Who Decides What in Marketing? A Governance Model for Modern Teams

Create clear marketing decision rights for strategy, brand, channels, content, data and AI using owners, evidence requirements and escalation rules.

For teams slowed by unclear approval or exposed by unowned risk, this guide turns governance into explicit decision rights. It provides a practical grid for strategy, brand, channels, content, data and AI, plus evidence and escalation rules.

Short answer: Give every consequential marketing decision one accountable owner, define the evidence that owner needs, identify who must be consulted, state what may be delegated and publish the conditions for escalation. Governance is not a chain of people saying yes. It is the allocation of authority before a deadline. Keep routine, reversible choices close to practitioners; elevate decisions involving material claims, customer data, large or automated spend, rights, vulnerable audiences or difficult-to-reverse public harm.

Marketing work crosses product, sales, finance, legal, data, technology, creators and agencies. Without defined authority, teams either wait for everyone or let the most available person decide. Both outcomes create hidden cost. Consensus can also obscure responsibility: when a claim fails, “the team approved it” does not identify who had the authority and evidence to release it.

This article is the canonical governance upgrade. It owns decision rights, not the wider operating model or organisation chart.

The Marketing Decision Rights Grid

The analytical spine is CDM's Marketing Decision Rights Grid. A useful row records six fields: decision, accountable owner, permitted delegation, required evidence, required consultation and escalation trigger.

Titles differ by organisation. What matters is that one person can make or stop the decision and can be held to the stated standard.

Decision rights are more precise than task responsibility

A contributor may prepare an asset, a specialist may review evidence and a leader may be informed, but only one role should own the final decision. Traditional RACI charts—responsible, accountable, consulted and informed—can help, yet they often become lists in which several people appear accountable. The grid adds the evidence and escalation needed to make authority usable.

Separate four rights:

• Propose: formulate an option and its evidence.
• Review: test a defined aspect, such as scientific accuracy or accessibility.
• Decide: accept, reject or require change.
• Release: execute the approved version under stated controls.

One person may hold several rights, especially in a small team. They should still be named separately when a release can deviate from an approval.

Govern by consequence and reversibility

Not every social caption needs executive review. Apply control in proportion to potential harm and difficulty of reversal. A low-cost, clearly labelled experiment with no sensitive data may sit with a channel practitioner. A health claim, use of children's data, synthetic public figure or automated budget change deserves specialist evidence and higher authority.

NIST's AI Risk Management Framework and Generative AI Profile both emphasise governance across the lifecycle. The principle generalises: decide acceptable authority before the system operates, monitor what happens and retain a path to contain and correct failure.

Create delegated limits in operational terms. “Marketing manager can approve paid activity” is vague. “Marketing manager may launch from the approved claim set to the defined audience up to £5,000 per week; new customer-data uses, new regulated claims or total exposure beyond the cap require escalation” can be applied and audited.

Put evidence beside authority

Authority without information produces arbitrary decisions. For each decision, specify what the owner must see. A content approver may need the exact final render, source claim, evidence date, audience, platform crop and disclosure. A budget owner may need incremental evidence, contribution margin, uncertainty and exposure limits rather than reported return on advertising spend alone.

The evidence requirement should be proportionate. A reversible headline test may need a brief and baseline. A company-wide repositioning needs broader customer, commercial and operational evidence. When information is missing, the owner should record whether the decision is to gather more evidence, run a bounded test or accept the uncertainty.

Design consultation so it can finish

Consultation should identify expertise, not create universal vetoes. State what each reviewer assesses and by when. Legal counsel may assess a legal question; data protection staff may assess a new processing purpose; a product expert may verify functionality. The accountable owner integrates advice and records the final decision within lawful bounds.

Silence must not be treated as approval for high-risk work. For routine work, pre-approved standards and response windows can prevent delay. Escalation should resolve disagreement about risk or authority, not merely move every late task upward.

Worked application: an AI-assisted product campaign

Consider a fictional software company using an AI tool to draft campaign variations. This is an illustrative governance case.

The product marketer may propose claims only from an approved register. The creative lead decides wording and imagery within the brand standard. A subject expert reviews technical meaning. The privacy owner must decide any new use of customer records. The paid-media owner may release approved variations within an exposure cap. Automated publication is disabled until the final rendered asset, audience and landing page pass a human release check.

If the tool invents a performance claim, the editorial owner stops the release, preserves the draft and checks whether derivatives exist. If a correct claim is phrased differently but retains scope, the creative lead can resolve it without executive escalation. If the campaign proposes a new customer-data enrichment, that leaves the delegated path and enters data-governance review.

The model is faster because ordinary authority is clear and exceptions are visible.

