Creative Precision

Predictive ad attention

Every creative you run.Every medium you run it in.

Creative Precision predicts where a viewer’s eye goes — on your Meta ads, your TV spots, your hoardings, your press — breaks each frame into its parts, and tells you how much attention every part holds. Then it checks that prediction against what the campaign actually delivered.

No person can do this, and it is not close. A strategist reviews a handful of creatives and forms an opinion. This reads every frame of every creative you have ever run and scores them all by the same standard — the ten-thousandth exactly as carefully as the first.

Creative drives 0% of incremental sales.

For digital alone it is 0% — a larger share than targeting, reach and recency combined. It is also the least measured part of the funnel.

NCSolutions (Nielsen Catalina), 2017; reconfirmed in a 2023 meta-study of roughly 450 campaigns across TV and digital.

Every frame

of a video, scored individually rather than averaged into one number

6 element types

located per frame: text and what it says, product, logo, person, face, call to action

Every creative

in the account — connected, uploaded, or imported from a report

Where the eye actually goes

Hover a row to find it in the frame. Every element gets a share of predicted attention, and the shares account for the whole creative.

  • Person38%
  • Branding31%
  • Headline18%
  • Product9%
  • The offer4%

Read the bottom two rows. Branding takes 31% and the offer takes 4%. That is a finding you can act on this afternoon, and no single score would ever have told you.

Illustrative layout and plausible figures, not a client’s creative and not a measured heatmap. Your own numbers come from your own ads.

Ad Focus

How much of the predicted attention your brand and message hold, rather than being spent on background.

Dwell Time

For video, how long attention is predicted to stay before it moves on — measured frame by frame, not averaged over the whole clip.

Attention Index

The two combined into one number you can rank a set of creatives by.

Attention over time

A video is not an image. We score every frame, so the question becomes when attention leaves — not whether the ad was “good”.

0.0s1.0s2.0s3.0s4.0s5.0s
0.0sAd Focus 86%

Face enters — attention locks on immediately

The branding lands last, on the fewest eyes. An average over the whole clip would have called this a 53% ad and told you nothing. The fix is a scheduling decision, not a creative one.

Illustrative curve over a synthetic storyboard. Your own curve comes from your own footage.

Which of these two should you run?

The question nobody arrives with is “what is in my ad”. This is the one they do.

Variant A — product led

61

Attention Index

  • Product34%
  • Offer26%
  • Branding22%
  • Copy18%

Lower overall score, but the offer holds a quarter of the attention and the branding is seen.

Variant B — lifestyle led

74

Attention Index

  • Person63%
  • Copy21%
  • Branding11%
  • Offer5%

Higher score, and almost all of it is the model's face. The offer is effectively invisible.

B scores higher and would probably lose. A single number ranks them and stops. Which element is carrying the attention is the thing that decides whether anyone sees the offer — and it is the reason we report shares rather than a verdict.

Illustrative figures. Your own comparison runs on your own creatives.

Every medium you buy, in one place

Upload the creative and the plan — a spreadsheet straight from your agency — and the same analysis runs on all of it. A 48-sheet and an Instagram story go through the identical pipeline.

Digital

Meta, Google, TikTok, DV360, The Trade Desk — connected or imported.

TV

Spots analysed frame by frame, against GRPs, spots and dayparts from your plan.

OOH & DOOH

Hoardings and screens, against sites, opportunities-to-see and plays.

Print & PR

Press ads and earned coverage, against circulation and readership.

One caveat we will state rather than bury. A billboard has no click, and neither does a TV spot. For offline media we compare predicted attention against exposure — how many people could have seen it — not against what they did next. That answers which creative earned its reach, and it is a weaker check than the digital one. We keep the two apart and never average them together.

Others tag what an ad contains.We measure what gets looked at.

A list of attributes tells you what is in the frame. It does not tell you whether anyone saw it. Every element below comes back with the share of predicted attention it actually holds — which is the number that changes what you do next.

Branding

Your logo — found, read, and weighted by how much of the eye it actually takes.

The offer

The call to action, transcribed word for word, with the share of attention it holds.

The product

Present or absent, and whether it is what the eye goes to or what it skips.

Message

Every line of copy, transcribed, with how much of it is read before attention moves.

Text density

How much of the frame is words — the single most common reason an ad is ignored.

People

Faces and figures located, with only a bounding box kept. No embedding, ever.

How we compare

Creative agencyAI tagging toolCreative Precision
What it measuresA senior opinion, formed on a handful of adsTags what the ad CONTAINSWhere the eye actually GOES, per element, per frame
Media coveredWhatever that agency buysMeta and GoogleDigital, TV, OOH, DOOH, print, cinema, radio and PR
Checked against resultsAfter the fact, if at allClaimed, rarely shownAgainst your own delivered numbers, misses included
When it is wrongNobody mentions itNobody mentions itWe show you, by name

What makes it different

It sees every element, not an impression of the whole

A person looks at an ad and forms a judgement. This locates the text and reads it, finds the product, the logo, every face, the call to action — and reports how much predicted attention each one holds, on every frame of a video. Nobody does that by eye, on any number of creatives, and nobody has ever done it on thousands.

It does not get tired, bored, or talked round

The tenth creative of the day is scored by exactly the same standard as the first, and so is the ten-thousandth. No preference for the campaign someone fought for, no fatigue at 6pm, no deferring to whoever is most senior in the room. Consistency at that volume is not a human skill.

It checks itself against your actual results

We connect to your ad account and pull what each creative really delivered — impressions, clicks, and the outcome the campaign was optimised for. The prediction is shown beside the result, including when the two disagree. A tool that only predicts can never tell you it was wrong, which is why most of them never do.

It reports what it could not read

Every sync counts what it skipped and why, by name. Carousels, catalogue ads, formats with no downloadable asset, video a read-only permission cannot reach. Silence about the gaps is how a tool reports success on half an account and nobody notices for a year.

Read-only by construction, and single-region by design

We ask for read permission and nothing else — the integration cannot alter a campaign, a budget or an ad, because it was never granted the ability to. Your creatives and results are stored in one AWS region, the one your account is provisioned in, and are never replicated to another, never sold, never shared between clients, never used to train models. Where analysis runs on a GPU in a different region we say which, on the privacy page, rather than leaving you to assume. Where a creative contains a face, only a bounding box is kept, so no one can be identified from what we hold.

Everyone says they predict winners.Ask two questions.

1. On how many creatives was that measured, and what was held constant?

“Scores concepts before you spend” is a claim about accuracy, and accuracy is a number with a sample size behind it. Format and placement move results enormously on their own — a mobile-native video beats a display banner whatever is in it — so a correlation that did not hold those constant is measuring the media plan and calling it the creative.

2. Show me the ones it got wrong.

Any tool can show you its hits. A tool that cannot show you a creative it scored highly which then underperformed is not validating anything — it is reporting its own output back to you.

Our answer: we do not quote an accuracy figure, because we have not yet measured one on a sample worth quoting. What we do is put the prediction next to your delivered results, on your own campaigns, and show the disagreements as plainly as the matches. When we have a number that survives those two questions, we will publish it — with the sample size.

Start by seeing your own ads

We will score the creatives you are running right now, and up to three competitors’, from public ad libraries. No account access, no login, no cost. Send us a brand name.