What we heard from ten agencies in New York, and what marketing procurement should do with it
By Christine Moore | AdForum Summit, New York, May 2026
In mid May, the AdForum Summit brought a group of search consultants through ten agency groups in three days in New York. Agencies like MSQ, Publicis Production, DEPT, Stagwell, Dentsu, WPP, Critical Mass, PMG and Media.Monks opened their doors (both physically and figuratively) to us.
These group visits would be immensely useful for Marketing Procurement looking to support Brands with agency selection, innovation and research, but it is very hard to implement in real life. Therefore, we have decided to bring our findings to you.
If you work in marketing procurement, you already know the headline: everyone has AI. That is no longer news, and frankly it is no longer useful.
The real question for anyone who buys agency services is how you tell these firms apart, and how you find the right partner when some of the best options are not on your list in the first place.
That second problem is real. During one session, a colleague of mine, a veteran consultant who does this for a living, leaned over and admitted that the group presenting “weren’t on my radar.” That group happens to manage billions in media. If the professionals who evaluate agencies all day can miss a firm of that scale, imagine what a procurement team juggling twelve categories is missing. So, this piece is about two things: what the agencies sounded like when you put them side by side, and what that tells us about how to search the haystack.
Where everyone agreed
Put ten agency presentations back to back and the vocabulary converges fast. Five themes showed up in nearly every room.
First, everybody has a platform with a name. MSQ demoed Assist, a model agnostic “harness” that sits over large language models. DEPT showed Deptify, its connective tissue across third party tools. Brandtech has Pencil, which it calls an AI operating system for marketing. Dentsu previewed an enterprise grade agentic platform that runs a brief through audience, strategy and media planning in minutes. Media.Monks walked us through a bespoke agentic content supply chain it built for a major manufacturer client. Critical Mass leans on its holding company (Omnicom) stack plus its own content orchestration engine. The names differ. The demos rhyme.
Second, everyone gave the same sermon on humans. AI is the scaling mechanism, not the idea. DEPT’s product leadership put it best when asked whether AI could ever produce a genuinely crazy, category defining campaign: it cannot, because “it can’t read the room. It’s not in the room.” Every shop, without exception, positioned human taste, judgment and craft as the thing that separates their output from the sea of generated slop. When ten competitors agree on something this loudly, treat it as table stakes, not differentiation.
Third, media and creative are being welded back together. Dentsu talked about intelligent story systems and showed a team that turned more than 360 briefs into roughly 53,000 assets in a year for a global technology client, with production and creative merged into one unit. VML’s most talked about campaign for a heritage skin care brand came through the media relationship, not despite it. After two decades of unbundling, every single group is now offering the reassembled version.
Fourth, the machines are now an audience. Dentsu was the most explicit, unveiling a large language model optimization practice and arguing that brands now have two targets: people and models. You are “seen first and felt second”, and if you miss either one, you lose. Jellyfish, inside Brandtech, calls it Share of Model and cites research claiming top quartile brands on that measure grew several times faster than the market. In the WPP room, the creative leadership pointed out that a certain community platform feeds a huge share of LLM training data, which is partly why an honest reviews campaign was built for one of their clients. Different labels, same insight: your brand’s next gatekeeper is an algorithm that summarizes you.
Fifth, commercial models are shifting toward outcomes. MSQ said plainly that it is less focused on day rates and have started to put meaningful fees at risk against business KPIs, including sales. PMG’s leadership framed its entire positioning around outcomes, not outputs, and accountability, not excuses. Brandtech said its best client conversations are about business growth, “not asset value or scope negotiation or what’s the commission level”. Whether every client and agency can agree on these more sophisticated remuneration models when the SOW lands on your desk is another matter, but the direction of travel is unmistakable, and procurement should hold them to it.
Where they split
The convergence is real, but the differences are where the useful information lives. Four areas of differentiation stood out to us.
Build versus connect. Brandtech sits at one extreme: it built Pencil starting in 2018 and has had over ten million ads produced and several billion dollars of media run through the platform, and sells it both in partnership with its agencies and as a stand-alone software. One global retail company has reportedly mandated the platform across its entire agency roster. Monks sits nearby, building proprietary client owned infrastructure. DEPT sits at the opposite pole. Its new chief product officer argued that holding companies spending $350 to $400 million a year on technology cannot outbuild a single point solution startup that raises more than that in one year alone, so DEPT instead builds only “the arrows” that connect best in class tools into a client’s process. It lets the client access the best-in-breed at any point in time and doesn’t bet on one or two horses in the AI race. MSQ and Dentsu sit in between, building platforms but stressing model agnosticism, composability and zero lock in. As a buyer, this is a genuine strategic choice: are you buying a product, a pipe, or a philosophy of assembly? Each has different switching costs, and the wise buyer need to explore those costs upfront, and not when contemplating an agency change.
