AI is making agency work faster to produce, but most brands and agencies haven’t agreed on who benefits. Bruno Gralpois draws on new WFA research to show why clawing back every efficiency backfires, and lays out a three-phase plan for AI agency compensation that measures, rewards, and shares the gains.
A procurement lead at a global CPG company opens a renewal proposal from her lead creative agency. The fee is flat to last year. She has heard, informally, that the agency now produces first-round adaptations with generative tools in a fraction of the time it used to take. So she asks the obvious question: where are the savings?
The agency’s response is equally reasonable. It has spent the past year licensing new platforms, retraining hundreds of people, hiring prompt engineers and AI governance specialists, rebuilding production workflow, and absorbing the cost of experiments that didn’t pan out. If anything, the agency counters, it is under-recovering. Both are right. Neither has the data to prove it.
Conversations like that are happening in thousands of client-agency relationships right now, usually without a framework, often without numbers, and sometimes without being spoken out loud. The result is an erosion of trust at the moment when both parties need each other to innovate.
What the WFA Study Tells Us
The new WFA and Agency Mania Solutions study, “How Brands Define, Scope and Reward Agency Work,”* puts numbers on the gap. AI has not yet reshaped the commercial relationship. Half of the largest global brand advertisers in the survey report minimal impact of agency AI use on fees, retainers, or scope, and 13% report no impact at all. Only 16% describe the impact as significant or very high. Just 20% have started evolving their commercial or remuneration models, although 61% intend to.
Yet AI is already reshaping how brands judge agencies. 82% agree that an agency’s poor use of AI could lead them to reconsider keeping it on the roster. 66% consider effective AI use an essential capability, table stakes rather than a differentiator. 54% say they are seeing at least some AI-generated efficiencies from their agencies.
Here’s where it gets uncomfortable. When asked how those efficiencies come back to them, we learned that:
- 33% receive reduced fees or credits.
- 25% have not yet had the conversation with their agency.
- 23% say the agency keeps the gain, with no change to fees or resources.
- 23% don’t know or have no visibility into how their agencies use AI.
- 23% get higher quality or faster turnaround at the same cost.
- 17% get more agency time or resources at no extra cost.
- 8% have moved toward value- or output-based pricing.
The picture is just as unsettled on who pays for the technology. 62% of brands say their agencies license third-party AI tools for their work, 51% say agencies use proprietary tools, and 34% provide or license AI themselves. Nearly half (46%) say AI costs are simply absorbed in agency fees or overhead, while 23% are still figuring out how to handle it.
Put simply: AI is now in the work, it influences roster decisions, and it creates value. But in roughly half of relationships, nobody has agreed on how that value is measured, who invested in it, or how it should be shared.
Clawing It Back Is the Wrong Instinct
The tempting response, especially under budget pressure, is to treat AI as a discount. If work takes less time, pay less. In hour-based and FTE-based models, that logic looks airtight, and it is precisely why those models are struggling. As the WFA study notes, AI makes time an increasingly weak proxy for value. A purely punitive approach has three predictable side effects.
First, it kills the incentive to innovate. If every hour an agency saves becomes an immediate fee cut, the rational move for the agency is to save fewer hours, or to stop telling you about them. You get less transparency, not more.
Second, a punitive approach ignores the investment. Agencies are paying for tool licenses, security reviews, legal and IP protection, employee training, and new specialist roles, among the most common costs. Some of those costs are shared across clients; many are not. Early productivity gains are often offset, at least initially, by the cost of getting there.
Third, the approach confuses efficiency with value. A campaign that’s produced 40% faster isn’t automatically worth 40% less. It may launch earlier, test more variants, produce better work, improve analysis, or free senior talent for strategy. Those are gains too, and in many cases, they are worth more than any immediate fee reduction.
It’s worth noting that the opposite extreme is just as damaging. If agencies quietly pocket the gain while billing the same hours, that tactic eventually surfaces, usually in an audit or a pitch, and the trust deficit that follows costs far more than whatever margin was kept.
Between the two extremes, a third way stands tall: a transparent, collaborative, measured, jointly governed model in which AI gains are visible, investments are recognized, and the dividend is shared in a way both sides agree to in advance. How do we bridge the divide?
What Brands Should Do About AI Agency Compensation
No matter how tempting it might be to jump into negotiations with agencies, none of this requires a new compensation model on day one. Instead, a collaborative approach requires structure and careful consideration to set up a solid foundation for a joint effort that benefits both parties. Here’s a practical three-phase sequence to help brands get started immediately.
Phase 1: Establish Visibility and a Baseline
Name an owner and form a joint AI working group. Assign a client-side lead, typically from agency management or marketing procurement, paired with a senior agency counterpart for each major agency. Without an owner, AI stays a hallway conversation. Keep marketing, legal, IT/security, and finance in the loop and involve them as needed.
Ask each agency to map AI across your campaign/project workflow. Come up with a list of questions. For example: where is AI used today, from briefing and research through ideation, production, optimization, QA, and delivery? For each step, capture the tools used, tasks affected, level of human oversight (if any), and areas where AI is excluded and why.
Build an AI tool and license register. Document which tools are used on your business, who licenses them (agency, client, or shared), whether costs are allocated across clients, where your data goes, and how outputs are protected. Reconcile that information against tools you already license, so you don’t pay twice for the same capability.
