AI Tech Paper

Introduction

The Power of AI Based on Neuroscience

Predicting consumer behavior is challenging because traditional methods fail to access the subconscious drivers of choice, while scientific methods like neuroscience have historically been too slow and expensive to apply at scale. Artificial Intelligence promises to overcome this by delivering neuroscience-level insights with the speed and scalability modern marketing requires. However, the critical pitfall is that an AI's predictive power is fundamentally limited by its training data, as models trained on generic information cannot forecast the nuanced emotional and cognitive responses of actual consumers.

This technical paper introduces Databrain AI, a predictive platform engineered to overcome the limitations of both traditional research and generic AI. The Databrain AI suite is built on a "golden foundation": one of the world's largest proprietary consumer neuroscience databases, meticulously curated over more than a decade of dedicated research. Our dataset is a highly specific and relevant corpus of human behavioral data, capturing how consumers attend to, feel about, and process marketing materials in social, out-of-home, print, and other media. The superior accuracy of Databrain AI models is a direct result of this unique data asset, which contains over 100 billion consumer behavior data points from more than 100,000 participants tested annually.

This document provides a transparent and in-depth exploration of the science, technology, and validation protocols that underpin the Databrain AI platform. It will elucidate the theoretical framework guiding the platform's development, detail the architecture of the predictive models, define the methodology behind each metric, and present the rigorous, multi-tiered validation that substantiates the platform's claims. The objective is to provide neuromarketing experts, insights managers, and marketers with a comprehensive understanding of the scientific and technical foundations of Databrain AI, demonstrating its validity as an indispensable tool for creative effectiveness.

The Science Behind It All

The platform was founded on the work of leading behavioral scientist Dr. Thomas Zoëga Ramsøy — PhD in Neurobiology, Certified Neuropsychologist, Assistant Professor in Marketing & Neuroscience, and CEO at Neurons.

The operating ecosystem is built on the technological integration of Neurons Inc.

• Data processing

  Databrain audits the raw biometric data supplied by the underlying technology and turns it into structured guidance for creative optimization.

• Empirical execution

  Algorithmic evaluation makes it possible to assess visual assets at scale, through repeatable and measurable protocols.

• Market efficiency

  Every compositional change rests on verifiable attention and cognitive-load metrics, which keeps media spend efficient and improves return on investment.

• Scientific authorship

  Bestselling author of four reference works in behavioral science.

• Institutional validation

  Models developed in collaboration with researchers from leading universities; 254 articles and academic publications cited around 5,000 times.

• Academic leadership

  Curriculum design and technical training in behavioral science and applied neuromarketing, with more than 145,000 students taught.

Why Do Implicit Responses Matter in Advertising?

Human Decision is Neither Optimal, Unconscious, nor Rational

Traditional Explicit Research Is Facing a Problem

It is no longer a secret that our choices and decisions are often driven by unconscious processes and influences. This understanding has been supported by decades of research, culminating in the 2002 Nobel Prize in Economics awarded to psychologist Daniel Kahneman.

A major portion of traditional market research is based on surveys, interviews, focus groups, and other queries about people's thoughts and feelings regarding their behavior. However, if consumer choices are even partially driven by unconscious processes, then we require a different set of tools to fully understand what influences decisions.

Predicting Consumer Response Through the Unconscious Mind

Implicit research methods uncover dimensions of consumer behavior that explicit measures might miss. Techniques like Fast Response Test (FRT), eye tracking, and neuroimaging tap into consumers' automatic reactions to brands, products, or ads. These methods capture quick, spontaneous responses, offering a more direct window into subconscious preferences and biases. For instance, Knutson et al.'s study demonstrates that brain activity can predict choices more accurately than self-reports. Similarly, Dmochowski et al. showed that brain responses in a small group can forecast broader market reactions.

The ˘4 Powers' of High-Impact Ads

Knutson et al. (2007) found that activation of specific neural circuits tied to anticipatory emotion precedes and shapes purchase decisions. Preference, for instance, is linked to activity in the nucleus accumbens (NAcc) during product exposure and price evaluation.

The temporal cascade of the response

1. 50-100 ms

   Attention — the power to orient

   The first threshold. A stimulus that lacks early visual salience is excluded from everything that follows.

