The wrong scoreboard
Job-cut announcements citing AI are now monthly news. Gartner’s recent data says most of those programmes are not delivering returns. This piece is about what actually ensures they do.
Most of the AI content conversation is about generation. How fast, how cheaply, how much. That rewards anyone with a prompt window and a stock library, which is partly why the market is full of platforms pitching “transformation” that, in reality, is just a prompt-to-image pipeline.
If your AI workflow cannot guarantee product accuracy, version control, and a portable source of truth, you have generation. If it can, you have a system. The gap between those two is where the repeatability dividend lives.
Reusable and repeatable are not the same thing
Reusable describes an asset: a 3D model or material that can be picked up again without being rebuilt. Repeatable describes a process: the ability to produce the same quality of output, on brand, at scale, week after week.
The stack looks like this. At the bottom: the digital twin, the geometry and materials. In the middle: the pipeline, the AI and governance that turns the source of truth into outputs predictably. At the top: the shot.
Get the bottom right and the middle can run. Get the middle right and the top drops to almost nothing in marginal cost. Get either wrong and the efficiency you thought AI was giving gets eaten up by the time spent checking, correcting, and reworking. Most brands are pouring money into the top while neglecting the two underneath.
“
"If your platform disappeared tomorrow, what would you still own?"Oliver Disney
Chief Growth Officer, Collective
What it takes to compound
Three conditions matter.
First, the product has to be right. Forrester predicts ungoverned generative AI will cost companies over $10 billion in 2026. Cosnova’s work with Collective is the clearest proof of the alternative: product managers usually need seven to ten feedback loops before an AI image resembles the product on the shelf. With a digital twin in the pipeline, they were hitting 96% accuracy on the first pass.
Second, the asset has to be portable. Most AI platforms offer reusability within their own walls. That is rented, not owned. Open standards like OpenUSD, OpenPBR, and MaterialX give 3D content the portability that PDF gave documents. A pack modelled in 2026 should still render correctly in 2030 because it was built in an open language.
Third, the process has to hold. A reusable asset on its own just sits on your DAM. The commercial return comes from the pipeline around it: governance, version control, brand guardrails, baked in rather than bolted on.
The commercial trap, and the 2030 test
Most brand content is still bought on an asset-by-asset basis. A shot, a render, a SKU. That works when everything is bespoke. It punishes repeatability. The shift is from buying outputs to buying systems.
Four questions worth asking about your current pipeline:
- Are your AI outputs accurate enough on the first pass that the process can run without supervision?
- If you changed agencies tomorrow, could the next team pick up your assets and keep going?
- Could a team in a different market run a campaign from your master assets without rediscovering how the pipeline works?
- Does your contract reward repeatable output, or penalise it?
The brands that compound over the next five years will not be the ones generating the most. They will be the ones who built content infrastructure. The difference will be obvious by 2030.
Read the full piece on LinkedIn: The repeatability dividend.
Related articles block for this post:
