Data Mesh

Data mesh is an architectural and organisational approach to data management proposed by Zhamak Dehghani. It addresses the bottleneck that occurs when a central data team is responsible for all data pipelines, transformations, and quality across the entire organisation. Data mesh is built on four principles: domain ownership (domain teams own and are accountable for their data), data as a product (data is treated as a product with clear interfaces, documentation, and quality standards), self-serve data platform (a shared platform provides the tools and infrastructure for domain teams to manage their data products), and federated computational governance (governance standards are defined centrally but implemented by domain teams). For AI initiatives, data mesh provides clear data ownership (knowing who to ask about data quality and semantics), well-documented data products (reducing the time to understand and use data), and scalable architecture (enabling AI teams to access data from multiple domains without waiting for a central team).

Traditional centralised data architectures create bottlenecks as organisations scale. The central data team becomes overwhelmed, data pipelines take months to build, and domain expertise is lost in translation. Data mesh addresses these challenges by distributing responsibility to the teams that best understand their data. For AI adoption, data mesh can accelerate access to high-quality, well-documented data across the organisation. AI teams can discover and consume data products from different domains, rather than waiting for a central team to build custom pipelines. However, data mesh is a significant organisational change, not just a technology choice. It requires domains to invest in data capabilities, governance structures to evolve, and cultural shifts towards data ownership and product thinking. Organisations should evaluate whether data mesh suits their scale and maturity before adopting it.