2. 300 ms

   Cognition — the power to convey

   Semantic decoding. Measures how closely the intended message matches what the viewer actually processes.

3. 400 ms

   Emotion — the power to motivate

   Affective reactivity sets the valence and the bias of the advertising experience.

4. 1000 ms

   Memory — the power to endure

   The asset is encoded into long-term structures, which is what makes later recall and brand equity possible.

Data Collection Protocols

At our core, we believe that groundbreaking insights are built on a foundation of unimpeachable data. Our multi-layered, scientifically-backed protocols ensure that every data point we collect is representative, reliable, and replicable. This commitment to quality means you can make critical business decisions with absolute confidence.

Representative & Robust Sampling

The validity of any study begins with its participants. We employ rigorous sampling methods to ensure our results are a true reflection of the target population.

Scientifically Determined Sample Size

Our baseline recommendation is a minimum of 30 participants per coherent group (e.g., a specific age range or income level) for lab studies and 100 participants for online studies. These numbers are derived from established power and effect size calculations, ensuring statistically valid, reliable, and replicable findings from all behavioral studies.

Targeted Demographic Representation

For studies requiring comparisons across different segments we scale our methodology accordingly. For example, a study comparing three distinct age groups would require a minimum of 3×30 participants for lab studies and 3x100 for online studies. We structure recruitment to meet specific demographic needs, such as ensuring gender balance or precise age brackets.

Built-in Contingency

To safeguard against potential dropouts and ensure a valid final sample, we systematically over-recruit by 10%.

Reliability of data by sample size

To ensure the association word ratings used in Databrain AI are trustworthy, we tested how measurement reliability changes with sample size. As sample size increases, the average standard deviation across all association words drops steeply before flattening out  demonstrating that at ca. 150 respondents, the data reaches a consistently low level of uncertainty and can be considered highly reliable.

Data Quality Assurance for our Predictive AI Models

Multi-Layered Quality Assurance

In-Lab Protocols (Eye-Tracking & EEG)

Online Panel Protocols

Our lab studies undergo a meticulous 7-step quality check to eliminate confounding variables and guarantee pristine data. Key checks include:

Our online testing platform integrates proprietary quality measures that go beyond industry standards.

Participant Screening: We confirm participants have not

Real-Time Performance Monitoring: We monitor participant reaction times to ensure engagement. Responses that are too fast or too slow fall outside our predefined norms (e.g., 0.3s - 2.5s for Emotional FRTs). Participants are warned, and if more than half their responses are outside the acceptable range, they are disqualified from the study. consumed alcohol, caffeine, or nicotine prior to the session to ensure unaltered cognitive and physiological responses. on all hardware and software before each session. We

Technical & Calibration Integrity: We run full diagnostics

Methodological Refinements: We actively identify and mitigate potential biases. For example, to counteract the common "first- word response time effect," we strategically place the word "Great" at the beginning of FRT sessions, a method proven to normalize subsequent reaction times. exclude any data from participants who fail our strict eye-tracking (e.g., <80% fixation accuracy) or EEG calibration procedures.

Data Signal Quality: Recordings with insufficient data

(e.g., less than 65% of eye-tracking data captured) or

Respondent Training: Every participant completes a practice task to ensure they fully understand the instructions before the study begins, leading to higher-quality responses. excessive noise (e.g., EEG signals consistently 3+ standard deviations from the mean) are flagged and excluded from analysis

Data Privacy & Anonymization

Data collection through the underlying technology is subject to the strictest international standards: the European Union General Data Protection Regulation and the ethical regulations set out in the Declaration of Helsinki.

All biometric and survey data is fully anonymized using a unique log number, with no link to personal information.

All staff handling data are bound by strict non-disclosure agreements.

Personal data files are securely discarded upon project finalization.

Attention and Visual Modelling

Why is Attention Critical?

Attention is paramount in marketing due to its role in engagement and visibility. It's the primary trigger of consumer decision-making, making your message stand out amidst digital noise. Furthermore, understanding what garners attention enables content personalization, thus boosting campaign effectiveness. Simply put, attention is the first and most critical step in any successful marketing strategy.

Eye-tracking is a great way to measure consumer attention.

In this award winning ad, the brand is not seen.

