Your best campaign was also the last campaign where your tactic fully worked.
You just didn’t know it yet.
Here’s what actually happens to every marketing tactic over time.
It starts working because there’s a gap: you know how the mechanism works, your audience doesn’t. That gap is the leverage. Scarcity feels real. Social proof feels organic. The offer feels urgent.
Then you scale it. More spend, more reach, more formats and here’s the part nobody talks about: the moment you reach maximum scale is also the moment your audience starts learning how the mechanism works. They’ve seen it too many times. They recognize it. The counter-reflex develops.
Peak performance and peak immunity growth are the same moment.
Your dashboard shows one of them.
It gets more specific than that. Your best customers — the ones who drive referrals, who pay premium, who tell others — they figure it out first. 12 to 24 months before everyone else. So the overall numbers still look fine. The tactic is still “working.”
Just not on the people who matter most.
Eventually the numbers drop. The team meets. Someone says the creative needs refreshing. Someone says the targeting has drifted. Someone says the channel is saturated.
None of that is why the numbers dropped.
The tactic expired. Not because it was executed badly. Because it was executed so well, for so long, that the audience learned what it was — and the mechanism stopped working.
The brief asks: how do we produce more motion?
The numbers are asking something different: why is motion costing more than it used to?
Those are different questions. Most teams keep answering the first one.
One practical test: graph efficiency — return per unit of spend, not total results — for any tactic you’ve been running more than 18 months. If the line peaked and then declined despite multiple rounds of optimisation, you didn’t have a creative problem.
You passed the inflection point during your best quarter.
The brief was thin.
The craft was beautiful.
Nobody asked the hard question.
That was the agency model for twenty years.
Not because strategy was strong. Because production was expensive.
Shoots cost money. Motion needed specialists. Scale needed headcount. Complexity created cover.
AI just removed the cover.
Campaign recently cited forecasts suggesting that up to 90% of web content could be AI-generated by 2026.
Once production gets cheap enough, the economics change.
Now clients are about to ask the question agencies spent two decades avoiding:
What are we actually paying for?
For a lot of shops, the answer is a layer of output that can now be prompted faster, cheaper, and at scale.
The agencies that survive won’t be the ones using AI fastest.
They’ll be the ones with something harder to replicate: real knowledge of the category, the customer, the culture, the tension.
Most built production capacity and called it expertise.
That bill is coming due.
The only question that matters now is the one volume kept burying:
What do we understand about this market that nobody else has said yet?
It was always the job.
The industry just got paid very well to avoid admitting it.
The shops that grasp this early have a real window.
What I’m curious about now is the client side.
Is that question starting to show up in the room yet?
Most marketers are debating how to optimize for AI answers.
That conversation starts too late.
The real competition begins inside the data models learn from. Not campaigns. Not slogans. Repeated patterns across sources the model treats as credible: media coverage, research, expert commentary, structured knowledge bases.
When the same idea shows up often enough across those environments, it hardens into background knowledge. The default answer.
That quietly changes what brand strategy means.
Winning is less about ranking a page and more about becoming the reference a model reaches for when someone describes a problem. Citation density matters. Structured knowledge matters. But those are tactics.
The deeper asymmetry is upstream.
Some brands have huge citation surfaces: documentation, analyst reports, developer communities, technical writing. Others mostly have recipes and retail listings. Same objective. Very different terrain.
Which is why the real move isn’t publishing more.
It’s owning the vocabulary.
The words people use to describe a problem shape what the model reaches for before retrieval even begins. If your brand defines the terms, not the answer but the question, you’re no longer competing inside the response.
You’re shaping the prompt.
That’s a different kind of moat.
Attention used to be the scarce resource in marketing.
In the answer economy, it’s the question.
WPP didn’t lose 64% of its value because it overspent, but lost it because the holding company model was built on coordination scarcity.
For decades, global brands needed an intermediary to orchestrate creative, media and production across markets. That orchestration was complex, expensive and hard to replicate. The margin made sense.
AI has reduced the cost of coordination close to zero.
That doesn’t eliminate agencies. It removes their structural advantage.
A £500m restructuring simplifies the machine. It does not answer the harder question: what is the machine for?
If a brand can generate, test and deploy at scale inside its own stack, the agency shifts from mandatory infrastructure to optional partner.
Optional changes pricing power. Optional changes leverage. Optional changes negotiation dynamics in downturns.
Until the industry defines what scarce capability a holding company truly owns in a world where intelligence is cheap and distribution is platform-gated, cost savings look defensive rather than directional.
What is a holding company for now?
