What is a data dictionary?
A data dictionary is a centralized description of the data elements in a database, application, or data platform, written so that a person can understand what each field means and how to use it. A schema says a column named DLVRY_DT holds a date. The data dictionary says it is the date the carrier reported delivery to the consignee, that it stays blank until the proof of delivery is scanned, that it differs from the promised date in another column, and that logistics operations owns the definition. A good dictionary also lists allowed codes, the source system, and when the definition last changed. Today the dictionary is usually part of a data catalog, beside automatically harvested technical metadata.
Why a data dictionary matters
Shared definitions prevent the quiet errors caused by reasonable but different interpretations. One logistics company found that three teams measured on-time delivery against three different date fields, each defensible, none documented. A dictionary ended that by stating which date is the delivery date. The same document makes AI assistants more reliable. Given the dictionary entry for each column alongside the schema, a language model’s text-to-SQL answers improve, because it no longer has to infer what an abbreviated name means.
What to watch for
- Assign an owner to each definition, since an unowned dictionary falls out of date within a year.
- Write definitions for business readers and include examples of allowed values.
- Store the dictionary where the data lives, so the semantic layer and AI tools can read it.