Structured and Unstructured Data

A customer support team shows both kinds side by side. The ticket system holds structured fields: ticket ID, customer, priority, opened and closed dates. Those are easy to count, filter and chart with ordinary database queries. The conversation itself, the customer’s email, an attached screenshot and a call recording, is unstructured. Before language models, most software could store that material but not really read it.

Large language models changed the economics. They can summarize, classify and extract fields from text, and multimodal models handle images and audio too. That is why so much enterprise AI work now starts with documents: chunking PDFs, embedding them and retrieving the right passages for RAG. Structured output runs the other way, turning a model’s free text into fields a database can store.

The distinction also decides where data lives. Structured data usually sits in databases and a data warehouse; unstructured files usually land in object storage or a data lake. Answering a real business question often needs both, such as joining a contract’s wording with the account’s billing history.

In episode 16 at 4:50, DevRev’s Manoj Agarwal describes company knowledge as “the structured data, unstructured data and so on,” with structured records such as those in a Salesforce Service Cloud or Zendesk support desk. Episode 16 is a Season 2 episode sponsored by Galileo; its co-host is Galileo COO Yash Sheth (intro at 0:24).

On this site, “structured data” can also mean schema.org markup for search engines. That is a different use of the phrase, so only the unstructured and semi-structured forms link here automatically.