How it works

The attention model

In short

The attention model learned from eye-tracking research where many people's gaze was recorded while they freely viewed images. It predicts a map of where a typical viewer is likely to look during free viewing, and a companion model estimates visual complexity. Focus, Spread, Clarity, Mental Demand and Attention are computed from these predictions.

The attention model predicts where a typical viewer is likely to look when your creative appears, and a companion model estimates how visually complex it feels. Together they feed every attention overlay and the attention KPIs.

What it measures

The model produces an attention map: for every point of your creative, how likely people are to look there during free viewing (looking at it without a specific task). The heatmap and fogmap you see on the results page are drawn from this map.

Each map is scaled to the creative's own strongest point, so the colours show where attention goes within that creative.

What it learned from

The model learned from eye-tracking research: large collections of images for which researchers recorded where many real people looked while viewing them freely. From these recordings it learned which visual features tend to draw the eye (faces, text, contrast, colour, composition and position in the frame).

The model used in Foresight is optimised to run fast enough to analyse every creative, including long videos.

The complexity model

Mental Demand comes from a companion model, trained on people's ratings of visual complexity, that estimates how complex a typical viewer would judge your creative: the number and density of elements, detail and clutter.

It looks only at what is visible in the frame. It does not read the story, the script or the meaning of the text, so a sophisticated narrative told through clean, simple shots gets a low Mental Demand. See Mental Demand.

What feeds it

Only pixels. The attention and complexity models don't receive text, sound, your sector, your objective or any information about your audience.

If you choose a platform, the input is your creative placed inside that platform's screen, with the interface softened. The model then predicts attention on the whole screen, so some attention can go to the interface or to empty space around your creative. See The analysis flow.

How the KPIs are built

All attention KPIs come from the predicted map and the complexity estimate. The logic, in short:

• Focus: is predicted attention concentrated on a small part of the frame, or spread over large areas?
• Spread: are the main points that attract attention close together or scattered?
• Clarity: built from Focus and Spread; high when attention gathers on a few, well-grouped areas.
• Mental Demand: how visually complex the creative looks, from the complexity model. The middle is best: 25–75 is the recommended zone.
• Attention: built from Clarity and whether the visual complexity sits in a comfortable range. It is highest when the eye is guided to a clear focal area and the creative is neither too bare nor too busy.

Attention is predicted on what the viewer would see, including the platform screen if you chose one.

Each raw value is positioned against reference creatives of the same media type, so 50 means typical. See the KPIs overview.

Scale & direction

The map itself has no numeric scale: red areas are more likely to be looked at, then orange and yellow; uncoloured areas less. The KPIs derived from it run from 0 to 100, higher is better, except Mental Demand, where the middle band is best.

Image vs video

• Image: the whole picture is analysed once. Images also get a gaze plot and an attention build-up animation, both simulations built on the map.
• Video: Foresight analyses sampled frames across the whole video and fills in the moments in between. You get heatmap and fogmap videos and per-moment curves. Each frame is analysed on what it shows; movement between frames, sound and the story so far are not taken into account. See How videos are analysed.

How to read it

• The heatmap shows where attention is likely to concentrate during free viewing; Focus, Spread and Clarity put numbers on how organised that attention is.
• Clarity reflects how concentrated predicted attention is on the creative as shown: a measure of visual organisation and hierarchy, not of whether the message is understood.
• Mental Demand is visual complexity only: narrative or conceptual complexity has no influence on it.

Common pitfalls

• Comparing two heatmaps by colour intensity. Each map is scaled to its own creative, so a "red" area in one is not "more attention" than a red area in another.
• Expecting text to be read. The model sees text as visual shapes: it predicts whether text draws the eye, not whether it is legible or understood.
• Reading the map as a specific audience. It predicts a typical viewer's free viewing; it doesn't know brand familiarity, scrolling intent or what else is in the feed.
• Comparing framed and unframed results. With a platform selected, the model saw a different picture: your creative plus the interface.