The post-production stack has not been replaced by AI. The tools that have carried campaigns from first edit to final delivery for over a decade, such as Autodesk Flame, Nuke, Adobe Premiere Pro, DaVinci Resolve and Pro Tools, are all still alive and well. What has changed is what happens inside those tools. AI has hollowed out the labour-intensive work within each of them. That matters because it tells you where the savings are real, where the hype outruns the evidence, and what to expect from your next production budget.
AI has compressed the manual layers in post-production. The biggest shift is that AI has taken on what McKinsey calls the microtasks, the repetitive steps that used to absorb hundreds of staff hours. The clearest measurements come from film and high-end VFX, because that is where those manual hours pile up highest, but the same ones running advertising, so the picture maps straight across. Rotoscoping is the cleanest example: isolating a subject frame by frame so that effects, backgrounds or new elements can be layered behind it. The work is so labour-intensive that studios have long sent the bulk of it offshore, with more than 90% of Hollywood's rotoscoping done in India. A Roland Berger study of AI in visual effects found that automated rotoscoping cuts the work by up to 65%. What took a skilled artist a full day now takes a click and a tidy-up.
The same compression is running through the rest of the production workflow. Clean-up and object removal have gone from frame-by-frame work to near-automated passes. Backgrounds can be generated and extended rather than rebuilt from scratch. Grading software suggests scene-matched looks, and dialogue repair and noise reduction now sit as features inside Pro Tools. These are the microtasks, and they are exactly where the hours come back.
In advertising itself, adoption is wide but shallow. ISBA found the share of UK advertisers with at least one live generative AI use case quadrupled to 41% in just over a year, and 62% say efficiency is the primary focus of their AI strategy. Depth is another matter: the WFA's research on in-house agencies found only 17% have fully integrated AI into their operations, while 61% are still in early testing. Where AI is being used, it is concentrated in those manual, time-consuming and much less glamorous layers.
Notice what is missing. None of this is new creative capability. Nothing here puts anything on screen that a professional team could not achieve before. It is the same work, minus the repetitive, boring tasks.
What AI cannot do is judgement. Creative direction, hero-shot VFX, complex character work, editorial pacing, and the trained eye that catches when a shot is not quite sitting right. None of that has moved.
Two numbers show how wide the gap is. Studio executives expect efficiency gains of as much as 80 to 90% in VFX and 3D asset creation. Yet the productivity increase leaders report actually achieving so far is just 5 to 10% in specific use cases. The difference between those figures is not a measurement error. It is the judgement layer, and it has barely moved.
A recent REI ad shows what happens when that layer is skipped. A real product photo of a Van Rysel road bike, from a professional shoot, was passed through an automated ad-personalisation tool. The tool redrew the image and gave the bike a second set of handlebars growing out of the saddle, along with extra chains and a distorted rider. The software flagged none of it. The ad ran on Instagram for about a week before REI pulled it in June 2026, by which point the comments had already torn into it. Nobody had looked at the output before it shipped.
In global campaigns the same failure is quieter and harder to catch. An automated clean-up pass comes back flawless for nine markets and subtly wrong in the tenth. The tool will not flag it, but someone who knows that market will. The manual tasks of production can shrink by 90% while the work of reviewing the result barely changes, because the review still has to happen, market by market. If anything, it grows, because the same speed that produces the savings also produces far more assets, and every one of them still needs a human to decide whether it is right.
This is the layer where quality is decided, and it is the one thing not to hand over. Faster output means more to review, not less, so keep experienced eyes on the work and budget for their time. Automation without someone reviewing the result is a risk waiting to go live.
The compression sits in the repetitive, manual layers: the clean-up, the masking, the rotoscoping, the noise reduction. That is where the hours genuinely come back, so that is where to push. A useful question for any partner is which of those steps they have already automated.
Most of the flashy standalone AI video tools are built for content creators: short clips, compressed files, a single 9:16 deliverable. Professional global advertising runs on uncompressed footage at 4K or above, layered project files, strict colour pipelines and versions for dozens of markets. These tools are not yet capable of handling the needs of an international campaign.
Freedman's latest roundtable brought senior marketers together on exactly this question: where to build AI capability, and where to buy it in. We've written up what came out of the room: [BUILD OR BUY LINK HERE]