A strong answer. “The bottleneck is real, and decentralising ownership only helps if the domain teams can carry it — otherwise you replace one slow team with twelve inconsistent ones and no way to join anything.

So I would ask three things first. Do domain teams have anyone who can own a data product, including on-call for freshness? Usually two or three do and the rest do not, and that asymmetry decides the design. Do we have a platform good enough that owning a data product is a reasonable ask rather than a research project? If not, mesh means every team builds their own pipeline badly. And who owns the conformed entities that cross domains — customer, account — because that is precisely where mesh is weakest and where the organisation's most important numbers live.

What I would actually do is neither, at first. Keep core conformed entities central and owned, because one definition of customer is worth more than any organisational principle. Push domain-specific marts out to the teams that understand them, with contracts and quality standards they must meet. And invest hard in the platform, because whether decentralisation works is almost entirely a function of how good the paved road is.

That is mesh-shaped in the places it works and centralised where it does not, which is what most successful implementations actually look like even when they use the vocabulary.”