Agency of Record Without Surrendering Control

An agency of record should centralize responsibility, not ownership.

An agency of record earns its keep through continuity. One operator knows the systems, carries an ongoing mandate and answers for the result, but vague authority ruins that deal; once AI agents can read customer data, change campaigns or write to a CRM, an unclear mandate becomes an access-control problem.

What does agency of record mean?

In marketing, an agency of record (AOR) is the organization formally appointed to manage a defined discipline on an ongoing basis and act for the client inside that scope. The original use was closely tied to media buying, where publishers and platforms needed to know which agency could place spend for an advertiser. Today the label can cover media, creative, public relations, lifecycle marketing or a broader operating remit.

The definition is narrower than “our main vendor.”

A monthly retainer alone does not create useful AOR accountability. The relationship needs a named scope, authority to make specified decisions and one party answerable for keeping the work running.

The 2025 ANA and 4As tenure study puts the average AOR relationship at about seven years, with integrated firms averaging 7.3 years and media-only firms 3.7 years. That is a long time to leave ownership, access and exit mechanics implicit.

Does a lean B2B team need an AOR?

I would put a 1-to-50-person B2B company on an AOR only when coordination has become a real operating job and nobody inside the company owns it. A founder juggling a content contractor, a paid media specialist, a CRM consultant and an outbound vendor has created a roster without creating an operator.

An AOR can remove that coordination load, but only if the mandate is specific.

“Own marketing” is too broad. “Run paid acquisition and lifecycle email against the approved budget, inside client-owned accounts, with weekly pipeline reporting” is a usable mandate because it defines the work, the budget boundary and the reporting duty.

I would stay with project work when the output has a natural finish line, such as a website migration, attribution audit or CRM cleanup, and I would also avoid an AOR while the company is still deciding what message or Ideal Customer Profile (ICP) works.

A standing operator can run a validated growth motion; it should not be mistaken for permission to invent the company’s go-to-market strategy.

The choice sits one level above ordinary vendor evaluation. First decide whether you need continuing ownership or a bounded delivery. Then use the growth agency evaluation criteria to assess the people and commercial model behind it.

What should an AOR agreement control?

Start with a responsibility map before anyone writes contract language. Every recurring activity should have an owner, an approval point and an evidence trail.

Control area	Write down	Avoid
Scope	Channels, systems, markets and excluded work	”Full marketing support”
Authority	Decisions the operator may make and spending limits	Approval by habit or chat history
Accounts	Client admins, manager access and credential rules	Accounts created under a vendor identity
Data and output	Export rights, retention, work product and derived assets	Data held only in a vendor portal
Exclusivity	Named discipline, geography and conflict rules	Blanket limits on unrelated specialists
Exit	Notice, handoff format, access removal and open-work treatment	A promise to “support transition”

Use this as a business checklist, then have counsel turn it into an AOR agreement that matches your jurisdiction and risk. Ask whether the company can replace the operator without losing its accounts, evidence or ability to keep working.

That ownership check complements the broader questions to ask a SaaS marketing agency. An AOR needs all the ordinary capability checks plus a much sharper definition of delegated authority.

Who should own the accounts, data and output?

The client should hold direct administrative access to every core account from day one. The operator receives the access needed to work, preferably through a manager role or its own named user. Shared passwords make the offboarding problem worse and erase the audit trail.

Google Ads shows why account architecture matters. Its manager-account documentation says a linked manager can receive administrative ownership, while the client account still owns its data and can remove the manager by unlinking. When an existing client account is linked, ownership does not transfer by default, and I would copy that pattern across the stack with the client account first and delegated operator access second.

Data rights need the same precision. HubSpot’s customer terms say the customer owns its customer materials and customer data. Your AOR agreement should go further and name the derived work the engagement creates: scoring rules, prompts, campaign history, audiences, dashboards, research notes and agent instructions. Platform terms define the relationship with the platform. Your agreement must define the relationship with the operator.

I would not accept a setup where the client receives PDF reports while the working history stays inside the operator’s private system.

Reports are evidence of work. The account history, rules and source data are what let someone else continue it.

