Data stewardship is the practice of assigning people clear responsibility for specific data - its quality, meaning, access, and lifecycle. Tools and policies enforce governance, but governance fails without an accountable human behind each dataset. A data steward is the person who can answer "what does this field mean, who's allowed to use it, and is it correct?" - and who's on the hook when the answer is wrong.
Stewards, owners, and custodians
Governance roles are often confused, and separating them helps. A data owner is typically a business leader accountable for a data domain and its policies - they decide who should have access and what the data is for. A data steward is the hands-on role that manages the data day to day: defining and maintaining its meaning, monitoring quality, and resolving issues. A data custodian is the technical role (often a data engineer) that operates the systems the data lives in. One person can wear more than one hat in a small team, but the responsibilities are distinct: owner sets policy, steward curates meaning and quality, custodian runs the infrastructure.
What a steward actually does
The work is concrete. Stewards maintain the business glossary - the agreed definition of "active customer" or "recognized revenue" so teams stop arguing about whose number is right. They own the documentation of their datasets (descriptions, grain, caveats), set and monitor quality expectations, approve or review access requests, and triage data-quality incidents in their domain. In a data-mesh organization the steward role often merges with the domain team that owns its data products; in a centralized one, stewards may sit in a governance function spanning domains.
Why it's the missing piece
Most governance programs buy a catalog and write policies, then wonder why data quality doesn't improve. The reason is usually that no one is accountable for any specific dataset - governance is everyone's job, so it's no one's. Stewardship fixes that by attaching a name to each important dataset. It's what makes compliance real (someone answers the auditor), what keeps lineage and documentation current, and what turns a data catalog from a stale inventory into a trusted, maintained resource. The tooling enables stewardship; it doesn't replace it.
Frequently asked questions
What's the difference between a data owner and a data steward? An owner (usually a business leader) is accountable for a domain and sets policy; a steward does the hands-on work of defining meaning, monitoring quality, and resolving issues.
What does a data steward maintain day to day? The business glossary and documentation, quality expectations, access-request review, and incident triage for their datasets.
How does stewardship work in a data mesh? The steward role typically merges into the domain team that owns its data products, rather than sitting in a separate central function.
Why isn't a data catalog enough on its own? A catalog is an inventory; without an accountable steward per dataset, documentation goes stale and quality issues go unowned. Stewardship is what keeps the catalog trustworthy.