Future of Marketing Briefing: The brands winning at AI started with process not tech

They’re deploying agents into workflows, mapping tools onto current organization structures and then wondering why the gains are marginal. A strategy agent here, a creative ideation tool there. Useful, but not really transformative. What they’re producing, as one person I spoke to put it, are faster versions of the wrong thing.

The brands that are actually getting somewhere are asking a different question first: not what agent is necessary, but what process needs to change before anything is built. It sounds obvious. Yet, few are doing it. And the ones that are tend to be the ones facing the most acute pressure. The automotive brands rattled by Chinese EVs, the financial services firms staring down fintech competition. The ones, in other words, that don’t have the luxury of moving slowly. For everyone else, the organization chart stays intact and the agents get slotted in around it.

Vinny Rinaldi, who has spent the last year building out AI infrastructure at Hershey’s, puts it better than most: “You’ve got the ability to plug AI on top of anything right now, but if you don’t plug it into something that’s structurally built, your outputs are not going to be relevant whatsoever.”

His analogy for what that means in practice: “When you build a home, it can be the most beautiful thing on the outside, but if you forget to pour the concrete foundation, the first storm is going to blow it over.”

The foundation, in his case, was a year spent cleaning data, fixing taxonomy, and building the infrastructure that makes it possible to prompt a system and get a relevant output in 10 minutes. Not glamorous work. “It’s the most unsexy part of the job,” he said. “And I think it’s the first thing people overlook.”

That point landed repeatedly at Digiday’s Programmatic Marketing Summit last week. On and off the stage, conversations kept returning to the same conclusion: none of it — not the agents, not the agentic workflows, not the innovation — works without a process underneath it. 

At one point, OMD’s managing director of data and technology solutions Emily Proctor put it plainly: “We really start with building a solid foundation in the workflow before we even get into the fun, innovative, transformative things.”

What clients often say they want when they ask for AI, she added, is actually their workflow to be more efficient. The more ambitious applications come in later if the groundwork is there. Jump ahead without it, and the problems compound. “You’re plugging agents into a very siloed workflow,” Proctor said. “Agents and tools that are not talking to each other, no orchestration. That’s where it really starts to break down.”

The guardrails, it turned out, had to come before the automation — not after.

Wes ter Haar, chief AI officer at S4 Capital’s Monks, watched this play out in a meeting with a group of CMOs recently. The board pressure was there. The mandate to cut costs — and in some cases headcount — was there. What wasn’t there was any clarity on what comes next. “They’re very much still in the ‘how do we use this, how do we adopt this,'” he said. 

It’s hard to answer those questions when CMOs haven’t yet asked the one that matters most: not how do we use this but what needs to change about the way we work before we do.

General Motors is a useful illustration of why that sequencing matters — and also why it isn’t always linear. Ter Haar’s team started working with General Motors on the obvious solvable problem: content costs too much and takes too long. That work generated real results. But it also surfaced the harder question that only became visible once the first one was solved. Once the supply chain bottleneck was removed the question of what to actually make with that capacity became more urgent, not less. 

“Just doing more doesn’t necessarily help the business,” ter Haar said. His team is now working through the next phase — persona agents and consumer versioning tools that inform what gets made, not just how fast it gets made.

Sometimes you can’t see the real problem from the boardroom. You have to build something first before it comes into view. True as that may be, the brands getting the most out of agents and agentic workflows are the ones that arrive at the structural questions sooner rather than later (ideally before they start building at all).

That’s the argument Market Singer, who leads marketing transformation work at Deloitte Digital in the U.S., keeps making to clients: “It’s not what agent do I need to build,” he said. “It’s what process do I need to change. And then go build an agent.”

In at least three instances, that argument is working.

The first, a large U.S. retail advertiser, has been in the process of shutting down most of its external agency relationships after building sufficient in-house AI capability. Not all of it — targeted support in specific areas remains — but the structural dependency on external agencies is going. The second is restructuring its entire global marketing function from scratch, applying what Singer described as a supply chain mentality to rethink how creative, media and data interact at every level. The third, a large tech company is consolidating a fragmented portfolio of agencies and freelancers into a more coherent model. 

In all three cases, the starting point wasn’t a conversation about which agents to deploy.  “The common theme amongst all of them,” said Singer, “is workforce transformation first.”

None of this is an argument against agents. It’s an argument for earning them. The brands that have done the process work — the data cleaning, the taxonomy fixing and the organizational restructuring — are finding that the agents, when they eventually arrive, actually work. The ones that skipped ahead are finding out why that matters.

—Sam Bradley contributed to this report.

Inside the AI lab taking on the agency model one brand brain at a time

In keeping with the AI theme (yes, there’s more) we spoke to Rob Wrubel, founder and managing partner of Silverside, the innovation lab behind some of the most-watched AI campaigns of the last two years.

The problem with the traditional agency model isn’t talent or ambition, it’s architecture, said Wrubel. But large agencies are built for a world where scale required headcount, and headcount required structure. “You spend four weeks just scheduling meetings,” Wrubel continued. “You’ve got everybody in a Slack channel with 50 people, not sure who has creative decision rights.” AI removes most of that friction. A three or four person agile team, as he puts it, can now do “what maybe hundreds of people could do before.” That’s not a marginal efficiency gain so much as a structural argument against the model itself. 

“I think the holding company and the agency model — there’s a lot that’s going to get reinvented over the next few years,” Wrubel said. “It’s a very tricky one to defend, that larger kind of model structure.”

