Advertising Week New York brought together brands, agencies, publishers and technology partners for four days of conversation, and it will come as no surprise that artificial intelligence was the common thread running through nearly every stage. What was encouraging this year was the shift in tone. The discussion has moved on from what AI might one day do to how it is already changing the way media is planned, bought, produced and measured.
Another difference from last year stood out to us in particular. No one is hiding the fact that AI is part of how the industry works anymore. Agencies, their employees, vendors and technology partners spoke openly about using AI across most of their workflows, and it is becoming accepted that AI will be built into nearly every process, provided the rules of the relationship, including contract terms, are upheld.
For those of us in marketing procurement, that shift has a large impact. Many of the most useful moments of the week were not about the technology itself, but about the commercial and operational questions that sit around it: who owns the data, how agency effort is valued when work becomes faster, who is accountable when automated decisions go wrong, and how success is measured. These are the questions that will shape agency relationships, scopes and contracts over the coming year.
We attended sessions across the Tech, Marketplace and Innovation stages and the industry lounges, and have identified six themes that we believe procurement teams, and the agency partners they work with, will want to have on their radar. Each is grounded in what was said on stage, followed by practical steps to consider.
The six themes at a glance
- Protect your data in the contract.
- AI is making agency work faster, and remuneration models need to evolve with it.
- Who owns AI mistakes is still contested.
- More of the media decision happens before the auction.
- Agency AI tools can lock you in.
- Brand building is back in the scorecard.
1. Protect your data in the contract
What we heard: A consistent message across the week was that AI models themselves are quickly becoming a commodity, and that a brand’s advantage will come from the quality of the data it brings to them. LiveRamp’s CEO described data as the engine and AI as the dashboard on top of it and encouraged brands to strengthen their first-party data and form second-party partnerships with other companies.
He then shared real contract language from a technology partner that allowed the partner to use client data for “research and development” and to “improve” its own products. In practice, that can mean training the partner’s models on the very data that differentiates the brand. His advice rested on three things: 1) consent, 2) contracts and 3) clean rooms, with clean rooms offering a technical safeguard where contract terms alone may not be enough. A GWI-led panel added that a vendor unwilling to explain what data sits behind its outputs should be treated as a red flag, and brand-side speakers described the AI review boards they now run on every new vendor, covering data storage, privacy compliance and security.
What to consider: Review data-use, model-training and “product improvement” clauses across agency, AdTech and AI agreements, including the vendors your agencies use on your behalf. Make clear that your data may not be used to train or improve third-party models without explicit consent and include data return and deletion terms at the end of the relationship.
2. Work is getting faster, and remuneration needs to evolve
What we heard: Agencies shared concrete examples of time savings. Horizon Media described turning a launch brief into a platform-ready campaign setup in about 45 seconds, a task that used to take hours. A retail media panelist said post-campaign reports that once took weeks now arrive about 90% complete. Meta’s agency lead said buyers see AI handling 15 to 30% of their work, and Goodway Group described search buyers whose time has shifted from mostly setup to mostly strategy.
Agencies were also candid that the commercial side has not caught up. On the agency leaders’ panel, one speaker said outcome-based models have not been worked out and that client expectations do not yet match reality. Goodway added that some clients have priced AI efficiencies into fees before the capabilities were ready, while agencies still have to fund the investment themselves. Fluency pointed to a newer cost line: AI models carry a real cost per transaction, which grows with volume, contrary to most general procurement norms.
Our perspective: There is no single answer, and moving away from FTE-based fees overnight is neither realistic nor always desirable. During this transition, we recommend a blended approach:
- Deliverables-based fees where the output is well defined. Creative, production and content work often lend themselves to pricing by asset or deliverable, which lets both sides benefit as AI shortens production time.
- A hybrid model for ongoing teams. For media, strategy and account teams, keep a resourced team, but expect it to include AI agents and tools alongside people. Agree upfront that hours will reduce over time as efficiency and quality improve and review the resourcing plan at set intervals.
- Transparency on what is automated. Ask agencies to show which tasks are now handled by AI, what that time is reinvested in, and how technology and AI usage costs are charged.
This gives agencies room to invest while making sure the efficiencies are shared fairly over the life of the relationship.