Data Foundation: The Source of Truth

The exceptional accuracy of the Databrain AI attention model is a direct consequence of the unparalleled data on which it is trained. The model is not built on generic saliency datasets but on the proprietary consumer neuroscience database, which, as of 2025, includes:

Over 1B eye-tracking data points. Over 8,000 ads tested, with 20% new data added every year. Participant panel size: 30 40 per test. Data from over 15 distinct consumer contexts, including digital advertising, social media ads, out-of-home (OOH), print advertising, packaging, and more.

The ground-truth data for the attention model is sourced exclusively from high-precision, laboratory-grade eye-tracking studies conducted by Databrain.

These studies utilize state of the art eye-trackers like the Tobii X2-30, Tobii Pro Nano, Tobii Glasses Pro 2, and earlier the Tobii T60-XL to ensure the highest possible fidelity in capturing fixation points, gaze paths, and attentional dwell time.

The high-quality data, as well as our model evaluation practices, ensure an accuracy of 95%+ compared to ground-truth eye- tracking data.

Deep Learning Computer Vision Model

The attention model is a fully convolutional autoencoder featuring a sophisticated encoderdecoder architecture that integrates state- of-the-art components like ConvNeXt and ResNet50 to ensure a balanced and accurate output.

A critical decision was to train the model "from scratch" exclusively on the proprietary Databrain database, which contains vast amounts of consumer-specific eye-tracking data.

This approach ensures the model's features are entirely optimized for predicting consumer attention, avoiding the biases of generic public datasets. The training process was rig orously monitored, with performance evaluated using statistical measures such as Kullback Leibler divergence (KLD) and Mean Squared Error (MSE) to guarantee a high level of predictive accuracy (95%).

The Database

The AI is built on consumer eye-tracking data from our research studies. The database is one of the world's largest single databases of high-quality eye-tracking data, with well over 20.000 participants from around the globe exposed to consumer-related stimuli, such as watching ads on social media or TV, or reading physical newspapers.

Machine Learning

Using our database, we generated attention heatmaps from the eye-tracking data, and almost 200 distinct machine learning models were trained and compared to produce the best possible model prediction. For each model training, one portion of the data was randomly selected for training and a second portion for validation test.

High Fidelity Attention Prediction

Through extensive testing and validation, uploading an image to Databrain AI now gives you a heatmap that is statistically equivalent to a heatmap built on actual eye-tracking data from showing ca. 150 people the image for 5 seconds with over 95% accuracy, and it only takes a few seconds.

Beyond a Simple Salience Map

Video Prediction Modelling

Predicting attention in dynamic video content presents unique challenges. To address this, a specialized video attention model was developed. This model is distinguished by its unique 2-second floating window operation, which analyzes video in overlapping temporal segments to predict attention with human-like dynamics. The development of this feature demonstrates a deep understanding of visual cognition. Early model iterations exhibited a flaw: they would predict an instantaneous shift of attention to a newly appearing object. However, human ocular behavior exhibits a slight delay; attention often lingers on the previous point of fixation for a fraction of a second after a scene change before saccading to a new target.

The floating window operation was specifically engineered to replicate these

˝nuanced lags in human attention,˛ resulting in a model that is not only accurate but also behaviorally realistic. This level of sophistication is not merely an architectural choice; it is an emergent capability that the model could only have learned because it was trained on thousands of hours of real human eye-tracking data exhibiting this exact behavior, thereby causally linking the superior data foundation to the model's as advanced technical performance.

Start and End Attention

Our models predict where consumers focus during the first and last 2 seconds of a 5-second ad exposure.

Separate AI models handle each attention window (start/end). This feature helps users quickly see if key elements like the brand or CTA grab attention instantly or need more time to be noticed.

Attention Dispersion: Focus

The focus score is a derived score from the attention heat map. A higher Focus score is when only a few items are likely to grab attention

The Focus score measures how viewers' visual attention is distributed across content elements, based on the concept of "attention dispersion" from eye-tracking research by Teixeira, Wedel, and Pieters.

The Focus score identifies whether viewers' eyes are likely to be concentrated on specific areas, indicating heightened focus, or if they will be scattered across the content, suggesting a more dispersed or distracted attention

Precision Where It Matters: AOIs & AOI Detection

What Are AOIs?

