AI is making the agency relationship more valuable in places that conventional scopes of work barely describe. The durable work is no longer only the campaign, report or asset. It is also the workflow that produced it, the customer context that shaped it and the evaluation logic that kept it on brand.
That creates a procurement problem. A client can own every finished deliverable and still be unable to reproduce the system after the agency leaves. Prompts may sit in agency accounts. Connectors may depend on agency credentials. Approval rules, evaluation sets and performance history may never appear on an asset register.
The next agency review should therefore test an AI operating model for portability as seriously as it tests capability. Embedded expertise is useful. Embedded dependency is a liability that often becomes visible only during a handover.
Table of contents
Jump to section:
- AI agency value now lives between the deliverables
- Portability is different from ownership
- Score the exit before you score the demo
- The best partner should survive a clean handover
AI agency value now lives between the deliverables
Agency reviews used to make the work relatively easy to see. A team could compare strategy, creative, channel expertise, staffing and price. AI-enabled service models introduce a less visible layer because the quality of the work increasingly depends on how client context is encoded inside the production system.
Singtel’s latest creative agency review makes the shift unusually explicit. AI sits across planning, production, personalization and capability building, while prospective partners are expected to work closer to the internal team. The agency is being asked to supply an operating capability, not simply a stream of outputs.
The value in that model accumulates between assignments. A useful prompt template contains decisions about the brand. An evaluation set records examples of acceptable and unacceptable work. A connector reflects how customer data is allowed to move. A review workflow encodes who can approve, reject or escalate an automated action.
None of those assets is impressive in isolation. Together, they become the institutional memory of an AI-enabled marketing function. They can improve speed and consistency with every cycle because the system remembers more of what the organization has learned.
That compounding effect also changes the dependency risk. If the agency controls the accounts, documentation and technical configuration, the client’s marketing capability can become stronger while its ability to change partners becomes weaker. The most important asset in an AI agency relationship may be the one that never appears in the deliverables folder.
Portability is different from ownership
Ownership clauses answer who has rights in an asset. Portability answers whether another authorized team can find, understand and operate what created it. The distinction matters because AI workflows often cross three commercial layers: the model provider, the agency and the advertiser.
Provider terms can settle only part of that chain. For example, OpenAI’s May 2025 business terms state that, as between OpenAI and the customer, the customer retains rights in inputs and owns outputs to the extent permitted by law. The same terms make the customer responsible for evaluating the accuracy and appropriateness of those outputs. They do not decide whether an advertiser can access a workspace purchased and administered by its agency.
The agency agreement has to bridge that gap. The 2025 IPA and ISBA Creative Services Framework Agreement added provisions addressing generative AI, a sign that established client-agency contracts are being updated for this layer of work. The exact terms remain a matter for commercial and legal teams, but the operating questions belong in the brief before lawyers begin drafting.
Access should cover more than final files. A client needs a defined route to the prompts, source materials, evaluation results, version history and workflow documentation that it is entitled to retain.
Execution concerns the dependencies required to run the system. Client-controlled credentials, transferable configurations and documented third-party services reduce the chance that a routine roster change becomes a rebuild.
Comprehension is the ability to use what has been transferred. A folder of exports has little value when no one can explain the decision logic, data boundaries or failure procedures behind it.
The point is not that every agency method should become client property. Agencies have legitimate background intellectual property, reusable tools and proprietary expertise. The contract should distinguish those assets from client-specific configurations and operational records instead of pretending that ownership of the final JPEG or article settles the matter.
Ownership without operability is an expensive archive.
Score the exit before you score the demo
An AI demonstration rewards fluency in the best-case scenario. A portability test exposes how the relationship behaves when a model changes, a subcontractor leaves or the client moves the work. That is a more demanding test because it makes hidden dependencies visible while there is still leverage to resolve them.
The review team should ask prospective partners to walk through a hypothetical handover using one representative workflow. The exercise should reveal what the client receives, what remains proprietary and which third-party permissions must be replaced. It should also identify who funds transition work and how long service continuity is maintained.
- Reproducibility: Can an authorized client team or successor partner recreate an approved output from the documented inputs, rules and model version?
- Account control: Are critical data stores, model workspaces and integration credentials held in client-controlled environments, or is there a defined migration route?
- Evaluation continuity: Do quality rubrics, test cases and exception logs transfer with the workflow so the next operator inherits judgment as well as automation?
- Change resilience: What happens if a provider retires a model, changes its terms or removes a feature on which the workflow depends?
- Transition accountability: Which party documents, exports, verifies and supports the handover, and which obligations survive termination?
This due diligence is becoming relevant beyond small specialist shops. Omnicom’s shift of Omni engineering teams to Endava shows how ownership and execution can sit with different organizations even inside a major holding-company platform. Omnicom says it retains the platform, client relationships, data science and AI intellectual property. For buyers, that still leaves practical questions about continuity, access and accountability across the delivery chain.
The World Federation of Advertisers’ 2025 guide to global media agency contracts places GenAI alongside data access, audit rights, governance and holding-company roles as issues requiring stronger oversight. Its useful message is commercial rather than technical: a contract works only when the advertiser owns the governance process and audits the relationship over time.
A portability score should not automatically favor the partner that exposes everything. Some proprietary systems may create genuine advantage that cannot be handed over. The buyer needs to price that dependency consciously and compare it with more open alternatives. A polished demo proves what the partner can do today; the exit test reveals whether the resulting capability will ever belong to the organization.
The best partner should survive a clean handover
Portability can sound hostile when it enters a pitch because it forces both sides to discuss termination before the relationship begins. Mature agencies can treat it as evidence of confidence. A partner that documents workflows, trains the internal team and separates client-specific assets from its own intellectual property is making its value easier to trust.
That does not make the agency disposable. It changes what earns renewal. The relationship continues because the partner keeps improving the operating model, brings outside judgment and solves harder problems, not because switching costs have trapped the client inside undocumented infrastructure.
Clients also have obligations. They need named owners for the accounts and data the agency uses. Internal teams must participate in training, approve documentation standards and maintain the assets transferred to them. Portability fails when the client demands an exit package but never builds the capacity to receive it.
The commercial tension is real. Agencies need room to protect reusable expertise. Advertisers need confidence that their own context, learning and operational history will not vanish with a roster decision. A clear boundary makes both positions stronger because it turns an eventual dispute into a design choice made at the start.
AI will keep moving agency value away from visible production and into connected systems. The winning partnership will not be the one with the most impressive black box. It will be the one whose value remains obvious even when the client is free to leave.