AI glossary · Beginner · Classic ML, recsys & search · Data engineering
Structured data
Structured data is data organized in a table with defined rows and columns. Like an Excel sheet or a database table, what kind of value goes in each cell is decided in advance.
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Basics
- Added
- 2026-10-09
How it works
In structured data, each row represents one entity (one customer, one transaction), and each column represents a feature of that entity (age, amount, region). Each column has a set format such as number, date or category, so computers can calculate and search it directly. Relational databases and SQL are the main tools for handling this kind of data. By contrast, data without fixed fields, such as documents, images and audio, is called unstructured data.
Structured vs. unstructured data
Structured dataTables with defined rows and columns
- Excel & database tables
- Calculate & search directly with SQL
- Gradient boosting tends to be strong
Unstructured dataNo fixed fields
- Documents, images, audio
- Handled by converting to embeddings, etc.
- Mainly processed by deep learning & LLMs
공통 Both are material AI learns from and makes decisions with.
Why it matters
A large share of real business data, such as company sales, inventory, customer and sensor records, is structured data. Tasks such as loan screening, demand forecasting and fraud detection mostly make decisions by looking at these tables. For tabular data, traditional machine learning methods such as gradient boosting are known to often still outperform deep learning. Recently, there are also more efforts to connect structured data with generative AI, such as Text-to-SQL, in which an LLM converts a person's question into SQL to query tables.
Good to know
Because LLMs read text in units of tokens, if you feed in a large table whole, they can be weak at calculating numbers accurately or comparing many rows. So a common approach is to leave calculations to the database or code and have the LLM handle interpreting questions and explaining results. Also, even structured data often has quality problems such as empty cells, unit mix-ups and duplicate rows, so cleanup is needed first.
Example
If you ask an internal data analysis chatbot "What were the top 3 regions by sales last quarter?", the LLM converts the question into SQL, queries the sales table (structured data), and explains the result in sentences. A bank's fraud detection system picks out suspicious transactions using a model trained on a transaction table containing amounts, times and locations.