The four principles

  1. Domain-oriented ownership — data is owned by the domain team that produces it (orders team owns the orders dataset), not a central data team.
  2. Data as a product — each domain exposes data with SLAs, documentation, and quality guarantees. Consumers can rely on it like an API.
  3. Self-serve data infrastructure — a platform team provides tooling so domains can publish and consume data without infra work.
  4. Federated computational governance — global policies (PII handling, retention) are automated and enforced at the platform level, not by manual review.

When data mesh makes sense

Data mesh solves the scaling problem of a central data team that becomes a bottleneck as the organisation grows. It is not appropriate for small organisations — the overhead is not worth it below ~5 domain teams.

Implementation starting point

Start with principle 1: assign dataset ownership to domain teams and hold them accountable for quality. Everything else follows from ownership.