Data product

A data product is a dataset, API, or feature built and maintained with an accountable owner, a defined interface, documentation, and quality expectations, treated less like a byproduct of some other process and more like a product with real users. Examples include a curated "customer orders" table with a stable schema, or a feature-serving API for a machine learning model.

The concept sits at the center of data mesh, where each domain team is expected to publish its data as one or more discoverable, trustworthy data products rather than exposing raw internal tables. A data product typically comes with an explicit data contract describing its schema and guarantees, and is registered in a data catalog so other teams can find and evaluate it before depending on it.

The distinction from a plain dataset or table is ownership and intent: a data product has someone accountable for its uptime, correctness and evolution, and is designed for reuse by people the owning team may never meet, whereas an ordinary internal table is often a side effect of one pipeline with no such commitment. This matters because it shifts data quality left, onto the producer, instead of leaving every downstream consumer to independently clean and reconcile the same raw data, though the discipline it requires, ownership, documentation, support, is easy to claim and hard to sustain in practice.