Human in the loop, or not. MSQ told us it deliberately built its platform with a human in the loop as a core principle. DEPT said “human in the loop” is a fallacy, because it implies you have outsourced the work and kept a person around just in case. DEPT’s alternative is a task-by-task framework: some work gets worse the more human it is, like data collection, where empathy just injects bias, and should be fully automated; other work, like insight generation, needs human empathy and should be augmented, with the tech in the loop rather than the human. These sound like semantics until you realize they produce different org charts, different staffing plans and different fees.
Volume versus restraint. Monks and Brandtech lead with scale: millions of ads, billions saved in physical production for clients with that kind of need, content engines that never sleep. Another agency barely demoed a platform at all. They showed campaigns, awards, craft, and put a client on stage. When I asked around the room afterward, most of us noticed the same thing: the agencies most confident in their creative product spent the least time on technology.
Selectivity as strategy. Two groups openly told the room which business they will not chase. One agency said it had disqualified itself from a high profile global integrated pitch because the client’s demand for offices in every market and its buying style did not fit. Another Group was even blunter: when a review “is a pure procurement exercise,” or “the client only wants efficiency”, they not going to be the best fit for that client, and they now win when the marketer genuinely wants transformation. There is an irony in saying that to people like pitch consultants, but the discussion should happen up front instead of discovering a mismatch six months into a review. An agency’s disqualification list tells you more about its strategy than its credentials deck does. When Procurement builds a Brand roster, it is important to listen “between the lines” and not only read the prospectus.
And one small caveat about press releases and industry press. The same global brewer’s media review was claimed as a headline victory by two different agency groups during the same week, with each group’s version arguably true because the outcome was split across different remits. If you are relying on trade press win lists to build your long list, you are reading PR about marketing. Go one layer deeper, always. Call someone in New Business at an Agency that you are hearing about in the press.
What this means for the haystack
Here is the uncomfortable part for procurement: the most interesting things we saw did not come from the two or three names that would survive a standard RFI screen on size and geography. A ten-year-old, largely employee-owned firm is crossing several billion dollars (!) in media under management and acts as global growth partner to one of the most famous AI companies in the world. Several agencies were new to people in a room of professional agency watchers. An independent, runs performance and media for major retail and luxury names and is hosting its own AI and technology venue at Cannes. The haystack now contains needles with billions in billings. Your screening criteria, if they were written more than one or two years ago, will filter these firms out before a human ever looks at them.
So look differently, and ask differently. Based on this week, here is what I would put in front of any agency, big or small, claiming an AI enabled offering:
- Ask for a login, not a demo. Brandtech’s founder offered the simplest fraud test of the week: if it is real, they can give you a login; if all they can give you is a demo, be suspicious. Any platform pitch should survive you poking at it yourself.
- Ask who actually uses it. I asked MSQ how many clients genuinely work inside their platform, and I got an honest answer: it is a mix, from daily users to clients who only ever see the output. That honesty is worth more than an adoption chart. Push for named use cases and how usage is measured. You do not want to buy more “shelfware” for your organization.
- Ask for results, in business terms. One consultant politely told a presenting agency that the previous hour had been “a results-free zone.” Every capability story should end in a number a CFO would recognize, and in a statement of what fees the agency will put at risk against it.
- Ask about token economics now. Brandtech gave the most transparent account of the new cost line nobody has priced: model usage. One agency view is that client fee mix is moving from mostly people toward a rough split of technology and people today, heading toward technology taking the larger share within a few years. Ask how tokens are bought, whether they are marked up or resold, who holds the license, and whether indemnification passes through to you. If the answer is fuzzy, you have found next year’s audit focus.
- Ask about lock in. Model agnostic or tied to one lab? Building blocks or all or nothing? What happens to your data, your templates and your trained brand agents if you leave?
- Ask where the humans sit, by task. Do not accept “human in the loop” as an answer. Ask for their framework: which tasks are automated, which are augmented, and how that shows up in the staffing plan and the rate card.
- Ask what they turn down. An agency that cannot name business it walked away from either has no strategy or will not tell you the truth about fit.
- Ask how the data led to the work. The sharpest piece of feedback all week came from a consultant who told one holding company that they had nailed the data story and nailed the creative story, and never once connected the two. Make them draw you the line (or arrow) from insight to output.
One last observation, from the PMG dinner. Their client guest, a senior media leader at a luxury retailer, argued that the biggest barrier to getting value from AI is not the technology but psychological safety: teams will not test, learn and adopt if they believe every efficiency they find shortens their own runway. The client described dashboards as heat and insight as light, and said the job is turning one into the other.
That applies to agencies too. The firms that impressed us were not the loudest platforms. They were the ones who could put a client on stage, defend a number, explain who does what, and name the business they will not pitch. That is what a needle looks like. Stop scanning the hay for shine, and start pressing on proof.