What is a data format?
A data format is the agreed arrangement of characters or bytes that tells software how to read a file or message. CSV holds rows of comma-separated values with no type information. JSON nests labeled fields and lists, which suits API messages. XML does the same with more ceremony and appears in older integrations. Parquet stores tables by column with compression and an embedded schema, which makes it the usual choice for analytics on a data lake. PDF and image formats hold documents that software can display but cannot query without optical character recognition.
Industries add their own. Supply chains exchange EDI documents such as the 856 ship notice, healthcare systems exchange HL7 messages and FHIR resources, and banks still move fixed-width files whose layouts have not changed in decades.
Why data format matters
Format decides how much work stands between a source and an AI model. A language model reads JSON or plain text directly, a Parquet table needs a query engine, and a scanned invoice needs an OCR step first. A logistics provider that receives ship notices as EDI from large customers, as spreadsheets from small ones, and as PDFs from a few converts all three into one JSON structure, usually defined by a canonical schema, before any downstream system touches them.
What to watch for
- Conversions lose detail silently, especially dates without time zones, decimals rounded to floating point, and text in the wrong character encoding.
- Proprietary formats tie the data to one vendor’s tools, so ask how to export in an open format before signing.
- A format carries little meaning on its own, so a JSON field named status still needs a documented list of allowed values.