Build a creative intelligence system around questions you can test

The two primary objectives of creative intelligence

In modern digital ad strategy, ad delivery is increasingly understood not simply as broadcasting a few dominant assets broadly, but as matching distinct creative concepts to different audience segments. Under this operating model, where creative serves as a primary targeting lever, creative intelligence focuses on two main goals: lowering cost per acquisition (CPA) within existing audiences, or unlocking new audience segments at an acceptable CPA to scale total spend.

When an account already operates at scale, performance gains rarely come from tweaking minor production details of an existing winning ad. Sustained growth typically requires testing horizontal concepts—divergent narrative angles, formats, and value propositions designed to resonate with buyer segments that existing creative ignores.

Why universal creative rules fail across accounts

Our cross-client discussion argues against assuming that a rule such as “shorter videos always win” transfers unchanged between accounts. The clips do not provide a comparative dataset; their practical point is to check the context before copying an apparent lesson.

Sweeping generalizations also fail to account for product and performance context. Stating that 'short videos don't work for prosumer brands' is not creative intelligence; it provides no understanding of why an ad succeeded or failed, nor does it account for the product's value proposition or campaign constraints. Without understanding the causal mechanism behind observed results, teams risk abandoning effective creative approaches based on flawed assumptions.

Defining brand ontologies beyond rigid naming conventions

Every brand operates within a unique creative ontology: the complete space of relevant creative decisions and variables available to that business. Variables that matter for one brand may be irrelevant for another. For instance, demographic variables like male versus female may be far less critical to track for an app like Flo Health, whereas for a brand like BlueChew, both male and female creators are deployed—even though female creators in that context may still speak to male buyers rather than targeting female audiences.

Historically, media teams attempted to track these dimensions through complex naming conventions, occasionally spanning a dozen or more taxonomy fields parsed via database queries. However, an analytical language is only as expressive as its predefined taxonomy. If a visual cue, pacing element, or framing angle outside the naming convention drives performance, manual tagging fails to capture it.

Our proposed approach combines structured metadata with multimodal analysis of the assets. Models can suggest patterns in pacing, visuals and message that a naming convention may omit. Those suggestions need to be checked against the actual creative and results; the model does not establish why an ad worked.

Investigating mechanisms instead of superficial labels

One conversation references Ray Dalio’s emphasis on examining how you know a conclusion is right. Applied to creative work, that means going beyond labels such as “authentic” to state a specific explanation and identify what evidence could test it.

The ElevenLabs example proposes that production craft may matter to an audience of editors and creators evaluating a creative tool. That gives a strategist a concrete hypothesis to investigate. It does not, on its own, establish the cause of the observed performance.

Another clip illustrates a possible mismatch in an AI video offer: “free to try” attracts people seeking free generation, but they encounter a paid gate. The example is not presented with account-level evidence. Its lesson is to connect the ad’s promise with downstream behavior and examine whether the acquired users want the paid experience.