While overall attention patterns are informative, marketing effectiveness ultimately depends on whether attention is directed to specific, commercially relevant elements. The Databrain AI platform provides powerful tools for this granular level of analysis through Areas of Interest

(AOIs). AOIs are predefined regions within visual content that aggregate attention data within those areas. They are pivotal for deep level analysis and are the foundation of optimizing creatives. developed a proprietary object detection model specifically for advertising.

Proprietary AOI Detection Model For Images and Videos

To automate and scale working with AOIs, Databrain has

This model, based on advanced deep learning techniques, is trained to automatically identify five critical AOI types:

Branding, Product, Headline Text, Body Text, and Call to

Action (CTA).

Unlike general-purpose object detection models, the Databrain AOI model is a trained specialist. It has learned the specific visual language of advertising how logos, products, and text appear in diverse and often cluttered commercial contexts. This specialization has resulted in a model that "outperforms the Google Vision logo detection model by 40%," providing a more reliable and accurate foundation for AOI-based metrics.

A Note on Gaze Mapping

Unlike some competitors, Databrain AI does not offer gaze mapping scores. This is a deliberate methodological choice, since gaze mapping exhibits too high individual variance to produce stable, reliable predictions. Offering it would mean compromising on the accuracy standards that underpin the Databrain platform.

Supplying Areas of Interest (AOIs) yields precise results on the elements that matter to your objectives: branding, headline, product, body text and call to action.

Precision Where It Matters: Attention Scores

Attention Scores

Total Attention

The Total Attention metric is derived from our proprietary Total Attention Heatmap algorithm. This algorithm processes raw attention data, normalizes it across different viewer demographics, and maps this data onto the AOIs of the tested asset. The final score is an aggregate measure of the intensity and frequency of customer attention on each AOI during the entire exposure.

Time Spent

The Time Spent metric uses our refined attention-tracking algorithm to model how long consumers pay attention on

AOIs. It captures the duration of attention over a 5-second exposure window, tracking moment-to-moment gaze fluctuations. The cumulative time spent on each AOI is then converted into a score, indicating the staying power of individual elements.

Percentage Seen

Derived from the Total Attention Heatmap, this metric estimates the percentage of viewers likely to notice a specific

AOI. It uses polynomial regression to link attention scores to the percentage seen and incorporates raw image data, predefined AOIs, and historical attention data to produce an output between 0-100%, indicating expected viewer engagement.

Availability

Availability measures the total amount of time in seconds that an AOI is present in the video. The cumulative AOI availability is then translated into a score, showing the staying power of individual elements in the video.

Relative Attention

Relative attention is calculated by dividing the sum of the attention the AOI receives, by the number of frames where the AOI was detected. This metric provides another perspective to AOI-level attention enabling users to ensure that when the AOI is visible, it receives enough attention to be impactful.

Attention scores are part of the metrics provided by Databrain AI. They are represented as a pixel-by-pixel heatmap of how human attention is distributed across a creative asset.

Validation: A Commitment to Scientific Accuracy

Our multi-tiered validation protocol

To ensure scientific rigor, we established a comprehensive, multi-tiered validation protocol to assess the accuracy of our attention models against the gold standard of laboratory eye-tracking. It relies on three distinct "pillars of truth," each measuring a different aspect of the prediction's quality.

Pearson Correlation Coefficient (CC): The "Synchronization Score"

The Pearson Correlation Coefficient measures the linear correlation between two variables (the pixel intensity arrays of the

Human map and the AI map). It is the ratio of their covariance to the product of their standard deviations. The value ranges from -1 (perfect negative correlation) to +1 (perfect positive correlation), with 0 indicating no correlation.

To explain CC, we use the analogy of ˝synchronized dancing.˛ J Just as two dancers might move in perfect rhythm, this metric measures how closely the AI's predictions rise and fall in sync with actual human eye movements. If humans focus intensely on a logo, the AI must predict intense focus on that same logo. A score of 1.0 would mean perfect synchronization. The Databrain AI model achieves a correlation of approximately 95% for total viewing time, indicating that the AI's "rhythm" is virtually indistinguishable from aggregate human behavior.

