Know where they look before you launch.
Every creative decision is a bet, and most are settled after the spend. Predicted attention moves the question from a post mortem to a gate you pass before the budget is committed.
Design that steers attention instead of guessing at it.
The standard sequence is still: design it, argue about it in a review, ship it, then read the analytics. By the time scroll maps and click data arrive the media budget is committed, and the learning applies to the next campaign rather than this one.
Traditional eye tracking was the honest answer to that problem and it is still the reference standard. It is also slow and expensive enough that almost nobody runs it on a routine landing page or a set of ad variants. Vendors put conventional lab studies at roughly two to four weeks and 1,000 to 5,000 US dollars per analysis, with 20 to 50 recruited participants.
Predictive attention modelling changes the economics of asking. Saliency models trained on millions of real human fixations return an attention estimate for a static creative in seconds. We build that into your review process as a pre-launch gate, and we are explicit about what these models genuinely know and what they cannot.
Predicted attention maps
Where the eye lands in the first seconds, on a landing page, an ad, an email, an app screen or a pack.
Scanpath and hierarchy
The order attention is likely to travel, and whether it reaches your call to action at all before it leaves.
Area of interest scoring
The share of total attention each element earns: logo, headline, offer, price, face, call to action.
Variant ranking before spend
A whole set of creatives ranked by predicted focus and clarity, so the review ends with a decision instead of an opinion.
The attention engine behind the work
We built our own attention tool because we needed to run this on every client launch, not on the occasional flagship campaign.
It returns four things for any creative: a predicted attention heatmap, a focus and clarity score from 0 to 100, an area of interest breakdown that auto detects logos, headlines, calls to action and faces and reports the share of attention each earns, and a predicted scanpath showing the order the eye is likely to travel. It reads an uploaded image or a live URL, and it handles ads, landing pages, email, packaging, app screens, social creative, banners, thumbnails, print and video.
It is tuned on live campaign work rather than lab material, and it runs inside a review process: four gates before launch, variant ranking, and a record of what we predicted so it can be checked against what actually happened.
Predictive UX
How to predict where a user will look before a page is ever published. A working guide to AI attention maps, eye tracking approaches and scroll depth data, built so design can steer attention instead of guessing at it. It covers what saliency models genuinely know and where they fail, the three signals worth reading, and a four gate pre-launch playbook with variant ranking and channel plays.
Frequently asked
Is this real eye tracking?
No, and we are careful about the difference. Real eye tracking measures where actual people looked. This predicts where people are likely to look, using saliency models trained on large sets of real human fixations. It is a screening and ranking instrument, and it is fast enough to use on everything.
How accurate is a predicted heatmap?
Accurate enough to rank variants reliably and to catch a hero or a call to action that is being ignored. Less reliable on anything that depends on intent, brand familiarity or reading comprehension. We use it where it is strong and tell you when a question needs a real test instead.
What can you analyse?
Static creative of almost any kind: paid social and display ads, landing pages, email, app and web screens, packaging, print, thumbnails. Video is handled frame by frame so you can watch attention move over time.
Does this replace user testing?
It replaces the argument about whether anyone will notice the headline. It does not replace usability testing, interviews or live experiments, and it is not meant to. It removes the cheap questions so your testing budget goes to the expensive ones.
Can you run it on our existing pages and ads?
Yes, and that is usually where we start. A pass over what is already live tends to surface the quickest wins, and it gives us a baseline to judge new work against.
Three guides. No charge, no filler.
Each one is the working method behind a service, written the way we brief a client rather than the way an agency writes a brochure. Sources are cited so you can check the claims.
The Search Evolution
Why a top ten ranking no longer guarantees a citation, what the research says earns one, and the audit and reporting steps to act on it.
Download now
The Conversion Lab
What share of A/B tests really win, how much traffic a test needs, and why last click attribution and platform reporting disagree.
Download now
Predictive UX
What saliency models genuinely know and where they fail, the three signals worth reading, and a four gate pre-launch playbook.
Download nowCreative and UX work from the portfolio
Attention prediction is a review tool. The output still has to be designed, and this is the work it sits inside.
See all casesSend us the creative
Give us a landing page or a set of ad variants. We will come back with the predicted attention map and what we would change before it goes live.
