OPINION | Everything advertising mandarins want to ask about AI (but are afraid to)

Advertising Mandarins

There’s a quiet joke circulating in the ad world: the hottest new AI team at agencies is just a group of “prompt jockeys” stringing together MidJourney, Runway, ChatGPT, and a handful of dashboards and then spending hundreds of man-hours offline to stitch together a single 30-second film. The output is shiny enough to win a pitch, but the economics are laughable. This is the dirty little secret of advertising’s “AI revolution”. Agencies are not building infrastructure. They are playing with wrappers.

Erosion of two defensible moats

Most AI investments in ad land today are wrappers around existing models, creative playgrounds that generate buzz, or PowerPoint decks filled with flowcharts on “AI-enhanced creativity” while actual delivery remains artisanal. Meanwhile, what clients really need is far more fundamental: accurate, licensed, culturally grounded datasets that can be used to train foundation models for advertising. Instead, agencies are applauding vanity projects that showcase novelty but don’t solve the brand’s actual problem: representing consumers truthfully and at scale.

Agencies once had two defensible moats. The first was creative genius. The myth of the Don Draper figure who could crack culture. The second was performance marketing expertise, an operational moat built on buying, optimising, and arbitraging media. Both are eroding.

On the creative side, brands now turn to TikTok / Instagram influencers who generate cultural traction in real time. On the performance side, platforms like Meta, Google, and TikTok automate bidding and targeting, reducing agencies to middlemen with dashboards.

Search for relevance

Agencies are scrambling to rediscover relevance, and AI has become the convenient fig leaf. But instead of grappling with the infrastructure challenge: how to build foundation models that truly understand local consumers, regional dialects, body types, rituals, and aspirations, most are stuck celebrating their ability to make “one ad in five styles.”

It’s reminiscent of the early 2000s, when agencies doubled down on banner ads long after click-through rates had collapsed, or when they clung to Facebook “likes” as proof of engagement. Today, once again, the industry is mistaking activity for strategy. Advertising is not about prompts. It is about people. If your generative pipeline produces an Indian bride who looks like a generic Western model in a red dress, you haven’t solved a client problem. You have created a liability.

What AI should be able to do

AI should be able to give brands accurate, nuanced representations of their consumers. It should allow a shampoo brand to test hundreds of creative variations across hairstyles, languages, and cultural contexts in days, not months. It should enable an e-commerce company to generate product videos that reflect local beauty standards without triggering backlash over unrealistic bodies. But that requires infrastructure, not playbooks. It requires datasets built with provenance, licensing, and cultural annotation. That’s not glamorous work, but it’s the bedrock. Without it, agencies will keep investing hours into patchwork tools while consultancies and tech companies eat their lunch.

Agencies, by and large, are still applauding GenAI vanity projects because they make good headlines. A surrealist beer ad. A fake influencer campaign. A CGI model who does not need hair and makeup. These stunts are echoes of the Levi’s partnership with Lalaland.ai or the Guess AI model that sparked outrage for being more artificial than diverse.

Wrong questions; clients who aren’t fooled

Instead of asking: “How do we build foundation models for advertising?” the industry is stuck asking: “Which prompt should we try next?” Clients are not fooled. They see the hundreds of man-hours behind every “AI campaign.” They know that cost and time are still unsustainable. And they are looking elsewhere, to players who are building the pipes, not just decorating the taps.

The right way forward is to treat AI in advertising like previous infrastructure shifts. Think back to the rise of digital media buying. The agencies that thrived were not the ones producing flashy websites; they were the ones that built systems for trafficking, reporting, and optimisation.

Rebuilding moats through datasets designed for use cases

In AI, the analogue is foundation models tuned to consumer datasets. Not generic text-to-video engines, but models that know how to represent a Filipino family dinner, a South Indian wedding, or an Emirati graduation ceremony with accuracy and respect. That is where the moat will be rebuilt: not in prompt hacks, but in datasets and training frameworks designed for advertising use cases. This is where Clairva is stepping in.

Agencies could choose to lead in this transition. They could invest in building or partnering on foundation models trained on culturally relevant, licensed datasets. They could make the leap from prompt jockeys to architects of AI advertising. But most are not. They’re still fiddling with tools, applauding vanity projects, and mistaking the sizzle for the steak. The moat of creativity is gone. The moat of performance marketing is gone.

The next moat will be infrastructure for AI advertising, the ability to generate accurate, diverse, and trustworthy consumer representations at scale. Agencies can either keep playing with prompts, or they can build the models that will define the next decade of brand communication. If history is any guide, the ones who choose infrastructure over stunts will be the ones left standing.

(Sunil Nair is the Cofounder of Clairva.ai.)

Views are personal and do not represent the stand of this publication.