Kullback Leibler Divergence (KLD): The "Allocation Error"

Kullback-Leiber Divergence is a measure from information theory that quantifies the difference between two probability distributions: the true distribution (Human) and the predicted distribution (AI). Think of the viewer's attention as a budget they have a limited amount of time and mental focus to ˝spend˛ on an image (100% of their attention). The different parts of the image (the logo, the face, the product, the background) are different stocks or assets they can invest in. The Human

Distribution (Truth) represents the market reality. Real humans might invest 60% of their attention budget on the face, 30% on the headline, and 10% on the product. The AI Prediction represents the fund manager's strategythe AI predicts how that attention budget will be allocated. KLD is the ˝allocation error.˛ It measures the mathematical distance between the manager's strategy and the market's reality.

If the AI predicts an investment of 50% on the face and 40% on the headline, t

We want the KLD score to be low (close to 0). A low score proves the AI understands not just that people look, but exactly how much they look at each element compared to reality.

Similarity Measure (SIM): The "Overlay Test"

The Similarity Measure, also known as histogram intersection, calculates the sum of the minimum values at each bin (or pixel) of two normalized histograms (distributions). This is a geographic assessment. If we were to print the human attention map on a blue transparency and the AI map on a yellow transparency, the SIM score measures how much of the image turns green when we overlay them. It quantifies the physical intersection of the two maps.

While the synchronization score confirms that the intensity matches, the overlay test confirms that the location is precise. using a hybrid dataset of open-source benchmarks and proprietary commercial assets, Databrain AI establishes a scientific basis for trust, moving predictive analytics from a "black box" to a transparent, verified business intelligence tool.

By validating the model against these three complementary metrics

A Multi-Dimensional Approach to Validity

Validating a predictive AI model goes beyond a single accuracy metric. Drawing on peer-reviewed research (Ramsøy, JAR

2019), Databrain AI has established validity across multiple dimensions:

Ecological validity  predictions reflect real-world viewing behavior, not just lab conditions

Construct validity  the model measures what it claims to measure

External validity  findings generalize across contexts, formats, and audiences

Internal validity  the model's logic and structure are sound

Beyond validity, we have assessed sensitivity and specificity to confirm the model detects attention signal accurately and avoids false positives. Reliability has been established through Monte Carlo simulation methods, which also inform recommended sample sizes and confirm high between-group consistency.

We have further run clustering analyses  including Cronbach's Alpha and PCA  to identify groups of highly interrelated metrics (e.g., across buy, try, and recommend behaviors), ensuring our scores are meaningfully distinct rather than redundant. not just statistical performance.

Finally, our metrics have been continuously calibrated with cli

Image Attention Model

The comparison of the image model is done against the MIT/Tuebingen Saliency Benchmark. It shows the performance of the Databrain image attention prediction against the leaders in predicting saliency maps based on the MIT open source dataset. Note that the Databrain AI model is specialized to advertising materials so it achieves higher accuracy scores than the MIT benchmark which includes natural images.

0.824

0.699

0.347

You can find the leader board here.

Video Attention Model

The comparison of the image model is done against the Video saliency prediction leaderboard. It shows the performance of the Databrain video attention prediction against the leaders in predicting saliency maps based on the different open source datasets. Note that the Databrain AI model is specialized to advertising materials so it achieves higher accuracy scores than the MIT benchmark which includes natural images.

0.5461

0.4068

You can find the leader board here.

Mixed Attention Model (Beta)

Our new mixed media ads attention model is in beta. Currently, there are no saliency prediction research models specialized in mixed media stimuli. KLD has not been calculated due to the limited testing dataset of the model. Databrain will work to gather more testing data in the coming period.

For CC and SIM, a higher value means closer agreement with real eye-tracking data. KLD works the other way around: it measures the distance between the human and the predicted distribution, so a lower value is better.

Predicting Emotion and Quantifying the Unconscious

Emotions vs Feelings

While the terms ˝emotions ˛ and ˝feelings˛ ar are frequently used interchangeably, it is crucial to understand the distinction between them.

Emotions are automatic and unconscious reactions in the brain and body automatically triggered by internal or external factors. Emotional responses to stimuli arise in less than a half-second, which is why they are often described as a gut feeling.