3. Who owns AI mistakes is still contested
When the industry talks about “responsibility” for AI, it is really talking about liability and accountability: who bears the financial loss, who answers to regulators and who carries the reputational damage when an automated decision goes wrong. That is why this question matters so much, and why it is one of the most negotiated points in agency agreements today.
What we heard: GWI shared research among senior marketing professionals showing that around eight in ten feel confident using AI for audience profiling, and many are already making business decisions on AI-generated outputs. Yet almost all reported having caught AI producing inaccurate or misleading results, and only about a third regularly verify what comes back. When asked who is accountable if AI gets it wrong, there was no consensus: respondents pointed to the technology vendor, IT, procurement and marketing.
Speakers on stage approached the question from different angles. Brand-side panelists argued that the brand ultimately carries the risk (not liability), because its name is on the work, and that any team should be able to explain where every AI-informed recommendation came from. Digitas’s chief investment officer said that people remain accountable for the work and that an agency cannot simply blame the agent or the AI it is built on. Technology partners focused on control: Fluency described writing business rules for anything predictable and keeping a human decision for anything that is not, and Meta stressed authenticated access, so it is always clear who, or what, made a change. Goodway Group described keeping tasks with almost no margin for error, such as delivering an exact budget in an exact market, on fixed rules, and only handing more to automation after it had run error-free across many campaigns.
What happens in practice: In contract negotiations, the prevailing principle is that the party that controls the work owns the liability. In our experience, agencies tend to push back on liability for AI-driven outcomes, accept some share of it, and look to pass the remainder on to the AI platform or model provider. That chain can leave the brand carrying the reputational risk without matching protection in the contract, particularly where the underlying technology provider’s own terms are narrow. On top of that, the provider’s terms are often unknown until a crisis happens.
What to consider: Agree liability and accountability based on control rather than leaving it to be settled after something goes wrong.
- Map control for each AI-driven activity: who configures it, who approves it and who monitors it.
- Set clear thresholds: which decisions an agent may take alone, and the spend or risk level at which a person must sign off.
- Separate read access from write access in platform and agent permissions and expand autonomy in stages as performance is proven.
- Require an audit trail that records what the agent did and on what data.
- Flow obligations down: make sure the agency’s commitments on accuracy, data and compliance extend to the AI vendors it uses on your behalf, and that the agency remains your accountable counterparty.
Industry bodies, including the IAB Tech Lab, are working on standards for agentic advertising, which may help settle some of these questions over time.
4. More of the media decision now happens before the auction
What we heard: For years, programmatic decisions were made largely on the buy side, at the moment of the auction. The Intelligence Shift panel, with Magnite, Dentsu, Roku and CVS Media Exchange, described a clear move upstream. Media owners and retailers are increasingly applying their own first-party data to inventory before it ever reaches a buyer’s platform. Roku described the viewing signals it captures as the operating system on the TV, and the retail media partnership it launched with CVS earlier this year, pairing Roku inventory with CVS shopper data.
Dentsu said its tests across consumer goods and beverage brands showed up to twice the return on ad spend, or 40% more efficient CPMs, in part because data no longer needs to be re-matched and re-charged inside the buying platform. The panel closed by agreeing that the most important programmatic decisions should now happen before the auction.
Other sessions explained why this is accelerating. Seedtag’s CEO noted that around 20% of global browser traffic already blocks third-party cookies and that most iOS impressions carry no device identifier, pushing the market toward context and publisher data. PrimeAudience warned that a precise brief is often diluted into broad audience segments before it reaches the platform, losing value at each step.
Panelists were also clear that these partnerships only work with the right commercial terms. Publishers often have little visibility of how their inventory performed once bought, and data shared without a way to feed results back to both sides tends to deliver little.
Why it matters for procurement: When data and curation move upstream, so do the fees. A media plan can now include a publisher or retail data fee, a curation fee, supply-side platform fees and buy-side platform and data fees, sometimes for overlapping services. These layers are less visible in a traditional agency reconciliation, and the audit trail can sit outside the agency’s own systems.
What to consider:
- Map the full supply chain for your largest buys, from brief to impression, and identify where data is applied and by whom.
- Ask for each fee layer to be disclosed separately, including curation and data fees on the sell side, and check that you are not paying for the same audience data twice.
- Extend audit rights in agency and platform agreements to cover curated deals and retail media partnerships.
