Business users in the warehouse, safely

For years the central data team was the only contributor. If a domain needed a new measure or a corrected definition, it filed a request and waited. The team that understood the metric best was the one team not allowed to touch it. The warehouse was a fortress, and the queue at the gate was always long.

Governed contribution changes that. The domains that own the data (finance, supply chain, marketing) propose measures and definitions directly, and a certification workflow decides what gets promoted. The central team stops being the only author and becomes the steward of the gate. Contribution opens up; quality stays gated.

Why now

Data mesh named the right problem: a single central team is a bottleneck, and the people closest to a domain understand its data best. Its three good ideas (data-as-a-product, domain ownership, self-serve tooling) are genuinely useful. The trouble was the packaging: most organizations heard “data mesh” and “reorg,” and a multi-year restructuring is hard to start and harder to finish.

What’s possible now is to keep the kernel and drop the reorg. You don’t need to redraw the org chart to let a domain own its measures; you need a place to propose, a way to test, and a gate that promotes only what passes. The same semantic layer that gives AI trustworthy definitions is the surface a domain expert can safely contribute to, because every definition there is reviewable, testable, and versioned. Self-serve stops meaning “raw and unowned” and starts meaning “proposed within guardrails.”

What it looks like

A supply chain analyst notices the warehouse has no measure for on-time-in-full delivery, the number her team lives by. Old world: she files a ticket and waits a quarter.

Instead she proposes it. She describes the measure in plain language, names the tables and the logic, and submits it as a candidate. The platform runs it against sample data, flags that one join would double-count returns, and shows her the corrected result. A data steward reviews it, confirms it matches the agreed grain, and certifies it. The measure is promoted, tagged with her name as proposer and the steward as certifier, and from that moment everyone, and every agent, reads the same on-time-in-full. It went from idea to certified in days, through a gate, not around one.

Where it’s heading

Toward a warehouse the whole organization improves rather than one a single team defends. As more domains contribute through the gate, the central team spends less time authoring measures and more tending the workflow: setting test standards, curating what’s certified, retiring what’s stale.

Contribution and correction start to blend: the same path that lets a domain propose a new measure lets it fix a wrong one, which is why governed contribution and natural-language writeback converge, both are audited, gated ways for the people who know the data to improve it in place.

How we think about it

Invite contribution, gate promotion. The instinct to protect quality by locking the warehouse is understandable and wrong: it doesn’t raise quality, it raises the queue, and it keeps the people who understand a metric from owning it. Open contribution wide and make promotion the controlled act: anyone can propose; only what’s reviewed, tested, and certified is promoted. That’s how you get the self-serve energy of data mesh without the chaos, and domain ownership without the reorg. The gate isn’t a wall; it’s the thing that makes the open door safe.