Ad Anatomy.

This is the most objective layer and the one with the most immediate upside. CreativeX analysed roughly 890,000 digital ads, representing about a trillion impressions and 1.37 billion dollars in media spend, and mapped where assets and placements diverge. Their findings: 35 percent of digital media spend went behind video with no visible branding in the first three seconds, 50 percent behind ads with no call to action, 12 percent behind assets not sized for their placement, 32 percent behind video outside the platform's recommended length, and 7 percent behind video whose core message lived only in the audio.

The upside is measurable. Every ten percentage point improvement in CreativeX's Creative Quality Score corresponded to a 6.3 percent reduction in cost per completed view, measured across 1.8 million video ads and 1.6 trillion impressions. Analytic Partners found that creative built natively for a channel is 2.2 to 3.6 times more effective than broadcast creative repurposed for digital.

Every check in this layer is deterministic. Aspect ratio, safe zone, length, resolution, presence of branding, presence of a call to action, presence of captions. An engine that assembles from specs should simply never get these wrong.

Attention is not a sliding scale. It has a threshold. Amplified Intelligence measures active attention, meaning eyes genuinely on the ad rather than the ad merely being viewable, and places the attention memory threshold at an average of 2.5 seconds. Roughly 85 percent of digital impressions fall below it. Every additional second of active attention above the threshold adds around three days of memory.

The most useful recent refinement comes from VCCP Media and Amplified, based on more than 20,000 views of 72 digital video ads, each tested in its original form and in a version with brand colours, characters, sonic branding and visual assets stripped out. Their finding: 1.5 seconds is enough to encode memory, provided strong distinctive brand assets are present.

Dentsu's 2026 study The Brand Reset, built with Kantar and Lumen Research across 40,000 respondents and ten video platforms, adds the ceiling. Returns on attention begin to diminish beyond roughly 20 seconds of active viewing. Duration is not the goal. Quality of attention is.

Google's ABCD framework, validated across more than 17,000 campaigns with Ipsos, Nielsen and Kantar, sets out how to design those opening seconds: get to the heart of the story fast, use tight framing and engaging pacing, support the message with audio and on-screen text without letting elements compete, and keep visuals bright and high in contrast.

This is where the most value sits unclaimed. Jenni Romaniuk of the Ehrenberg-Bass Institute defines distinctive brand assets as the sensory triggers that bring a brand to mind without the brand name: colours, shapes, characters, sounds, words, silhouettes. They are assessed on two axes. Fame is how many people link the asset to your brand. Uniqueness is how many of those people name only your brand.

CreativeX published the largest asset-level study on this in April 2026, covering more than 2 billion dollars in ad spend over four years, across 300,000 creatives within 1.4 million ads, from 176 brands in 105 countries. Ads carrying four or more distinctive brand assets showed 28 percent higher ad recall and a 45 percent reduction in cost per completed view. Currently 61 percent of ads use only one or two brand cues.

Ebiquity's analysis of 1,300 campaigns from its ROI benchmark database found 62 percent stronger short-term profit ROI for campaigns built on recognisable brand ideas and assets. Ebiquity also identifies the compounding mechanism: brands that build and maintain distinctive assets create memory structures that last longer and decay more slowly, so they need to spend less later to restore awareness.

One caution worth stating plainly. Research published in the Journal of Brand Management, testing more than 400 brand elements, found that marketers' judgements about their own brand assets are rarely accurate. Familiarity with your own category tends to make you overestimate fame and underestimate uniqueness. This is a layer to measure, not to assume.

Kantar, working with Affectiva's facial coding, found that digital ads generating strong emotional responses are four times more impactful on brand equity. System1's central finding is the mirror image: neutrality, the absence of emotion, produces neither short-term nor long-term commercial effect. Their modelling puts the additional media cost of emotionally flat advertising at around 189 billion dollars globally.

In the Netherlands, Energize and professor Peeter Verlegh of Vrije Universiteit Amsterdam ran a large study into which factors predict the voluntary attention people give to brand communication. It produced six attention triggers: emotion, help, novelty, inclusion, status and reward. Of the six, emotion is the strongest predictor.

Google's guidance in this layer is short and holds up well: humanise the story by putting people in it, focus on one message rather than three, and use emotional levers such as humour, surprise and intrigue.

Half of the digital budget in the CreativeX analysis sat behind ads with no call to action at all. The underlying assumption is that calls to action belong to performance advertising, which leaves brand communication without a next step even when a soft one would do: follow, sign up, learn more, watch the longer cut.

Google's guidance: be intentional and say what you want people to do, reinforce the on-screen call to action with voice-over, and adjust by objective. Plant urgency at consideration. Present the ask once the context is set at the action stage.

One call to action per asset. Multiple next steps dilute the response.

This layer did not meaningfully exist two years ago. Meta introduced Andromeda, a new ad retrieval engine, via its engineering blog in December 2024, with global rollout completed by around October 2025. Andromeda reads the content of the creative itself, using computer vision and audio analysis, and uses that to predict which user the ad suits.

The practical consequence is significant. Assets that look alike, for example the same template with a different text layer, are collapsed into the same internal entity. Fifty small variations on one concept perform roughly like one asset. Ten genuinely distinct concepts perform like ten.

This is the difference between resizing and assembly. A campaign set is not one idea in many formats. It is several genuinely different angles on the same proposition, each correctly executed for its placement. The six attention triggers from the Vrije Universiteit Amsterdam research work well as an axis for that: six angles on one proposition, each with its own hook.