In contrast, feelings are the conscious experiences of those emotional reactions. They only arise once unconscious emotional responses are brought into conscious awareness. Feelings can therefore never arise without prior emotional reactions. Databrain metrics are focused on emotions, which are stronger predictors of actual choice and behavior, both individually and at the market level.

In-depth: The Fast Response Test (FRT) Methodology

The scientific engine behind the Engagement score is a proprietary Fast Response Test (FRT). The FRT is a type of implicit association test grounded in decades of cognitive psychology research, which demonstrates that the time it takes to make a judgment is a reliable indicator of the strength of the underlying mental association.

Instructions: In a typical FRT study, a participant is briefly exposed to a creative asset. Immediately following the exposure, they are presented with a series of attribute words or phrases (e.g., "Engaging," "Trustworthy," "Innovative"). Their task is to respond "Yes" or "No" as quickly and accurately as possible to indicate whether the word describes their feeling about the asset.

The core principle is that a faster "Yes" response to a positive attribute indicates a stronger, more automatic, and more deeply held positive association. Conversely, a slower or "No" response suggests a weaker or non-existent association.

FRT Calculation

The methodology combines three key elements:

A series of binary (yes/no) questions about an asset. Response Time (RT): A precise measure of response latency recorded in milliseconds. RT is recorded as a proxy for certainty. Group Consensus: We calculate group consensus as a proxy for in-market effects.

What is Response Time (RT)?

Response Time (RT), also known as reaction time or response latency, refers to the time it takes for behavioral responses to occur during a particular task (Donders, 1969; Luce, 1991). The RT usually refers to the time between the presentation of the external stimulus and the appropriate response (Posner, 1978). Critically, RT is affected by emotional and cognitive processes, thus allowing it to be used as an index of unconscious emotion, motivation, and cognitive processing.

Data Weighing

We weigh Yes/No responses with the response time to assess not only whether participants say yes or no, but how certain they are about their response. Faster responses indicate higher certainty and are up-weighted, while slower responses indicate lower certainty and are down-weighted. Studies have shown that group coherency is extremely good at predicting in-market effects. That's why we also weigh the agreement among all yes/no responses.

For example, if 50% answer "yes" with high variance in certainty, and 50% answer ˝no˛ with comparable certainty, the ˝no˛ weighs more.

Metrics calculation and prediction

The precise calculation of FRT-based metrics is designed to move beyond simple agreement to capture the true strength of an unconscious association. The final score is a sophisticated calculation that integrates three key data points: the binary 'yes' or 'no' response, the response time measured in milliseconds, and the consensus across the entire participant group.

In this model, faster response times for affirmative ('yes') answers are given greater weight, as this indicates a stronger, more automatic positive association.

By normalizing these individual response times and weighting them against the group consensus, the methodology produces a highly reliable and nuanced score that accurately reflects the collective gut reaction to an asset.(

To build our FRT-based predictive models, Databrain AI collects extensive ground-truth data from FRT studies with over 25,000 participants. By training advanced machine learning models on this vast and specific dataset, the platform can predict these implicit responses directly from a visual asset with an outstanding accuracy of over 90% when compared to the actual human response data.

Explainable AI in Action: Opening the Black Box

A core principle of the Databrain AI platform is a commitment to transparency and actionability. To show users why an asset received a particular score, the platform employs the "Grad-CAM (Gradient-weighted Class Activation Mapping)" technique.

Grad-CAM is a visualization method that produces a heatmap highlighting the specific regions in an image that were most influential in a model's decision. This explainable AI (XAI) approach is applied differently depending on the metric.

For the FRT-based models like Engagement, Grad-CAM heatmaps visualize what drives the predicted implicit response. In these heatmaps, red zones signify areas that strongly contribute to a high positive emotional response, while green zones represent areas with lower emotional impact. This transforms an abstract score like Engagement into tangible, visual feedback.

Engagement Trust Avoidance Visual engagement is a quantitative

Trust score measures how much a

The Avoidance score measures the measure of the emotional appeal of a creative based on FRT responses to creative feels credible, sincere, and degree of emotional rejection, disinterest, emotional association words such as emotionally aligned to the viewer. It or irritation viewers feel toward a creative.

