Data Governance: What It Is, and What It Means for Marketing Data
Data governance is the rules, roles and controls that make data trustworthy. Here's what a program contains — and what changes for marketing data.
Data governance is the set of rules, roles and controls that decide how data is defined, created, checked and used — so that the people relying on it can trust what it says.
Most governance writing assumes the data already sits in a warehouse. Marketing data does not. It is created upstream, by many hands, in the ad platforms themselves, and by the time it reaches the warehouse the decisions that determined its quality were made weeks earlier by someone who has moved on to the next launch.
What is data governance?
The rules, roles and controls that keep data trustworthy enough to act on.
Governance answers four questions about every piece of data that matters: what it means, who owns it, what counts as correct, and what happens when something is wrong. The enterprise definition is stable across publishers (IBM, “What is data governance?”, accessed 2026-09-11).
Where the published treatments differ from practice is scope. They describe governance as something applied to data at rest: cataloged, access-controlled, lineage-tracked. That is real and necessary. It is also downstream of every decision that determines whether the data was worth governing.
The four pillars
Ownership, definitions, quality standards, and enforcement.
| Pillar | Decides | Fails as |
|---|---|---|
| Ownership | Who is accountable for this data domain, and who may approve a change | A RACI nobody consults; decisions escalate or stall |
| Definitions | What each field means, in business language | Two teams reporting different numbers and both being right |
| Quality standards | What counts as a correct value, and how it is measured | A dashboard of green checks over data nobody trusts |
| Enforcement | Where the rule is applied, and what happens when it is broken | A published policy with no mechanism; compliance decays quietly |
Informatica’s treatment of the pillar taxonomy is representative of the standard framing (Informatica, “What Is Data Governance?”, accessed 2026-09-11).
The fourth pillar is where programs actually fail. Ownership, definitions and standards are all decisions, and decisions are the part organizations are good at making in a workshop. Enforcement is a systems problem, it sits with a different team, and it is routinely treated as an implementation detail to be worked out later.
The framework in detail: Read: data governance framework — how the pillars assemble into an operating model.
A worked example
One data domain, its owner, its definitions, its allowed values and its check.
Abstraction is what makes governance hard to start. Here is a single domain, fully specified.
Domain: campaign metadata. Owner: Director, Marketing Operations. Approves new permitted values; accountable for the domain as a whole. Steward: Campaign Operations Manager. Maintains the definitions and handles day-to-day value requests.
| Field | Definition | Allowed values | Check | Applied |
|---|---|---|---|---|
channel | The media category the placement ran in | 8 values, closed list | Value must be in list | Campaign setup form |
market | The country the placement targeted | ISO 3166-1 alpha-2 | Format and list check | Campaign setup form |
campaign_name | The initiative this placement belongs to | Pattern {region}_{brand}_{initiative}_{YYYYQn} | Regex | Campaign setup form |
agency | The partner that built the placement | Approved partner list, owned by Procurement | Value must be in list | Campaign setup form |
Escalation: a request for a ninth channel value goes to the owner, is decided within five working days, and is recorded with a date and rationale.
That is a governance program. It fits on a page, it names people rather than functions, and every rule has a place where it is applied. Scaling it means repeating it per domain, not replacing it with something more elaborate.
At enterprise scale
Federated ownership across regions, brands and third parties.
At scale the single-owner model breaks, and the usual replacement — a central governance team owning everything — breaks differently. Central teams become bottlenecks, regions route around them, and the governance program becomes something that happens to other people.
Federated governance keeps a central function for the decisions that must be global (which domains exist, which fields are enterprise-wide, how conflicts resolve) and pushes value-level authority to the people closest to the work. A regional lead can add a market-specific value without a global forum. Nobody can add a new dimension without one.
Two failure modes are worth naming. Over-centralization produces a queue and a shadow taxonomy. Over-federation produces twelve regional variants of the same field and a reconciliation project. The boundary between them is not a principle; it is a list, and writing that list down is most of the work.
Taxonomy governance, ownership and change management comes up across 63 enterprise accounts in our customer conversations, and it is consistently the ownership question rather than the taxonomy question that stalls programs.
Standards vs policies vs procedures
A policy says what must be true; a standard says exactly what “true” looks like; a procedure says who does what.
| Policy | Standard | Procedure | |
|---|---|---|---|
| States | An intention or obligation | A specific, checkable requirement | A sequence of actions |
| Example | “Every campaign must record its channel” | “channel must be one of these eight values” | “Request a new value via the intake form; owner decides in five days” |
| Audience | The organization | Systems and the people entering data | The people operating the process |
| Machine-checkable | No | Yes | No |
The middle column is the one that connects governance to anything. A policy cannot be enforced by a system because it does not say what the value should be. A data standard can.
What good governance looks like in practice
It is invisible: the right value is the easy value to enter.
Governance fails when it is designed as an audit. It works when it is designed as a default.
