Regulation, and Sovereignty
Key regulatory frameworks and security concepts
| AI Act |
The European regulation on artificial intelligence, which governs AI uses according to their level of risk. |
|
Asset freeze |
Official lists of individuals and entities subject to financial sanctions. |
|
ISO 27001:2022 |
The international standard for information security management. AITECA is ISO 27001:2022 certified. |
| KYC |
Know Your Customer: the identity checks a company performs when onboarding a new customer. |
| AML/CFT |
Anti-money laundering and countering the financing of terrorism (LCB-FT in French). |
| PEP |
Politically Exposed Person, subject to enhanced due diligence. |
| 100 % Santé reform |
A French healthcare reform that eliminates out-of-pocket costs for a range of optical, dental, and hearing care. |
| RGAA 4.0 |
The French government's digital accessibility standard, based on WCAG 2.1. The AITECA website and service comply with it. |
| GDPR |
General Data Protection Regulation: the EU regulation on personal data protection (RGPD in French). |
| SecNumCloud |
A qualification issued by ANSSI, the French national cybersecurity agency, to cloud offerings that guarantee a high level of security and protection from extraterritorial laws. AITECA offers SecNumCloud hosting. |
| Digital Sovereignty |
Control over the applicable law, technologies, teams, and capital, beyond the hosting location alone. |
| Vendor lock-in |
Dependence on a technology or vendor that is hard to replace. AITECA's modular architecture is designed to avoid it. |