Data Mesh Without the Chaos

Data mesh has suffered the fate of most good ideas that become fashionable: the label now travels far more widely than the discipline behind it. In many organisations it has quietly come to mean giving every team its own data and stepping back — decentralisation as an end in itself. That is not data mesh. That is fragmentation with a more respectable name.

The genuine idea is more demanding, and more interesting. It is that the people closest to a domain are best placed to own its data — to understand it, model it, and stand behind its quality — and that ownership should therefore sit with them rather than with a distant central team. But ownership is only half of it. The other half is that everyone owns to the same standard, so that data produced in one corner of the organisation can actually be found, understood and trusted in another.

Federation is not the absence of standards

The failure mode is predictable. An organisation embraces the ownership half of the message, distributes responsibility enthusiastically, and neglects the standard half because it is the harder and less popular part. The result is a hundred local optima and no interoperability — every team confident in its own data and unable to rely on anyone else's. This is worse than the centralised bottleneck it was meant to replace, because at least the bottleneck was consistent.

What makes federation work is a small number of things held firmly in common while much else is left to local judgement. Shared definitions for the concepts that cross domains. Agreed expectations for quality and for how data is described. Clear interfaces, so that consuming another team's data does not require knowing the team. This is the same instinct behind treating data as a product rather than a byproduct: something with an owner, a consumer, and a promise attached.

None of this is primarily a technology question. Platforms and tooling help, but the hard part is the operating model — who is accountable for what, how standards are set and enforced without a central chokepoint, and how a domain team is held to a promise it makes to the rest of the organisation. Get that right and the technology is comparatively straightforward. Get it wrong and no platform will save you.

The prize is real: data that scales with the organisation instead of collapsing into a central queue, owned by people who actually understand it. But it is earned by holding two ideas at once — distributed ownership and shared discipline — and resisting the temptation to keep only the half that is easy. Mesh without the standard is not agility. It is just chaos that has learned to describe itself well.