Build the grid in seven steps

1. Inventory recurring and high-consequence decisions, not every task.
2. Name one accountable owner for each decision.
3. Define proposal, review and release roles.
4. State the minimum evidence and authoritative record.
5. Set delegated limits using spend, exposure, data, claim and reversibility.
6. Name an escalation owner and expected response time.
7. Review incidents and delays to refine the grid.

Publish the current grid where employees and partners work. Link it to the claim register, access system and incident process. Review it when roles, markets, regulation, providers or automation change.

Audit whether governance is real

Sample recent decisions. Can the team identify who decided, which version was approved, what evidence they used and how an exception was handled? Measure unowned decisions, approval waiting time, rework caused by missing evidence, releases outside authority and time to contain an error.

Do not optimise approval speed in isolation. A faster process that bypasses meaningful review is not better governance. Conversely, a perfect record produced after every opportunity has passed is not useful control.

Limits and risks

Internal decision rights do not override law, regulation, contracts or directors' duties. Formal legal responsibility may remain with the organisation or a statutory role even when operational decisions are delegated. Obtain specialist advice for regulated claims, employment, financial promotions, health, children, privacy and other consequential areas.

Governance can become theatre if owners lack competence, time or power. It can also suppress useful dissent when hierarchy replaces evidence. Preserve a challenge route and protect staff who raise material concerns. Finally, a grid cannot rescue a culture that rewards speed while punishing correction.

CDM verdict

Marketing governance should move routine choices closer to qualified practitioners and make exceptional risk easier to see. Name one owner, put the required evidence beside the decision, bound delegation and preserve escalation and correction. The goal is not more approval. It is a decision that can be explained, challenged and repaired.

Related guides

• Which Marketing Work Should Be In-House, Fractional or Agency-Led?
• How Do You Prove a Synthetic Spokesperson Is Not a Fake Testimonial?

Frequently asked questions

Should legal approve every marketing asset?

Usually not. Legal review should follow defined triggers such as regulated claims, new contractual rights, sensitive audiences, novel data use or material dispute. Approved claim libraries, templates and guidance can allow trained marketers to handle routine work. Counsel should help define the boundary and remain available for exceptions. The accountable business owner still decides commercial action within legal advice; legal is not a generic quality-control department.

Can two people be accountable for one decision?

Shared leadership may contribute, but one operational decision needs a final owner. Two accountable names often mean unresolved authority when they disagree or are unavailable. Split the decision if it genuinely contains separate authorities—for example, a privacy owner decides whether processing is permissible while a marketing owner decides whether the campaign is worth running. Record how a conflict is escalated. Collaboration is compatible with singular accountability.

How is a decision-rights grid different from a RACI?

RACI maps participation in tasks. The CDM grid starts from a consequential decision and adds evidence, delegated limits and escalation. It can use RACI roles inside each row, but it rejects multiple ambiguous “accountable” roles. The grid also links approval to the version and conditions that were actually decided. Teams may continue using a RACI if it produces those outcomes; the name matters less than operational clarity.

Who should approve AI-generated marketing content?

The accountable owner for the underlying task should approve it, supported by relevant specialists. AI use does not create a new universal approver. A product claim needs product and evidence competence; personal-data use needs privacy authority; synthetic media may need rights and disclosure review. The reviewer must inspect sources and the final output, have enough time and possess authority to reject it. High-risk automated release requires separate control and fallback.

How often should marketing governance be reviewed?

Review at least during major planning and role changes, and immediately after a material incident, provider change or expansion into a new market or regulated activity. Use decision samples rather than merely checking that a document exists. Update delegated thresholds as experience and controls improve. Preserve earlier versions so the organisation can explain which authority applied at the time of a disputed decision.

Next decision: How Marketing and Sales Should Work Together When Buying Journeys Are Nonlinear

Related reading: The Modern Marketing Operating Model · What Capabilities Does a Modern Marketing Team Need? · The Marketing Claim Register

Sources and research notes

• UK Government Service Manual: Governance principles for agile service delivery — authoritative guidance on clear decision making, outcomes and proportionate governance. Checked 12 October 2026.
• NIST AI Risk Management Framework and NIST Generative AI Profile — primary risk-governance sources for AI-enabled decisions. Checked 12 October 2026.
• ICO: Accountability framework — UK regulator guidance on responsibility, risk records and oversight. Checked 12 October 2026.
• ASA/CAP: Substantiation — authoritative requirement for evidence supporting objective advertising claims. Checked 12 October 2026.
• Limitations: The Marketing Decision Rights Grid is CDM editorial analysis and does not allocate legal liability. Roles and escalation must be adapted to the organisation, jurisdiction and regulated activity.