volume is not the same as coverage
An ad is a point in a large design space: whom it speaks to, what change it promises, how it proves that promise, which mechanism it names, what format carries it, and how it looks and sounds. Generating twenty crop variants explores almost none of that space.
The creative-production research asks how an agent can produce genuinely distinct hypotheses, package each for the real channel, and measure what the audience actually did. The Resolve Studio report explores production mechanics; the creative-signal work analyzes how apparent authority gets manufactured in a market. Together they suggest a research route from creative taxonomy to production system to observed response.
The public version should demonstrate the state space with rights-cleared material and label which results have been measured. The private commercial corpus remains evidence for designing it, not a ready-made public gallery.
count hypotheses, not files
A campaign can contain fifty assets and one idea. The editor may change the aspect ratio, shorten the opening, swap a font and export a new file each time. That work may be necessary for distribution, but it has not tested fifty different reasons someone might buy. A useful creative system counts distinct hypotheses about the audience and the offer.
One way to map the space is to name the dimensions before production: audience, situation, promise, mechanism, proof, objection, opening, voice, format and visual treatment. Some combinations are nonsense. Some are risky because they imply a claim the product cannot support. The remaining combinations are candidates. The goal is to choose variations that teach something different, not to fill every cell of a spreadsheet.
Consider two ads for the same product. One speaks to someone afraid of wasting money and proves the claim with a cost comparison. Another speaks to someone embarrassed by a recurring problem and uses a demonstration. Those are different hypotheses. Two versions of the first ad with different music are mainly production variants. Both kinds of variation have a place, but they should not be reported as the same thing.
production has its own contracts
Once a concept is chosen, the real channel imposes constraints: a vertical opening must work in the first seconds, captions must survive the crop, audio stems need to be editable, and the revision has to retain a link to the concept it was meant to test. An agent can help generate storyboards, selects, captions and edit instructions. The project still needs source rights, a human review of claims, and an artifact another editor can revise.
The Resolve Studio research investigates how far that production chain can be made legible to software. The creative-signal research investigates another side of the market: how ads borrow the appearance of authority. Taken together, they suggest a creative system that can produce at volume while keeping provenance and claims visible. That is more demanding than one model generating one video.
learning closes at the audience
A creative taxonomy is only valuable if the results can update it. Track the concept, the exact edit, the channel, the audience, the spend and the response. Then ask what changed. A click may tell you the opening earned attention; a purchase may say more about the promise and offer; a refund may overturn the apparent win. Attribution is imperfect, but a recorded test is still better than remembering which ad “felt strong.”
There is a practical stopping rule. If a new asset does not test a new audience belief, a new proof route or a meaningful production constraint, say it is a variant. That keeps volume honest. It also makes the next creative brief sharper: which part of the space have we actually explored, and which bet would teach us the most now?
make the brief a map of bets
A useful creative brief can show what the team already believes, what it has actually tested, and where the next bet sits. It should name the buyer state and the objection, then connect the proposed concept to a proof route the product can support. The editor receives a production problem. The media buyer receives a test hypothesis. The founder can see which claim the company is putting into the world.
This avoids another common waste: winning one format and copying its surface into every channel. A long demonstration, a short testimonial and a static comparison may express the same hypothesis differently. Conversely, two visually similar ads may test different objections. The taxonomy should follow meaning and observed behavior, not just file type.
The final learning should feed back into the next brief. If an opening earned attention but the offer did not convert, the team should not declare the whole concept dead. If a high-converting ad brought refunds, the promise may be attracting the wrong expectation. A creative system becomes useful when it can make those distinctions before the next fifty files are produced.