Normalization organizes data to reduce redundancy and improve integrity. The goal is to store each fact once. If a customer's address appears in every order record, an address change must be updated in hundreds of places. Miss one and the data is inconsistent. Normalization moves the address to a customer table. Orders reference the customer by ID. The address is stored once. Change it there and every order reflects the new value.

The process follows normal forms. First normal form eliminates repeating groups. Second normal form removes partial dependencies. Third normal form removes transitive dependencies. Each form adds constraints that reduce redundancy further. Higher forms exist but are rarely used in practice. Most transactional databases aim for third normal form. Normalization has trade-offs. More tables mean more joins, which can slow queries. For transactional systems, the integrity benefits outweigh the performance cost. For analytical systems, the opposite is often true. Denormalized schemas with wide tables and redundant columns query faster because they avoid joins. That is why data warehouses use star and snowflake schemas that deliberately denormalize. The right level of normalization depends on the workload. Transactional systems normalize. Analytical systems denormalize. Both are correct for their purpose.

Normal forms

• First normal form — no repeating groups
• Second normal form — no partial dependencies
• Third normal form — no transitive dependencies
• Boyce-Codd normal form — stricter version of third

Normalization is about integrity. Store each fact once and the data stays consistent. Store it many times and inconsistency is inevitable.