- Agree data and performance feedback terms with media and retail partners, so results flow back to you and your agency.
- Measure incrementality independently before scaling a new curated or retail media route on the strength of partner-reported results.
5. Agency AI tools can create lock-in
What we heard: Nearly every agency on stage described its own AI platform or orchestration layer, and several have been building them for years. On the agency leaders’ panel, one speaker set out a principle we would encourage every brand to adopt: if an agency builds something genuinely differentiated on a client’s data, it should be portable into the client’s own technology stack. Otherwise, it becomes lock-in.
Brands are also rationalizing their own tools. A speaker on the AI Is Collapsing the Marketing Stack panel described brands canceling point solutions that duplicate data across legacy stacks and offered a simple adoption test: if a team is not using a new tool daily or weekly after 30 days, it is unlikely to stick.
What to consider: Clarify ownership of any models, agents, prompts and workflows built on your data and include portability and transition support in exit provisions. For AI tools you license directly, consider adoption and usage metrics alongside price at renewal.
6. Brand building is back in the scorecard
What we heard: In the FQ Lounge, Blizzard Entertainment’s CMO described moving over two years from roughly 70% performance spend to roughly the inverse, after concluding that performance marketing captures demand but does not create it. Another CMO on the panel described having to prepare her board for brand investment that would appear not to be working for up to 12 months before results came through.
On the Tech Stage, Meta noted that once AI agents are given margin and loyalty data, they can optimize for the most profitable customer rather than the highest volume of conversions, and agency leaders said AI makes customer lifetime value far easier to model. The Show Me the Outcomes panel, with iProspect, Adverb and ABCS Insights, proposed a shared model of accountability: the brand sets the outcome, the agency sets the learning agenda, the media owner is responsible for signal quality, and an independent party validates the result, so no one grades their own homework. iProspect added a warning that applies to every scorecard: do not confuse what is easy to measure with what is meaningful.
What to consider: Align agency KPIs and incentive fees with business outcomes rather than channel metrics alone and avoid separate scorecards for brand and performance that pull in different directions. Share the margin and lifetime-value data agencies need to optimize toward profitable growth and keep independent measurement in the contract.
Finding the balance between protection and progress
The openness we saw on stage brings its own challenge. Now that AI use is out in the open, the conversation moves to the contract, and that is where we see the most friction. Legal teams, and at times procurement, often seek protections that go further than AI implementation and experimentation can realistically live with. Blanket restrictions on any AI use, approval requirements for every tool, or liability terms an agency cannot pass on to its own technology providers can stall progress, or simply push AI use out of sight again.
The aim should be terms that protect what matters most while leaving room to test and learn. In practice, that can mean:
- Protection proportionate to risk. Hold firm on data rights, confidentiality, brand safety and accountability for outcomes, and be more flexible on lower-risk internal uses such as drafting, analysis and reporting.
- An agreed framework rather than case-by-case approval. Set out approved categories of AI use, the tools in scope and the disclosure expected, and review it at regular intervals rather than renegotiating each new tool.
- Room to pilot. Allow defined pilots with clear boundaries, success measures and an end date, so experimentation can happen inside the contract rather than around it.
- A seat at the table for all sides. Bring legal, procurement, marketing and the agency together early, so the terms reflect how the work will actually be done.
The themes above point to where the protections need to be strongest. The challenge for procurement is to secure them without closing the door on the innovation that brands and agencies are both looking for.
Questions to put to agencies and partners now
The next agency review or contract renewal is the place to turn these themes into terms. Seven questions worth asking:
- Which of our deliverables are now partly or fully automated, and how will that be reflected in fees or team resourcing over time?
- Where in our workflow do AI agents act without human sign-off, and at what spend or risk threshold does a person step in?
- Can any partner in the chain use our data to train or improve its own models? Show us the clause.
- Who owns the tools, models and agents built on our data, and can they move to our stack if we part ways?
- Where are curation, data and platform fees applied in our media supply chain, and what does each layer cost?
- How much of your AI workflow runs on LLMs versus conventional software, and who carries that cost as volume grows?
- Which measurement in our scorecard is independent of the media owner being measured?
As one independent agency put it on stage, signing a contract is very different from onboarding a capability. Procurement is now the function best placed to close that gap.