Set a baseline before you measure gains. Pick a manageable set of relevant deliverables (for example, a social asset, a 30-second video, a promotional email, a market research summary) and record today’s cost, hours, turnaround, revision rounds, and quality scores.
Have the conversation you haven’t had yet. Open it as an exploratory exercise and joint value-creation discussion, not as a negotiation or a basis for preliminary commitments.
Phase 2: Measure, Test, and Verify
Agree on a short list of efficiency and value metrics. Keep it realistic. Typical candidates: cycle time per deliverable type, cost per asset, output volume, first-time-right rate, revision rounds, and a quality or effectiveness indicator (based on the discipline). Balance speed and cost metrics with quality ones, so efficiency never comes at the expense of the work.
Run two or three structured pilots. Choose well-defined, repeatable workstreams where AI is likely to matter (adaptations, versioning, transcreation, reporting, social content). For the pilot, define the hypothesis, the success criteria, the duration (60–90 days), and, critically, how gains will be shared. Consider a modest innovation pool, co-funded or funded from a portion of early efficiency gains, to cover prototyping, tools, and training tied to the effort.
Review the results. Agree on how results will be validated, such as time and cost data from agency systems, sample-based reviews of deliverables, and spot checks by your internal team (or a neutral third party if preferable).
Remember to look at your own side of the process. The WFA study makes a point that too many clients skip: AI cannot compensate for poor briefs, fragmented feedback, or slow approvals. Measure client-caused rework and delays alongside agency efficiency. Some of the biggest gains will come from fixing how the brand, not the agency, works.
Phase 3: Codify, Contract, and Govern
Add an AI addendum to the MSA. Most master agreements predate generative AI or are limited in scope to address regulatory, brand safety, corporate compliance, and risks. Update the agreements to address:
- Disclosure: an obligation to disclose material AI use on the brand’s business and maintain the tool register.
- Data protection: what client data may and may not be used in AI tools, including restrictions on training third-party models.
- IP and ownership: ownership of AI-assisted outputs, indemnities, and clearance of training data and third-party rights.
- Cost allocation: who pays for which tools, how shared-platform costs are allocated, and what can be billed separately.
- Gain-sharing: the agreed principles for measuring and sharing efficiencies, and how they will be reviewed.
Create a specific AI service level agreement (SLA). The MSA sets the rules; the SLA governs day-to-day. A practical AI SLA might include quarterly AI usage and impact reporting against the agreed metrics; turnaround and quality standards for AI-assisted deliverable types; human-in-the-loop checkpoints and accountability for final outputs; incident reporting (response times for data, IP, accuracy, or brand-safety issues); a change process for introducing new tools or new AI-enabled services on the business; and training and capability commitments for teams assigned to the account. Make AI a standing agenda item in QBRs and annual planning.
Include AI topics in your performance evaluation, both ways. Brands should be rated on whether they enable AI-driven value (better briefs, data access, faster decisions) as much as agencies are rated on delivering value.
Adjust scope and compensation where the evidence supports doing so. Use pilot data to re-price specific deliverable types, move suitable workstreams to output-based fees, or redefine the scope for the next fiscal year. Change what the evidence justifies and leave the rest until it matures.
Agreeing on a gain-sharing model is critical. Decide, explicitly, how verified efficiencies will be allocated. Brands will land on a blend such as a defined share returned as savings, a share reinvested in innovation, and a share retained by the agency as reward and recovery of its investment. What matters is that it is measurable and agreed upon.
Share the Dividend, Don’t Fight Over It
If the plan works, a brand should be able to answer, for each major agency, five questions without guessing: Where is AI used on our business? What are the tools and who pays for them? What has AI improved and by how much, and how do we know? How are gains shared, and does the contract spell that out? When do we next review it together?
That model is a very different position from the one most brands are in today. It replaces suspicion with evidence and guesswork with transparency. It gives agencies a clear reason to bring their best innovation to your account first.
AI will reshape agency economics; the WFA data leaves little doubt about the direction. The open question is whether it will happen through a series of tense renewals and quiet margin grabs, and even painful agency reviews, or through a deliberate, transparent partnership in which both sides measure the gains, recognize the investment, and agree on how to share what they create together.
Agencies may not yet be paid differently because of AI. But brands that build trust and transparency around AI will get more from it: lower cost when warranted, faster and better work where it matters, and partners who are motivated to keep innovating on their behalf.
That’s the real AI dividend, and it only pays out when it’s shared.
*Source: WFA–Agency Mania Solutions research, “How Brands Define, Scope and Reward Agency Work,” September 2026. Online survey conducted in May 2026 of 69 multinational companies across six industry sectors, compared with similar studies in 2011, 2014, 2018, and 2022.
Our clients’ ongoing accomplishments result from cutting-edge practices in the area of client–agency performance evaluations. See how stronger relationships contribute to better marketing.
To drive greater value from your client–agency relationships, check out industry reference Agency Mania.
And sign up for our complimentary, bimonthly Industry Update.
Author, Speaker, Thought Provocateur, Client-Agency Guru, Entrepreneur, Innovator
Bruno Gralpois is the co-founder of Agency Mania Solutions, a premier service and technology firm specialized in helping companies realize the transformational value of managed partnerships. Bruno is the author of bestseller Agency Mania and the former chair of the Association of National Advertisers Client/Agency Committee and a faculty member of the ANA School of Marketing.






