The centralised data platform model — a central data engineering team responsible for ingesting, transforming, and serving data from across the enterprise — worked well when organisations had a handful of data sources and a modest analytics workload. At enterprise scale, with hundreds of data sources, dozens of consumer teams, and a central team that becomes a bottleneck for every data request, it breaks down.
Data mesh, a paradigm articulated by Zhamak Dehghani, proposes a different organisational and architectural principle: distribute data ownership to the domain teams that create and understand each dataset, hold them accountable for data quality and accessibility as a product, and provide a federated governance layer that ensures global standards without creating a central bottleneck.
The four principles of data mesh are domain ownership (each domain team owns and serves the data it produces as a product), data as a product (data products have defined consumers, SLAs, documentation, and quality guarantees — just like software products), self-serve data infrastructure (a platform provides teams with the tools to manage their data products without requiring central engineering help), and federated computational governance (policies about security, privacy, and quality are federated — enforced consistently across all domains without requiring central coordination for every decision).
In practice, a data mesh deployment looks like this: the Sales domain team owns the customer and opportunity data product, documents it in a data catalogue, maintains its quality, and exposes it through APIs and query interfaces. The Finance team owns transaction and P&L data products. The Product team owns usage and event data products. Any team that needs data discovers it in the catalogue, requests access, and uses it without going through a central data team.
The transition from a centralised model to data mesh is organisational as much as technical. It requires domain teams to accept accountability for data quality and accessibility that they previously could delegate to central engineering. This cultural shift is the hardest part of the transition.