Engaging, Happy, Interesting and reflects subconscious, intuitive

This is the first AI-based metric designed

Attractive. perceptions of honesty and authenticity, to quantify negative sentiment from a gut- based on quick reactions to words like Trustworthy, Honest and Sincere. level, subconscious perspective. It is derived from associations with the

This metric provides a new way to words Annoying and Boring, and reflects a assess brand credibility, moral alignment, creative's potential to repel, frustrate, or and emotional resonance in advertising fail to connect with viewers. essential signals for building brand equity and long-term loyalty.

Accuracy Stats

Engagement

Trust

Avoidance

0.91

0.93

0.89

How the emotional response unfolds

1. Stimulus

   Neural processing

   The amygdala and the thalamus process the stimulus before any conscious elaboration.

2. Unconscious

   Unconscious emotions

   Bodily responses fire without any involvement of the will.

3. Conscious

   Conscious emotions

   Global activity across the cortex, including the ventromedial prefrontal cortex, gives rise to conscious experience.

The three emotional metrics

• Engagement

  A quantitative measure of a creative's emotional appeal. It is grounded empirically in FRT responses to positively valenced association words such as appealing, happy, interesting and captivating.

• Trust

  Measures how credible, sincere and emotionally aligned with the viewer a creative feels. It reflects intuitive, unconscious perceptions of honesty and authenticity, derived from fast reactions to attributes such as trustworthy, honest and sincere. It gives brands an analytical standard for credibility, moral alignment and emotional resonance — the vectors behind long-term brand equity and loyalty.

• Avoidance

  Measures the emotional rejection, disinterest or irritation viewers feel toward a creative. It is the first AI-based metric designed specifically to quantify negative sentiment from a visceral, unconscious perspective. Derived from empirical associations with concepts such as annoying and boring, it surfaces the operational risk that an ad repels, frustrates or simply fails to connect.

Cognitive Demand: Decoding Visual Complexity

What is a Cognitive Response?

Cognitive responses are the mental processes triggered by stimuli, such as ads. Understanding these responses is crucial for ad success, as they reveal how people comprehend, form thoughts, and potentially behave. Ads should balance simplicity and complexity to avoid confusion and boredom, ensuring they engage and maintain interest.

Cognitive Demand and Cognitive Load

The Cognitive Demand score is a quantitative measure of the visual complexity and information load of an asset. A high score indicates that an asset is dense, cluttered, or difficult to process, which can lead to viewer confusion, message abandonment, and poor memory encoding. The technical foundation of this score is rooted in the robust mathematical concept of "Shannon entropy" from information theory (Bundesen, 1999). In this context, entropy serves as a proxy for the amount of information or "surprise" in an image; higher entropy corresponds to a more complex and unpredictable visual signal that requires greater cognitive resources to process. This mathematical underpinning ensures the score is objective, reliable, and consistently reflects the cognitive effort required to parse a visual scene.

Managing cognitive load is essential for effective advertising. Cognitive load refers to the mental demand placed on working memory, which can only handle limited information at a time. Overloading ads with too much information can lead to cognitive overload, resulting in loss of attention, reduced comprehension, and negative brand perception.

Application in Video: The Doorway Effect

In the context of video, the Cognitive Demand score becomes a powerful tool for narrative and pacing analysis. The platform calculates the score on a frame-by-frame basis, revealing fluctuations in cognitive load throughout the video's duration. Abrupt spikes in the Cognitive Demand score often correspond to scene changes or hard cuts. These moments are critical because they can trigger a cognitive phenomenon known as the "Doorway Effect" or "conceptual closure". This effect describes a temporary lapse in working memory that occurs when transitioning between distinct contexts. By identifying these high-risk transitions, the platform provides creators with actionable insights to optimize the pacing of their video, ensuring that key information, such as branding or a key message, is not presented immediately following an abrupt scene change where it is more likely to be forgotten.

Example of how a single ad can have a few notable scene shifts, as marked as abrupt shifts in the Cognitive Demand score between green and purple phases. Many ads have more rapid scene shifts. Each abrupt scene shift is related to an increased risk of conceptual closure, where key information can get lost.

How the model was selected

Among many candidate scoring models for cognitive demand, we selected the one with the best predictive ability for comprehension and information processing.

