MIRA uses several kinds of evidence. Their sources, limitations and comparison context determine how much weight to give a result in your creative review.
Evidence sources
Predictions and participant results have different meanings.
Evidence
Where it comes from
How to use it
Creative Strength
An AI prediction against a human-coded creative framework.
Use the overall read with its model provenance and eligible reference comparison. Human training labels do not mean people tested this particular ad.
Brain Mapping
A modelled response curve generated from the creative.
Inspect relative rises and falls within the film. The curve is not a participant brain scan or a forecast of campaign results.
Predicted visual attention
A model-generated spatial heatmap over the creative.
Examine where attention may land. It is separate from gaze collected from participants.
Creative coding
AI analysis of scenes, brand cues, product, text and audio where available.
Check the detected evidence against the film. Incorrect brand or scene identification can affect the interpretation.
Participant evidence
Responses collected in a defined MIRA Validate study, using the enabled measures.
Interpret the actual answers and recordings with the sample, method and quality checks. Confirm supported measures before commissioning a study.
Reading the report
Read the number with its context.
Score and percentile
A score describes the model’s assessment of the creative. A percentile describes its position within a particular reference set. A high percentile in one set does not establish the same position in another.
Check the model, version and reference scope recorded with the report before comparing results.
Supporting diagnostics
Diagnostics such as memorability, relevance or message comprehension help frame a question for review. They are model-derived assessments unless the report explicitly identifies a collected participant measure.
Use a recall or comprehension study when the decision requires direct evidence of what people remembered or understood.
Current validation evidence
What has been reported for Creative Strength?
Reported by MiAtlas
More than 70% agreement on ad-pair rankings.
Ads in the test were excluded from model training.
MiAtlas reports that Creative Strength ranked ad pairs in the same order as expert human ratings more than 70% of the time, in a test using ads excluded from model training.
This result concerns the order of ad pairs against expert ratings. It does not establish individual-score accuracy or predict sales.
This result has not yet been confirmed for social-media ads.
Scope of the result
This evidence is for Creative Strength only, not Brain Mapping, predicted gaze, memorability or likability.
Training a model on expert ratings and testing a finished ad with an audience answer different questions.
Details still needed
The full test report is not yet available in MIRA. We still need the test size, uncertainty range and confirmation that it covers this model version.
How close are the scores? MiAtlas has not yet supplied the share of test ads scored within 5 or 10 points of the human rating.
Read the reference scope with every percentile. A comparison is most useful when its category, market, format and duration are relevant to the decision.
The reference set
Check the reference identity, eligible ad count and available context in the report. Avoid treating a broad reference as a matched category or placement comparison.
Social and TV scope
A social-media result may use a TV Creative Strength reference. That comparison provides TV-reference context; it does not establish a social-specific norm.
A private comparison set
Your scored library can provide context from your own work. Keep creative versions, model versions and test conditions compatible when comparing results.
Where a supported comparison is unavailable, the report should explain the missing context. A reference set’s size alone does not establish predictive validity.
Evaluating MIRA
Agree what success means before the pilot.
01
Define the decision
Choose a question such as variant ranking, brand timing or a specific edit.
02
Choose suitable ads
Include work with known outcomes and agree a consistent analysis context.
03
Review independently
Record your team’s assessment before comparing it with MIRA’s findings.
04
Check the result
Assess the agreed question. Campaign outcomes also depend on audience, offer and delivery.
Can MIRA tell me which ad will deliver the best ROAS?
MIRA supports creative review. It does not establish a ROAS forecast. Media outcomes depend on factors beyond the film, including targeting, offer, delivery and the conversion experience.
Does Brain Mapping measure my audience’s brain activity?
Brain Mapping generates a prediction from the ad itself. No participant brain scan is used in that workflow.
Is the heatmap measured eye tracking?
The predicted visual attention view is generated by a model. Participant gaze is a separate study output, with its collection method, calibration and usable sample.
Does a higher revised score prove the edit worked?
It shows a change in the model assessment under the recorded conditions. An audience study or campaign experiment is needed to assess the corresponding audience or performance outcome.
Can I see how this applies to my ad?
Upload your own video ad. Your first two analyses include the full report, with scores, supporting evidence and recommendations.
See it on your own work
Discuss the evidence your decision needs.
Bring the creative question, intended placement and audience. We can discuss a suitable evaluation and the available evidence.