AI Didn't Replace the Ad Industry's Work. It Just Showed Which Half Was Never That Hard.

Carol Tan, the 36-year-old founder of Singapore's AI Studio, expected her AI-generated content business to be a small-business play. "We thought that mostly small companies would use it," she told CNA. She was wrong. Her clients are enterprise, some multimillion-dollar, some publicly listed, and enquiries have grown four to five times since late 2025.

Agoda's Singapore country director, Pranjal Kalra, told CNA the travel platform used generative AI to scale one recent campaign's video adaptations from 64 to 720 a month, an elevenfold increase that would have been, in his words, "costly and time-consuming" to produce any other way. IdoLive, a Singapore business managing AI avatars for shopping livestreams on Shopee and TikTok Shop, found something almost as stark on the labour-substitution side: an ordinary human livestream host in Singapore costs roughly S$3,000 to S$4,000 a month for four to five hours a day of work. An AI avatar costs S$458 a month and streams continuously. Both, according to founder Kevin Zhang, draw about the same ten concurrent viewers at any given moment, because the audience already knows what it's shopping for and doesn't especially care who, or what, is hosting. Enquiries for AI livestream avatars have grown 20 to 30% a month since the start of 2026.

None of the real, measured gains above touch the layer of the job that involves knowing whether AI should be used at all, and that is where the story gets much less tidy. Charlie Tan, the 28-year-old founder of AI agency Touchmkt, told CNA he actively advises some clients against using generative AI, specifically in high-trust categories, a tuition centre's parent testimonials, for instance, should use real parents on camera, not synthetic ones, because the credibility of the claim depends on it being visibly real. NTU's Assistant Professor Looi Jie Min found the backlash against AI-generated ads is sharpest for large, resourced brands and, within that group, specifically for luxury brands, because a brand selling an aspirational, high-effort story loses that story the moment its ad visibly required no effort at all. NUS sociologist Dr Jun Yu put the sharpest version of this argument on record: when a well-resourced global brand chooses AI despite being able to afford real production, audiences start wondering "what else has been quietly automated, customer service, quality checks, the handling of complaints."

This is exactly why Circles.Life built its own mascot in-house, in both 2D and 3D, rather than generating it, and runs every AI-assisted asset through the identical brand, accuracy and compliance review as everything else it produces, with an explicit policy to scale back AI use the moment it introduces inaccuracies or erodes trust. It is also, tellingly, exactly the logic Agoda's own Kalra used to describe the limits of his own success story: "If AI does not improve the work or cannot be used responsibly, it should not be used." The clearest job actually being displaced here isn't a strategist or a creative director, it is a livestream host earning S$3,000 to S$4,000 a month, replaced by a S$458-a-month avatar. Gartner found fewer than 1% of layoffs attributed to AI in recent headlines were actually due to measurable productivity gains, consistent with a production-layer substitution story that is real but narrower than the industry-wide panic usually implies.

This is where the industry's own biggest, most resourced player has quietly admitted it hasn't solved the problem either. WPP told investors and clients, through CEO Cindy Rose, that outcome-based pricing, paying for what a campaign delivers instead of the hours behind it, was a core pillar of its turnaround. Eighteen months later, by Rose's own account, exactly one WPP client pays that way: Jaguar Land Rover. If the world's largest agency network, with every incentive and every resource to make outcome-based pricing work, has managed to convert one client, the honest conclusion isn't that outcome-based pricing is coming any minute. It is that nobody, including the people betting their entire turnaround story on it, has actually cracked what the unit of value is supposed to be once hours stop being the currency.

What the Singapore examples suggest clients are already, functionally, paying for instead is judgment under uncertainty, the ability to say "don't use AI here" and be right about it, to spot when a luxury client's authenticity is about to collapse before the backlash happens, to build a quality-control layer that catches bad output before a client sees it. Separate research from Stanford's Social Media Lab and BetterUp found 40% of workers received AI-generated output that looked finished but wasn't in just the past month, each incident costing real time to catch and fix. Circles.Life's insistence that "AI does not make final creative decisions or publish content independently" isn't caution for its own sake, it is the actual current product: not the AI, and not the hours it takes to babysit it, but the judgment that decides which AI outputs are safe to ship.

Look at your own account team's AI usage the same way Agoda and Circles.Life do, not as a blanket policy but as a category-by-category judgment call. High-trust categories (testimonials, healthcare, anything where visible authenticity is the actual claim being made) deserve the Charlie Tan test: would a client's core credibility survive this being visibly AI-generated? Production-layer tasks (format adaptation, localisation, always-on presence) are where the real, measured ROI already exists, per this reporting, don't hesitate there. And if you're still pricing AI-assisted work by the hour purely because outcome-based billing feels unsolved, you're in the same position as WPP with 17 clients still on the old model, that's not a failure specific to your shop, it's the entire industry's unsolved problem, worth naming honestly to a client rather than pretending it's further along than one converted account globally.