Where marketing agencies are going in the age of AI

For most of the last two decades, the agency business ran on one quiet arrangement: hire a lot of junior people, bill their hours to a lot of clients, and keep the spread. Plenty of the work was real. But a large slice of it was commodity: resizing the same banner nine times, writing the fifteenth meta description, rebuilding last month’s report, filling a deck with slides nobody reads past page four.

That arrangement is coming apart. Not because clients got smarter, though they have. Because the commodity slice is exactly what AI is good at now, at a fraction of the cost and with none of the coordination tax.

I run a small agency, so weigh this how you like. I’m not writing from a threatened corner office. I’m writing because I think the change is healthy, and I’d rather say where it goes than pretend nothing is happening.

The floor is falling out of “hours”

The billable hour was always a strange unit. It rewards slowness. An agency that takes forty hours to do a thing earns more than one that takes four, which is a quietly insane way to price expertise.

AI agents are collapsing that. The repetitive execution — the ad variations, the first-draft copy, the campaign QA, the weekly “pull me the numbers” — is getting absorbed by tools that don’t sleep and don’t need a status call. If your agency’s value was the volume of hands, that value is draining away in real time.

What doesn’t drain away: knowing which thing to make, whether it is any good, and whether it actually worked. Judgement, taste, measurement. AI is a spectacular intern and a poor director.

Search is turning into answers

Here is the shift most owners are underpricing. People are stopping at the answer. Around 60% of Google searches now end without a single click, and inside AI chat the ratio is worse: you ask, you get told, you move on. The ten blue links you spent years optimising for are becoming something the model reads on your behalf.

So the game moves from “rank on the page” to “be the source the model quotes.” That is answer engine optimisation, and it is going from edge to table stakes fast. The upside is concrete. The sliver of traffic that does click through from an AI answer tends to convert three to four times better than a cold search visitor, because the model already did the qualifying. I wrote the long version on the AEO page. Short version: if your brand is not legible to the machines writing the answers, you are invisible to the fastest-growing way people decide what to buy.

Research gets a synthetic first draft

The other quiet change is how decisions get validated. The classic focus group — recruit eight people, feed them coffee, watch through glass, wait three weeks — is slow and pricey enough that most clients skip it and guess instead.

You can now simulate a good part of that. We built a tool for exactly this: the SimulatOR spins up 100 to 200 AI personas modelled on a real audience, has them argue over a concept, a name, a price, a landing page, and returns a report in 24 to 72 hours. It does not replace talking to real customers. It replaces the version where you launch on a hunch and find out in market. Much cheaper to be wrong on a Tuesday than across a quarter of ad spend.

The moat moved to data you own

When everyone has the same models, the same prompts, the same ad platforms, the edge stops being the tools. It is the data feeding them.

The agencies that look smart in two years will be the ones that got their clients’ measurement in order early: first-party data collected cleanly, server-side tracking so the signal survives the browser, conversions fed back to the platforms complete instead of guessed at. AI bidding is only as good as what you show it. Feed it half the picture and it optimises confidently toward the wrong thing. This is unglamorous plumbing, and it is becoming the line between spend that compounds and spend that leaks.

Clients want fewer, better people

Put it together and you get the thing I see across the board: clients want fewer people in the room, and they want them senior.

They have noticed they can get a competent first draft themselves. What they are paying for now is the person who has seen the pattern a hundred times and can say “don’t run that, run this, here is why.” Nobody is thrilled to fund a nine-person team where three do the thinking and six format slides. The six are the ones AI replaced.

Who wins

Not the shops with the biggest floors. The small senior teams that ship AI instead of talking about it in pitch decks, and stay honest about what is working and what is not.

That is the bet Raw Ideas is built on, and I will say it plainly because we already put it in public. We keep a handful of clients on purpose, so the work stays senior. We publish our prices instead of hiding them behind “let’s chat.” We build our own tools rather than reselling someone’s dashboard. None of that is a growth hack. It is just what the next version of this business looks like, and I would rather be early to it than defend the old one.

None of this kills agencies. It kills the ones that sold hours instead of judgement. If you want the longer argument for how we work, it is on the founder page. And if you have a decision worth pressure-testing before you spend on it, the SimulatOR is the fastest place to start.