Run a four-question ownership test before you sign:

1. Is the client a top-level administrator on every core account?
2. Can the client export the source data, working history and current instructions in usable formats?
3. Can the client revoke every human and agent identity without asking the operator to do it?
4. Could a replacement continue from the documentation and logs without access to a private vendor portal?

Any no answer is a dependency the agreement should resolve or price explicitly.

What changes when AI agents do the work?

Traditional AOR agreements delegate work to a company and its people. Agent-run delivery adds non-human identities that may hold credentials, call tools and change live systems. The contract should describe those agents as controlled operators, not as invisible implementation detail.

Microsoft’s AI agent responsibility model keeps customer accountability for data, identity, authorization and human oversight across deployment models, while recommending least privilege for each tool, authorization at action time, human gates for sensitive or irreversible actions and logs for tool use.

I would ask for six operational answers before an agent receives write access:

1. Each agent has a named business owner and a distinct identity.
2. Its allowed systems, records and actions are listed.
3. Read, propose and execute permissions are separated.
4. External sends, spending changes, deletes and production writes require explicit approval.
5. Every action produces a timestamped log tied to the identity and instruction that caused it.
6. The client can revoke access and stop the workflow without the operator’s help.

Human review also needs a real decision point. The NIST Generative AI Profile warns that people may over-rely on generated output as systems appear more reliable; a person clicking “approve” without seeing the source, proposed action and expected effect is ceremony, not oversight.

Pricing quotes should expose the operating layer too. The AI growth agency pricing guide explains what part of a fee pays for monitoring, failure handling and human judgment after a workflow launches.

Is an agency of record or a project agency the better fit?

Project work buys an output. An AOR buys continued responsibility for a function. Ask what happens the week after delivery.

A CRM cleanup can end after the agreed fields, deduplication rules and validation report pass acceptance. Ongoing CRM hygiene has no natural end; somebody must monitor imports, resolve conflicts and keep lifecycle definitions consistent. The first is a project. The second may justify an AOR-style mandate or a managed CRM agent.

I would begin with a project when the team has not worked with the operator before. A bounded assignment creates evidence about quality, response time and handoff discipline. Convert it into an AOR only when continuity has proved more valuable than optionality.

For a wider comparison of operating shapes, the growth agency vs demand generation agency guide separates discovery work from repeatable pipeline execution. The same distinction matters here. An ongoing mandate works best when the work itself is understood.

Can a company have more than one agency of record?

Yes. The scope matters more than the count.

National Vision’s 2025 AOR appointment gave VML a broad remit across creative, customer relationship management, social strategy, brand identity, design and experiential work. Windstar Cruises used its 2025 AOR for a narrower public-relations remit covering media relations, influencer collaboration and brand partnerships.

Large organizations split the model further. In its October 2026 marketing operating-system review, Coca-Cola retained WPP Open X across global media outside selected markets, chose Omnicom for North America and kept Dentsu in Japan and South Korea, while WPP remained the global creative partner.

A lean company does not need that roster. It does need the same clarity. If one operator runs outbound and another runs content, write the boundary at shared systems, audience definitions and reporting. “They will coordinate” is weak. Name who resolves a conflict and whose definition wins.

How should an AOR relationship end?

Design the exit before the appointment.

I would require a current access register, exportable data, documented workflows, open commitments, named subcontractors and a handoff owner. The notice period should match the time needed to move live work without losing campaign history or interrupting customer communication.

Google’s unlinking guidance offers a useful test: the individual account keeps its campaign history after a manager is removed, but shared lists, cross-account tracking or manager-paid billing can still break. Apply that logic to every system in the stack. Keeping the account is insufficient if the tracking, billing or automation disappears with the operator.

I would judge the AOR agreement by whether replacing the operator looks boring. The client keeps the systems and evidence, the outgoing operator hands over a current map, and the next operator receives only the permissions it needs.

Momentum Nexus runs managed growth agents inside defined permission and approval boundaries. Review the operating model on the AI agents page before you turn recurring work into a standing mandate.