As a company grows, the central data team becomes a bottleneck: every request goes through it, and the queue only grows. Data mesh is a response to that problem — a way of organizing data that decentralizes responsibility without falling into chaos.
The problem data mesh solves
In the traditional model, a central team handles all data for the whole organization. It works at small scale, but it does not scale: the central team does not deeply know each business area and becomes the point everything must pass through. The result is delays and data nobody feels is theirs.
The core idea: data as a product, by domain
Data mesh proposes that each business domain (sales, logistics, marketing) owns its data and offers it as a product — well documented, reliable and easy for others to consume. Who knows sales data best is the sales team; so it is they who curate and serve it.
The four principles
- Domain ownership: each area owns its data.
- Data as a product: treated with the care of a product, not as a byproduct.
- Self-service platform: common infrastructure that gives teams autonomy.
- Federated governance: global rules (security, quality, interoperability) applied locally.
It is not for everyone
Data mesh shines in large, complex organizations, with many domains and an overloaded central team. In a small company, it is over-structuring — centralization is still simpler and more effective. Adopting it as a fad, without the scale that justifies it, creates more complexity than it solves.
Mind the balance
Decentralizing without common governance leads to silos and incompatibilities. The secret of data mesh is in the word "federated": local autonomy within global rules. It is as much an organizational change as a technological one — perhaps more.
In practice
If your data team is the bottleneck of the whole company and there are domains mature enough to own their data, data mesh is worth studying. If you are still building the foundation, focus on that first. Is your biggest data problem a lack of technology, or is it organizational?