Working governance does not feel like governance to the people inside it. The campaign manager picks a channel from a list of eight instead of typing one. The agency’s setup form will not submit without a market. Nobody reads a policy, because the policy is expressed as the shape of the form.
Failing governance is conspicuous. It has a steering committee, a quarterly compliance report, a mapping table maintained by an analyst, and a documented standard that the data does not match. Every one of those is a symptom of the rule being applied somewhere other than where the data is created.
The practical test: count how many people have to remember something for the data to come out right. If the answer is more than zero, the control is in the wrong place.
Governing data you do not own
Marketing data is created by agencies and campaign managers in systems the data team does not administer.
This is the case the published governance literature does not cover, and for marketing data it is the majority case.
| Warehouse-resident data | Marketing data | |
|---|---|---|
| Created by | Systems, on a release cycle | People, continuously, at launch |
| Created in | Platforms the data team administers | Ad platforms, agency tools, spreadsheets |
| Authors | A handful of writing processes | Every campaign manager, region and agency |
| Access control reaches them? | Yes | Frequently no |
| When errors are visible | At load or profiling | Weeks later, in a report |
| Cost to correct | Reprocess | The campaign already ran |
Access control, catalogs and lineage tracking all assume you administer the system where the data lives. None of those levers reaches an agency building a placement in a platform you do not control.
Cross-agency governance and taxonomy compliance failures come up across 74 enterprise accounts, making it one of the most prevalent problems in our customer conversations — and the structural reason is in the table above. Approval workflows, audit trails and access control are raised separately, across 39 accounts, usually by teams who have implemented all three internally and discovered they cover a minority of the records.
The governance lever that does reach outside the perimeter is the one that constrains what can be submitted: a form that offers the approved values, to whoever is filling it in, regardless of who employs them.
A Fortune 100 retail company’s media science team described what that foundation made possible.
“Our big task on the media science team and responsibility in our early years are over data standards and governance. Claravine helped lay that foundation…helping to increase paid media tracking by 65%.” — unnamed, media science team, Fortune 100 retail company
Governance for marketing data: Read: data governance in marketing — the practice applied to data created outside your systems.
How a governance program starts
Pick one data domain, agree its definitions, and enforce them at creation before widening scope.
- Choose one domain that hurts. Campaign metadata, customer records, product data. Pick the one where a wrong value has cost someone a bad decision recently; that argument funds the program.
- Name the owner and the steward. Two people, by name. Not a committee and not a function.
- Write the definitions in business language. One sentence each. If it needs a paragraph, the field is doing two jobs.
- Set the allowed values. Closed lists where possible, format patterns where not.
- Apply the check where the value is created. Not in the warehouse. This is the step that converts the previous four from documentation into governance.
- Publish the escalation path. How a new value is requested, who decides, how fast, and where the decision is recorded.
- Measure conformance and widen. Only after the first domain holds.
Enterprise-wide governance frameworks designed top-down tend to still be in design when a single-domain program has been catching errors for two quarters. Scope is the most common thing to get wrong at the start, and it is always wrong in the same direction.
The reason step 1 says “a domain that hurts” rather than “the most important domain” is political rather than technical. A program justified by a recent, specific, expensive mistake survives its first budget review. One justified by general principle does not, and governance work is slow enough that it always meets at least one budget review before it shows a result.
Compare the tooling: Read: data governance tools — the categories of governance tool, and which problem each solves.
Where governance applies beyond compliance
Reporting, attribution, AI-readiness and asset reuse all depend on the same agreed definitions.
Governance is often funded as a compliance activity, which sells it short and tends to produce the audit-shaped version that does not work. Google Cloud’s framing treats governance as an enablement discipline rather than only a control (Google Cloud, “What is data governance?”, accessed 2026-09-11), and that is the more useful read.
Four things depend on the same agreed definitions, and none of them is a compliance outcome.
Reporting groups by them. Attribution joins on them. AI-readiness depends on them entirely, because a model trained on inconsistently labeled campaign data learns the inconsistency. Asset reuse requires that the same creative be findable by the same description twice.
That is the argument that funds a program, and it is available without invoking a regulator. Data quality is the property all four are really asking about.
Frequently asked questions
What is meant by data governance?
The rules, roles and controls that decide how data is defined, created, checked and used.
What are the four pillars of data governance?
Ownership, definitions, quality standards, and enforcement.
What is the difference between governance and management?
Governance decides the rules; management operates the systems that follow them.
Who owns data governance?
A named data owner per domain, supported by a steward who maintains the definitions.
Does governance apply to marketing data?
Yes, and it is harder there, because the data is created outside the systems the data team controls.
Sources
Outbound citations, named and dated:
- IBM, “What is data governance?” (accessed 2026-09-11) — the category definition, cited and then differentiated on scope.
- Informatica, “What Is Data Governance?” (accessed 2026-09-11) — the pillar taxonomy.
- Google Cloud, “What is data governance?” (accessed 2026-09-11) — governance as an enablement discipline rather than only a control.
