Volume went up. Quality didn’t. The bottleneck moved somewhere nobody is looking.

Every marketing team I speak to in Southeast Asia has the same story about generative AI: we make ten times more creative than we did eighteen months ago, and we’re not convinced any of it is better. That instinct is now measurable. New research from WARC with TikTok and LIONS Advisory, reported by PPC Land, surveyed 400 marketers and found that while 88 per cent are producing more creative since adopting AI, only 45 per cent see a meaningful lift in quality.

The interesting part isn’t the gap. It’s the reason for it. Two-thirds of those marketers said they brief generative models primarily with demographic data. Nearly six in ten of the same group said demographic segmentation no longer works. They are feeding the machine inputs they have already told researchers are broken. Only 17 per cent consistently brief with anything richer — community context, behavioural signal, actual audience tension.

The report’s framing is the sharpest line in it, and it deserves repeating in plain terms: this is not a technology gap. It is an intelligence gap.

The evaluation layer nobody costed

Here is what changed, and why it is easy to miss.

A brief has never really been a document. It has been the start of a conversation. It went to a strategist, who pushed back on the segment. It went to a creative director, who asked what the person actually feels at the moment of purchase. It went to an art director, who threw out the first three routes. By the time work reached a client, a thin brief had been quietly repaired four or five times by people whose job was, in part, to notice that it was thin.

Those stages were slow, and slowness was the point. They were also expensive, and so they were the first thing compression removed. When production timelines collapse from three weeks to three days, the repair layer goes with them. The brief no longer passes through a series of sceptical humans. It passes into a model, which is constitutionally incapable of scepticism about its own inputs and will produce forty confident variants of a bad idea as readily as forty of a good one.

So the weakness that used to be absorbed by the process now lands directly in the output, at volume, and with the polish of professional work. That is a much worse failure mode than the one it replaced. A bad brief used to produce visibly bad work that somebody caught. Now it produces plausible work that nobody catches, because it looks fine.

Also Read: Moving past the chatbox: The hidden risks of agentic AI and MCP in enterprise infrastructure

Demographics survive because they are already in the template

Why do teams keep briefing with data they don’t believe in? Not conviction — inertia.

Age brackets and income bands are already sitting in the planning deck. They are already in the media plan, the audience field, the campaign naming convention. They require no new work, no new tooling, no argument with anyone. Behavioural and community insight requires all four. Under deadline, the default wins every time, and the default is a demographic.

GWI made this point publicly last week in a rather good line — that age brackets are the laziest segment in marketing, and the differences inside a generation are larger than the differences between generations. They are right, and it is worth noticing how rare that argument is. Look across a week of published content from the marketing-technology category and a clean division appears. The audience research firms publish findings that stop at the statistic. The workflow and listening platforms publish features that start at the publish button. Almost nobody addresses the space between the two, which is precisely where the intelligence gap lives.

Why Southeast Asia feels this first

Because the compression here is more severe. A regional team in Singapore is routinely running six to eleven markets, in several languages, against budgets that would cover two markets in Europe. Local nuance is not a refinement; it is the entire job. And it is exactly the layer that demographic briefing flattens.

Feed a model “women 25–34, urban, middle income” and it will return something that could run in Jakarta, Manila or Kuala Lumpur and land properly in none of them. The output will be grammatical, on-brand and completely generic. Multiply that by eleven markets and a weekly cadence and you have built a very efficient machine for producing content nobody remembers.

Also Read: No fans, no fridges, just paint: ZERC’s founder on cracking SEA’s cooling crisis

Rebuilding the layer, cheaply

The teams pulling ahead have made one structural change: they treat the brief as the product. Not the deck, not the asset — the brief. Whoever controls the quality of the input now controls the quality of everything downstream.

That change is showing up in the tooling at both ends of the pipeline. At the front, platforms that decode live category data into audience tensions and evidence-backed briefs, so the input carries something observed rather than something assumed. At the back, a newer class of businesses built purely around execution — Touchigh, for instance, which helps Chinese cross-border sellers reach American buyers by automating both AI-search visibility and native English social content, and scores that content for predicted performance before it publishes rather than reporting on it after.

That last detail matters more than the category it sits in. A young execution company, serving SMEs at a few hundred dollars a month, has independently arrived at the same conclusion: the judgement call has to happen before the asset ships, or it doesn’t happen at all. Front end and back end are solving different halves of one problem — stopping weak inputs entering the system, and stopping good ones degrading on the way out.

What the operating numbers suggest is that the repair layer can, in fact, be rebuilt cheaply enough to survive a deadline. Pitching a UK hospitality group against considerably larger shops, the agency SAMY documented category planning collapsing from weeks to roughly thirteen minutes per brief, with predicted click-through accuracy running about three times better than human estimation. They won the retainer.

Read that carefully, because the speed is not really the story. The story is that the judgement nobody could afford is now something you can afford on every single brief — and a judgement that only happens when there’s time for it isn’t a standard, it’s a luxury.

The question worth sitting with

The uncomfortable version of the WARC finding is that most teams already know their inputs are wrong and ship on them anyway, because fixing the input costs more this week than shipping the output does.

That maths is changing. When evidence-grade audience intelligence takes minutes rather than weeks, the excuse for briefing on a demographic quietly disappears — and so does the defence when the work underperforms.

So: of the briefs your team wrote this quarter, how many could you trace back to something you actually observed about the audience, and how many were the template with a new date on top?

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post Your AI isn’t producing bad creative, your brief is appeared first on e27.