Data Governance: Who Owns the Numbers?
Key takeaway
Governance becomes real when every important metric and data domain has one named human owner who can settle a definition dispute. Start with the ten metrics leadership actually uses, assign owners, publish the definitions where people read the numbers, and grow from there.
Most organisations discover they need data governance in a meeting where two teams present different revenue figures. The instinct is to buy a catalogue tool. The fix is to decide who gets to say what a number means.
The four things governance has to settle
- Ownership — one named person per data domain and per headline metric, accountable for its definition and quality.
- Definitions — what exactly counts as an active customer, a completed order, recognised revenue. Written down, versioned, and visible.
- Quality — the rules a dataset must satisfy, checked automatically, with someone notified when they fail.
- Access — who may see what, especially where personal or financial data is involved.
Start with ten metrics, not the whole estate
Take the numbers that appear in the board pack. For each, name the owner, write the definition and the exact query behind it, and note the source system. This is a week of work and eliminates most recurring arguments — enterprise-wide cataloguing before this is done is where governance programmes go to die.
Quality checks that earn their place
- Freshness — did the data arrive when it should have? Stale numbers presented as current are the most damaging failure.
- Volume — row counts within an expected range, to catch a partial load.
- Uniqueness and referential integrity on key columns.
- Business rules — no negative quantities, no orders without a customer, no future-dated transactions.
- Alert the owner, not a shared inbox nobody reads.
Making it stick without bureaucracy
- Publish definitions where the numbers are consumed — in the dashboard, not in a separate wiki.
- Let anyone challenge a definition; route the decision to the owner, and record the outcome.
- Review access quarterly; it only ever drifts in one direction.
- Add governance to the definition of done for new data pipelines rather than retrofitting it later.