
The question I get asked most, by clients and other agency owners, is how much of the work AI can do now.
It almost always comes down to speed and cost. Can we produce the same volume of ads with fewer people? Can we turn a campaign around in two days instead of five? Should the retainer come down, now the technology is doing half the work?
Catalyst produces high-volume video advertising for around a hundred brands using a network of more than 10,000 creators. AI has been part of how we work since we started three years ago. Every new tool gets a three-month trial, whatever it costs. If it proves its value, we build our own version of it in-house.
So yes, AI makes us faster. The more interesting question is what happens once the technology has done its job. Output multiplies. The number of people making judgment calls doesn’t. The pressure hasn’t gone anywhere. It has moved off production and onto decision-making, which is the part of most businesses that hasn’t been upgraded. A team that ships twice as fast but makes the same, or worse, decisions hasn’t gained anything. It’s just found a faster way to be wrong.
AI is very good at spotting patterns in data it has seen before, and much worse at navigating a new situation. In advertising, which is rarely predictable, that’s a recurring problem. We’ve had campaigns where an AI model immediately recommended switching off the ad because click-through had dipped in its first few days. Someone who’s spent years running hundreds of Meta campaigns knows that’s usually the platform testing its way into a new audience. People who have enough advertising scar tissue (that’s how I think of experience) should be the ones reviewing the AI.
That’s harder than it sounds at Catalyst, where most of the team are under twenty-five, and many are in their first job. They’ve never worked without these tools. What they haven’t had yet is years of watching campaigns recover from a bad first week, so we have to put that experience within arm’s reach.
Our strategists, designers and editors work together in small ‘pods’, so a bad instinct usually gets caught by a colleague. Putting those people round the same table has solved more problems than any process document we’ve ever written.
The trap that most agencies and brands fall into is letting speed become the metric instead of the by-product. If AI produces an answer in seconds, there’s an expectation that all decisions can be made in seconds. We deliberately make sure we’re not delivering work at a superhuman pace, so if anything goes wrong, we don’t disappoint clients.
Volume is the real threat, not quality. When everyone can generate ten times more ideas, edits, reports and creative variations, everyone’s job becomes about sifting through AI output. An editor can’t assess fifty AI-generated cuts with the same care they’d give five. Either headcount or standards have to flex; otherwise, attention gets thinner and thinner until it snaps.
As output has grown, we’ve tried to keep ownership clear and manageable.
Early on at Catalyst, we were completely siloed. Design, animation, copywriting, strategy and performance were each responsible for their own part of the project. When everyone owns a stage of the process, nobody feels responsible for the final outcome.
We fixed it by giving one person responsibility for the whole thing, from strategy through to signing off the final creative.
Automation makes it very easy for responsibility to blur, particularly when something goes wrong. Before any work goes live, someone owns it. If an AI-assisted campaign underperforms, there shouldn’t be any debate about who was responsible for making the final call. We’ve been explicit with the team that AI is a tool, and a tool can’t be held responsible.
Speed metrics are seductive because they’re easy to measure and they always go up. Turnaround times and volume produced look great on a dashboard, but tell you very little about whether the work is any better. We weight our internal reporting towards campaign performance, client retention and win rates instead. They’re slower-moving, and they’re far less exciting to present in a meeting, but they’re the ones that show whether the business is improving or simply speeding up.
None of this is an argument for using less AI. We’ll keep trialling every new tool that comes along. But when a client asks me how much of the work the machine can do now, the honest answer comes with a second half: it hands your team more decisions, made faster, on more information, with less time to think. Most businesses haven’t planned for that yet, and I suspect it’s what will decide which of them actually see a return.



