Decision Rights

Decision rights are the authority to settle a defined class of question, held by a named role within a stated scope. In data governance they are what turns a list of owners into an operating model: not who is responsible for this table but who may decide what a paid order means, who may grant access to it, and who may approve removing a column.

The distinction matters because a role without decision rights is a contact. Someone to ask, who then asks someone else. Most governance that feels slow is not short of owners; it is short of people who may answer without convening anyone.

What a decision right has to state

• The question class. Meaning, access, retention, change approval, quality response — stated specifically enough that someone can tell whether a given question is in it.
• The scope. Which data, which systems, which jurisdictions. A right over customer data means nothing until someone says whether that includes the copy in the marketing platform.
• The limit. What this holder may not decide — usually anything that spends money they do not control, accepts risk on the organization’s behalf, or crosses a legal or contractual constraint. Being an owner does not confer authority to lift a rule imposed from outside.
• The escalation. Who decides when the holder cannot, and how long that takes. An escalation path with no name in it is the reason questions sit for weeks.

Seniority is not the same as proximity

Decision rights sit at different altitudes for different questions, which is why they are distributed rather than collected. The UK government’s data ownership model puts owners “at the leadership level” while also stating that they “do not need a granular understanding of the data,” with stewards “responsible for day-to-day operational activities in their data domain that support data owners’ decisions.” Read as decision rights, that is a split: the authority to commit the organization sits high, and the authority to settle daily questions sits close to the work.

Collapsing the two produces predictable symptoms. Put every decision at the top and the leader becomes a queue; put every decision at the bottom and commitments get made that nobody was authorized to make. The same model notes that “the responsibilities of ownership are not exclusive to a single person and require close collaboration across organisational levels,” which is the same point stated as a warning against tidy org charts. The roles themselves are set out in data owner, steward, and custodian.

Global or local

In any organization with more than one domain, the hard part is not assigning rights but drawing the line between what each domain settles for itself and what must be settled once for everyone. Zhamak Dehghani frames the federated model around exactly this difficulty: domains hold “autonomy and domain-local decision making power” while adhering to a set of global rules, and the governing group’s job is to decide “what decisions need to be localized to each domain and what decisions should be made globally for all domains.”

A workable test for the global list: a decision belongs there if leaving it local would prevent two domains’ data from being used together. Identifier formats and classification standards pass that test. Internal column names do not. Keeping the global list short is not a stylistic preference — every item on it is a decision someone must wait for.

Writing them down

Derive the list from questions rather than from assets. Going dataset by dataset produces a register nobody reads; going question by question — what does this mean, who may see it, what may consumers rely on, what happens when it is wrong, who approves a change — produces a short list of rights that covers most of the estate, because the same questions recur everywhere.

Record each answer where the next person will look for it: a meaning decision belongs with the metric definition, a promise to consumers belongs in a data contract, and an architectural choice belongs in a decision record with the conditions that would reverse it. And track the questions that have no holder at all — a count of unowned decisions is the most honest governance metric available, because it is the only one that gets worse when nothing is happening.

How decision rights are arranged across centralized, federated, and domain-owned models is worked through in Who Decides What.

References: GOV.UK, Data ownership model; Zhamak Dehghani, Data Mesh Principles and Logical Architecture.