A data product is a curated, self-contained dataset or data service that is owned, documented, and maintained so that others can use it reliably. The idea is to treat data with the same discipline as a software product, complete with a clear owner, quality commitments, and a defined way to consume it. The concept is central to data mesh.
A well-formed data product includes documentation, service-level expectations, and a stable interface, so consumers do not have to understand the raw source to use it. For example, a payments team might publish a transactions data product that other teams can query without knowing how the underlying systems work.
A data product defines ownership and packaging for a dataset. Whether that dataset is usable and reproducible inside a specific AI run remains a separate readiness question.