Where does an organisation without any start?

With the two or three bodies of material that actually matter, naming an owner for each and deciding permitted use. That is achievable in an afternoon and produces answers. Starting with a framework covering everything produces a document, and the question that arrives still has nobody to answer it.

Why did AI make this urgent?

Because material that sat quietly is now indexed, fed to tools and used to answer questions, and each of those raises the questions governance exists to have already settled. The absence was tolerable while nothing was reading everything; it stops being tolerable the moment something is.

Is quality really part of it?

It became central. A tool answering from your material inherits its errors and repeats them confidently at speed, so an outdated document that was harmless in a folder is now a source of wrong answers. Deciding what is authoritative is the part that changed most with AI.

Glossary

Data governance

How an organisation decides who owns which material, who may use it, and what standard it is kept to, so those questions have answers before somebody needs them.

In plain terms

Somebody has decided, in advance, who is responsible for each significant body of material, who may use it and for what, and how good it is expected to be. Most organisations have some of this by custom rather than by decision, which works until a question arrives that custom cannot answer.

Why it matters

Because AI turned a slow-burning gap into an immediate one. Material that sat quietly in a system for years is now being fed to tools, indexed for search and used to answer questions, and each of those raises exactly the questions governance exists to have already answered. Organisations frequently discover they have none at the point where they most need it.

How it works

Ownership is the part that does the work, and it means a named person rather than a department. Somebody who can answer whether a particular use is acceptable, and who is expected to be asked, converts an open question into a decision. Material owned by everybody in general is owned by nobody in particular, which is the state most organisations are actually in.

Quality is a governance question and is usually treated as a technical one, which AI made costly. A tool answering from your material is only as good as the material, so an outdated policy document or a duplicate customer record now produces a confident wrong answer at speed rather than sitting harmlessly in a folder. Deciding what is authoritative is suddenly load-bearing.

The useful version is narrow. Deciding ownership, permitted use and expected quality for the handful of bodies of material that actually matter is achievable and produces answers; attempting it across an entire organisation produces a framework document and no answers. The narrow version also tends to reveal which material genuinely matters, which is informative in itself.

It is a set of decisions rather than a technology, and tools that claim to provide it are providing enforcement or visibility for decisions somebody still has to make. That is worth knowing before buying anything, because a catalogue with no owners recorded in it is an inventory rather than governance.

The moment it becomes urgent is recognisable: somebody proposes pointing a tool at a body of material and nobody can say whether that is allowed. That question has no technical answer and no vendor can supply one, which is why it tends to be the point at which organisations start doing this rather than talking about it.

The question that arrives, and what it needs

Describing this practice through the question it answers is more useful than describing it as a discipline, because the discipline framing invites an organisation to build a framework and the question framing invites it to name three people. The second is achievable this week and produces the thing that was actually missing. It also correctly locates why AI made this pressing: the question on the left has always been askable and was rarely asked, because pointing something at a large body of material used to require a project and now requires an afternoon. The rate at which the left column arrives went up sharply, and organisations that had comfortably deferred the right column found the deferral had become expensive without anybody deciding to make it so.

Seen in the wild

Deciding which internal documents a search deployment may index, which requires knowing who owns them.
Glean
Establishing which workspace pages are authoritative before a tool starts answering from them.
Notion AI
Deciding which records an automation may send onward to a model, and who gets to say so.
n8n

Common misconceptions

People assume

It is a document we need to write.

In fact

It is a set of decisions that need owners. A written framework with no named person against each significant body of material records an intention, and the question that actually arrives, may this tool use this material, still has nobody to answer it.

People assume

We can buy a tool for it.

In fact

Tools provide visibility and enforcement for decisions somebody still has to make. A catalogue listing every data source with no owner recorded against any of them is an inventory, and the useful part, who decides, is exactly the part no product supplies.

Telling them apart

Data governance vs AI policy

Data governance

Who owns which material, who may use it, and to what standard.

AI policy

What staff may and may not do with AI tools.

One is about the material and predates AI; the other is about the tools. Both get asked at the same moment.

Questions

Where does an organisation without any start?: With the two or three bodies of material that actually matter, naming an owner for each and deciding permitted use. That is achievable in an afternoon and produces answers. Starting with a framework covering everything produces a document, and the question that arrives still has nobody to answer it.
Why did AI make this urgent?: Because material that sat quietly is now indexed, fed to tools and used to answer questions, and each of those raises the questions governance exists to have already settled. The absence was tolerable while nothing was reading everything; it stops being tolerable the moment something is.
Is quality really part of it?: It became central. A tool answering from your material inherits its errors and repeats them confidently at speed, so an outdated document that was harmless in a folder is now a source of wrong answers. Deciding what is authoritative is the part that changed most with AI.
How do we know we need it?: Somebody proposes pointing a tool at a body of material and nobody can say whether that is allowed. That question has no technical answer and no vendor supplies one, and it is reliably the moment organisations move from discussing this to doing it.