What is data classification?
Data classification is the practice of assigning each dataset, document, or field a sensitivity label from a small, agreed scheme. A common scheme has four levels: public, internal, confidential, and restricted. The label drives rules: where the data may be stored, who may read it, whether it may leave the company, and whether it may be sent to an AI model. Labels are applied by hand when a document is created, inferred by scanning tools that recognize patterns such as account numbers, or inherited from the source system.
Data loss prevention, access control, and retention policies all depend on it. A tool cannot block the export of confidential data until something has said which data is confidential.
Why data classification matters
AI projects force a classification question many companies have postponed. Before a retrieval index is built over a shared drive, someone must decide which folders a model may read and which it may never see. Before an agent gets a tool that queries the customer database, someone must say which columns are restricted. Without labels, whoever builds the integration makes those decisions ad hoc.
A law firm classifies its document management system by matter and sensitivity before indexing it for a research assistant. Documents behind an ethical wall are labeled restricted and excluded from the index. Confidential client documents are indexed but served only through permission-aware retrieval that checks the user’s matter access on each query. The firm’s data catalog records the label on every collection, so the next project starts from known ground.