How it works

The analysis flow

In short

Foresight checks your file and choices, prepares the creative (and frames it in a platform screen if you picked one), runs the attention and emotion models in parallel, turns their predictions into 0–100 scores, adds the benchmark comparison, and then writes AI insights. Results appear only when both models have finished.

This page follows one creative from the moment you click Start analysis to the moment its results and insights appear.

The steps at a glance

1. Checks. Foresight checks the file (type, size, video length), your plan's remaining analyses, and that the benchmark you chose exists for that media type. Problems here are shown immediately and the file is not analysed.
2. Preparation. The creative is converted to a standard working format so that every creative is processed the same way. If you picked a platform, a framed version is also made: your creative placed inside that platform's screen.
3. Content description. A short description of the creative's content is produced; the emotion model also uses it as context.
4. Two predictions in parallel. The attention model predicts where people are likely to look and how visually complex the creative is. The emotion model predicts which emotions it is likely to evoke.
5. Scores. The raw predictions are turned into 0–100 KPIs by positioning them against a large set of reference creatives, and the E-Score is computed from Emotional Engagement and Attention.
6. Benchmark comparison. Next to each score, Foresight shows the average of the benchmark group you selected, and the difference. With no group selected, no comparison value is shown.
7. Results. When both models have finished, the results page opens up: scores, overlays and, for videos, the timeline.
8. Insights. Foresight Copilot reads the finished scores and writes the summary and recommendations. They appear automatically a little after the scores.

From upload to insights: both predictions run in parallel, then become scores positioned against real creatives.

What feeds each step

Your wizard choices each reach a different part of the flow:

Your choice	Where it acts	What it changes
The file	Every step	Everything. The file type decides whether it is analysed as an image or a video.
Platform	Preparation (framing) and the benchmark display	What the attention model sees, so attention scores can move; a framed image loses its creative type and sector comparison.
Creative type and sector (images)	Benchmark comparison	Only the comparison values shown next to the scores.
Objective	Insights	Only how the summary and recommendations are framed.

How the platform choice flows through

Choosing a platform does not change the models or the way KPIs are calculated. What changes is the picture the attention model analyses.

• Your creative is fitted inside the platform's content area without cropping. If its shape differs from that area: for example a vertical video in a square feed slot, the remaining space is filled with the platform's background.
• The interface is shown in a simplified, softened form.
• Attention is predicted on the whole framed screen, and the overlays you see are drawn on the framed view.
• The emotion model always reads your original creative, never the framed one. A small, fixed contribution from the platform's interface is then added to its results.

That is why, when you test the same creative with TikTok selected, emotion scores shift only slightly while attention scores (Clarity most of all) can move noticeably. A framed image has no creative type or sector comparison group, and videos show no comparison value with or without a platform. The colour bands are the same for every platform. See What changes when you pick a platform and TikTok.

How long it takes

Most analyses take a few minutes. Videos take longer than images, and longer videos take longer than short ones. You don't need to keep the page open: the dock at the bottom of the screen tracks progress and you get a notification when results are ready.

When something goes wrong

Once started, an analysis can still fail, for example if the file is corrupted or uses an unusual encoding, if the creative can't be placed in the chosen platform frame, or if one of the prediction steps can't complete. Some steps are retried automatically. A failed analysis is marked Analysis failed: re-export the creative as a standard MP4 (video) or PNG/JPG (image) and upload it again, rather than retrying the identical file.

If AI insights can't be generated, your scores are still available, but the insights may be missing. For videos, Foresight splits the video into scenes automatically; occasionally the scene breakdown may be unavailable. An analysis that stays in progress for an unusually long time should be reported to support.

How to read it

• There are no partial results: scores appear only when both the attention and the emotion predictions are complete.
• Insights always arrive after the scores, because they are written from the finished scores.
• A different benchmark would change the comparison values, not your scores: the scores are computed before the benchmark comparison is added.

Common pitfalls

• Expecting the objective to change a score. It only shapes the insights.
• Comparing a framed and an unframed analysis of the same creative as if they were the same test. The attention model saw two different pictures.
• Renaming a file's extension to change how it is analysed. Foresight checks the real file format, and an animated GIF is always analysed as a video.