Document AI Glossary

These are reference definitions for the vocabulary used in document and content AI. Each entry explains what a term means, how it differs from the terms it is most often confused with, and what tends to go wrong in practice.

The definitions are written to be vendor-neutral. Where the industry disagrees about what a term covers, the entry says so rather than picking a side.

Capture and understanding

• Optical Character Recognition (OCR) - converting images of text into machine-readable characters, and why that is not the same as structured data.
• Intelligent Document Processing (IDP) - the pipeline that turns unstructured documents into validated, structured data.
• Document Classification - deciding what a document is, so the system knows what to do with it.
• Document Data Extraction - pulling specific fields, tables and entities out of a document.
• Confidence Score - what a model’s certainty value does and does not tell you.

Managing and governing content

• Document Management System (DMS) - versioning, access control, retention and audit for documents.
• Content Services Platform (CSP) - the analyst category that succeeded ECM, and what changed with it.
• Records Management - retention schedules, legal holds and defensible disposition.

Acting on content

• Document Workflow Automation - routing, approvals, exceptions and straight-through processing.
• Retrieval-Augmented Generation (RAG) - grounding generated answers in your own content, with access control enforced at retrieval time.

How to read these entries

Every entry opens with a one-sentence definition that stands on its own, then expands on the mechanics, the boundaries against adjacent terms, and the failure modes that show up once a system is in production. Entries cross-reference each other where the concepts genuinely connect, because most of these terms only make sense in relation to the others.

Terminology in this field shifts, and vendor marketing shifts it faster than practitioners do. Where a term has been redefined over time, such as the move from ECM to content services, the entry records the older usage as well so that older documentation still makes sense.