The brand brain is the destination

What Silverside is building toward — and what the more progressive brands are starting to demand — is a trained AI system built around a brand’s assets, visual identity, product knowledge and compliance rules that can produce content across channels without starting from scratch each time. As Wrubel explained: “Once you learn and work with a creative director or brand strategist and embed their insights and their qualities and what they believe is important into an AI system, you really are building, in a sense, that brand brain that can then do a lot of helpful work.”

This Panasonic example makes it concrete: a Porsche-designed washing machine, launched across all of China, built entirely in Chinese, three and a half weeks from brief to live. The Svedka fembot campaign is the other proof point: two broadcast spots in three weeks, with a trained system that then powered point-of-sale, social and TikTok integrations when the Super Bowl came around. One system. Multiple channels. No re-briefing.

The business model is shifting under everyone’s feet

The time-based fee — the bread and butter of agency and studio economics for decades — is giving way to something Wrubel calls a “tech-powered service model.” The progression he’s watching in real time: brands come in for a proof of concept, get comfortable with the results, then ask if Silverside can just run their social permanently because it’s 60% cheaper and 90% faster than the alternatives. “Then once they got used to it,” he said, “they said, well, can’t you just run and manage our social all the time?” From there, the brand brain becomes the infrastructure — and the question shifts from what do we want to make to how do we want to operate.

“It’s a new hybrid breed of company,” Wrubel said, “combining the best of craft and technology and brand expertise, but with systems that are automated.” His argument against pure SaaS as the answer: the technology moves too fast for brands to navigate alone.

“People don’t just need technology that no one helps you with,” he said. “You need the expertise and the services as everything changes so quickly.”

The model that wins, in his view, is the one that bundles both — and owns the client relationship across the entire journey from concept to scaled production infrastructure.

Transformation accelerates in downturns

Every major technology shift of the last 25 years — SaaS, cloud, paid search — emerged not from optimism but from pressure. “When SaaS emerged, when cloud emerged, when paid search emerged, it all came out of economic downturn,” Wrubel said. His prediction is the same pattern plays out with AI. Most brands don’t have to transform yet. Things are working well enough. But “at a moment when things are more challenging economically and from a growth standpoint, you will see the adoption rate move at an accelerated pace,” Wrubel added. The brands quietly building their foundations now, in other words, will have a significant head start on the ones still running proof of concepts when that moment arrives. 

“That’s just the nature of technology adoption,” he said. “It happens in recessions.”

What we’ve heard

“We’re not reducing headcount. We’re not reducing costs. We’re changing what people do and we’re refocusing on the higher value things. That’s super important. We had a bunch of panels with clients recently and a lot of it was about agentic and AI but one consistent thread through the whole thing was that they’re great but ‘I buy relationships, people and strategy.’ They still want teams of people who understand who understand their business as a marketer and can therefore help them think through how they can drive that.”

—Tyler Romasco, evp of of commercial at OpenX at eMarketer’s Ad Buyer Strategies Summit this week.

Numbers to know

78%: Percentage of millennials that will use a second screen during the World Cup matches, more than Gen Z (70%).

£3.99 ($5.39): Monthly cost of the new ad-free subscription for TikTok in the U.K.

€135 billion ($158 billion): Forecasted total that the European creator economy could reach by 2032.

1.8X: Affiliate link CTR from YouTube creator integrations is 1.8x higher by the eighth brand partnership than it was on the first.

What we’ve covered

TikTok launches MCP server to let AI agents run campaigns

TikTok has launched an MCP server that lets AI agents connect directly to its ads system. This allows campaigns to be created and optimised automatically without using Ads Manager.

OpenAI makes it easier to run shopping ads in ChatGPT

OpenAI is making it easier for retailers to turn product catalogues into ads inside ChatGPT using automated “product feed” tools. The aim is to let brands quickly launch and scale shopping campaigns without manual setup.

Ad tech is lining up behind OpenAI (it’s been here before)

Ad tech companies and platforms are increasingly aligning behind OpenAI as ChatGPT becomes a new ad distribution channel. This mirrors a familiar pattern, where intermediaries initially rush in before the platform ultimately builds and controls its own advertising stack.

‘There’s a big shift’: Amazon is turning the upfront into a pitch for its ad tech, not just primetime

Amazon is reframing its annual upfront pitch to advertisers around its ad tech stack, not just its premium TV and streaming inventory. It is positioning tools, data and programmatic capabilities as the main value, signalling a shift from content-led selling to technology-led ad sales.

What we’re reading

Elon Musk’s Grok Is Losing Ground in AI Race

Anthropic has signed a deal to use computing capacity from SpaceX’s data centers to power its Claude AI models. The partnership also underscores how far Elon Musk’s Grok is lagging rivals like ChatGPT and Claude, with SpaceX now leasing excess AI infrastructure to a competitor instead of using it internally, according to The Wall Street Journal.

Components of AI-enabled advertising

AI-enabled advertising is becoming a fully automated system across targeting, creative generation, measurement and media buying. The end goal is a model where AI agents autonomously run and optimize campaigns, potentially opening advertising to far more businesses, per Mobile Dev Demo.

YouTube Upfronts: 4 Alex Cooper Projects and Dude Perfect Among Avalanche of New Shows

The line up for this year’s YouTube upfronts signals how the platform is positioning creators — not traditional Hollywood talent — as the centerpiece of its entertainment and advertising strategy, according to The Wrap.

LinkedIn is cutting 5% of its workforce in a reorganization

LinkedIn is laying off roughly 875 employees — as part of a broader reorganization despite continued revenue growth. The cuts reflect how tech companies are reshaping teams and spending priorities around AI infrastructure, efficiency and higher-growth business areas, according to Quartz.