A few months ago, I was one of several reviewers on a piece of creative work that had been produced fast with AI assistance.
I added my comments and sent it back. Later I scrolled up the email chain to check a detail and found an interesting pattern. Other people had already left near-identical notes before it reached me, across both the agency and the client side.
Same file, same sort of comments, same polite send-back, same twenty minutes gone. We’d spent longer discussing the asset than the time it had taken to make it. I was another layer of it.
A few years of my working life have gone into helping organisations put AI into their production workflow alongside humans. The speed it delivers is real. That’s what makes the rest of this awkward.
In February this year, the National Bureau of Economic Research surveyed almost 6,000 senior executives across the US, UK, Germany and Australia. The utility layer is established; seven in ten companies use AI, but about nine in ten said it had made no difference to productivity or employment in three years.
B&T ran the Australian version 89-in April under a headline that didn’t bother being polite: AI’s impact on businesses is “essentially zero”.
Sitting inside that study is a smaller number that explains more than the headline does.
The executives who use AI spend an average of 1.5 hours a week with it; you probably spend more time in WIPs each week than you spend with AI.
So before we blame the technology, it’s worth asking what we actually changed around it.
Think about airports. Self-service check-in worked exactly as promised, and checking in genuinely got faster. You still don’t spend less time at the airport, because the queue moved to bag drop and security. The airline solved its own bottleneck and handed you a fresh one.
That’s roughly what our industry has done with AI. We bought the check-in kiosk and kept the queue.
The Handover
Many measure AI adoption by licence numbers, weekly logins, training completed and tools approved. Useful admin. But none of it tells you whether the work changed shape.
Here’s what I think is actually happening. There’s a fault in the moment work passes from one person to the next. That’s where the time gets lost.
Every time AI-accelerated work crosses a human boundary that was built for slower production, some of the time you saved gets placed straight back, and it gets charged in the currency nobody is counting: calendar days, waiting on someone else’s inbox, versions that exist only because four people had a preference on a Tuesday.
And this thing shows up fastest in creative.
A team that used to produce four routes over several days can now explore 40 in an afternoon. That acceleration is real, and it gets a round of applause in the status meeting.
Then somebody has to review 40 routes.
Then the account lead reviews them.
Then brand.
Then legal.
Then the client’s brand team.
Then someone joins halfway through with a surprisingly strong opinion about the blue that needs to be a bit bluer.
Production and attention didn’t move at the same speed. While production became abundant, attention stayed exactly as scarce as it always was.
What handovers do to the idea
I wrote here in June about the research Deakin and Swinburne ran across 371 films entered into DISRUPT, Australia’s first AI film festival.
Full disclosure again: I helped create it. The finding was that intention, taste and judgement separated the best work from the merely impressive.
Those things live in people.
Your operating model decides how much room those people get.
Every handover is a small renegotiation of the original intention.
Our approval chains, briefing templates, review meetings and agency-client workflows were built for a world where making things was slow and expensive. That was sensible at the time. Then making things got fast, and the workflow stayed exactly where it was.
What to do on Monday
None of this needs a restructure. It needs an afternoon.
- Take one live job and count every point where somebody can stop the work or send it backwards. Most of us guess low by half.
- Name the person who decides at each of those points. One name. Where you can’t name one, you’ve found where an inefficiency tax is likely being collected.
- Ask which of those steps improves the work, and which one exists because that’s how we’ve always done it.
- Remove a step rather than speeding it up. A faster meeting is still a meeting.
That last one is where most efficiency projects quietly fail. The same ten-slide brief becomes an AI-generated ten-slide brief. The same status meeting gets an AI summary. The same approval chain runs through a nicer tool. The same internal feedback arrives in a tidier format. All helpful. The shape of the process stays exactly where it was.
And ask the question underneath all of it: do we need 40 routes because the brief needs 40, or because the tool can make 40?
Useful friction makes ideas better. Inherited friction just makes them later.
Here’s the part nobody wants on the slide.
Approval is how a lot of us prove we were in the room: the comment that shows you read it, the small change that shows you added something, the invite that confirms you’re senior enough to be there, the reply-all that puts your name in the chain. Ask a business to remove approval stages and you’re asking well-paid adults to volunteer for less visible influence.
It’s a management decision, and it’s far less fun than a demo. I know, because I’ve been the extra stage.
The executives in that study expect AI to deliver an average 1.4 per cent productivity improvement over the next three years. Maybe they’re right. Forecasts are free, and nobody has to remove a single approval to make one.
The 1.4 per cent is a prediction. The approval chain is a decision.
Lucio Ribeiro is the chief AI and innovation officer at Omnicom’s TBWA\Australia. He previously led AI and innovation at Optus, Seven West Media and Nine, and lectures in AI at RMIT University.