Applied to video

In video, the cognitive demand score becomes a high-leverage instrument for auditing narrative and pacing. The platform computes the metric frame by frame, exposing how cognitive load fluctuates across the full duration of the piece.

Sharp peaks usually correspond to scene changes or hard cuts. Those moments are critical because they can trigger a cognitive phenomenon known as the event boundary effect: a brief gap in working memory that occurs when moving between distinct contexts.

By identifying these high-risk transitions mathematically, the platform provides actionable guidance for pacing — making sure that critical information, such as the brand appearing or a key message, is not delivered right after a hard cut, where the odds of it being missed and forgotten are highest.

A single ad can contain several of these shifts, visible as abrupt changes in the score. The faster the cut sequence, the higher the risk of event boundaries and of key information being lost.

Predicting Memory with AI

What is Memory?

Non-declarative memory, or implicit memory, refers to the unconscious memory processes that influence our behavior and skills, like riding a bike or forming habits. It cannot be expressed verbally, as it cannot be consciously brought into awareness. Surveys are ineffective in assessing this type of memory, but it can be measured through observing automatic patterns or behavior.

Declarative memory, also called explicit memory, involves the conscious recall of facts and events, such as remembering what you did yesterday, or the meaning of the word

˝neuroscience˛. . It can be measured by testing how much and what people remember through tasks like spontaneous recall and recognition with cues for instance.

The Implicit Brain

How do we know the existence of implicit associations? Beyond the findings showing that implicit associations are related to behaviors, neuroscience studies have also demonstrated this effect.

For example, we* recently used fMRI brain imaging to show that brand associations unconsciously triggered the brain's memory network, including the hippocampus. In the study, stronger activity in the hippocampus to the viewing of brand mascots such as the Michelin man predicted a higher number of associations to the mascots in a later task.

That is, even when we look at brands or brand derivatives in a passive manner or while doing something els e, our brain automatically activates the associations we have with these brands .

Ad memory

Brand memory

Ad memory refers to consumers' ability to recall and recognize ads they have seen.

Brand memory refers to consumers' ability to remember and associate an ad with its brand. (

It is crucial for assessing the effectiveness of ad campaigns, indicating how well an ad leave s an impression on consumers' minds. he brand behind the ad, indicating how effectively the ad embedded the brand in consumers' minds.

This involves identifying t

Ideally, each ad should aim to achieve top-of-mind (TOM) memory, where consumers immediately recall the brand without any cues. Such recall suggests that a brand has a strong presence in the consumer's mind.

MEMORY

The Memory score predicts ad recall, providing a probability score on a 0-100 scale that represents the likelihood of an asset being remembered by a group of viewers after brief exposure (<2 sec) and distraction.

The methodology of predicting ad memory

The Databrain Memory AI utilizes a data-driven approach derived from the Visual Memory Game methodology developed by MIT. The original paper, along with subsequent research using this methodology, has provided a strong scientific foundation and initial validation for our metric.

The study involves exposing participants to a rapid sequence of images. Participants must indicate when they see an image for the second time. This rigorous methodology captures both reaction time and accuracy, thereby integrating cognitive engagement factors into the memory score. The datasets used to train the model include a wide range of advertising materials, covering various industry use cases  from print and Out-Of-Home (OOH) displays to digital platforms such as websites and social media.

Databrain has gathered data from over 7,000 general population participants. This dataset is continuously enriched to ensure that the model captures the latest participant behavior and evolving trends in advertising.

Scatter Plots of Model Memory Score vs.

Predicting Memory on Video

Memory Recall Studies Score

For video assets, each frame is scored individually using the image prediction model. This model operates under the constraint of not utilizing sequential video data or audio signals.

To further ensure the robustness of our memory metric, we correlated predicted video scores with ground-truth data from more than 100 videos. These videos were evaluated for ad recall by over 1,000 participants using our Memory Recall Test. This methodology is based on highly reliable approaches used to measure memory functions and is adapted from proven methods originally developed for diagnosing Alzheimer's disease.

The Databrain AI Memory Score shows strong alignment with human testing results, suggesting that sustained memory engagement throughout an ad is a strong predictor of ad recall and overall memory performance. This alignment between human testing and AI predictions indicates that the Databrain AI Memory Score is a reliable proxy for ad memorability in video content.

From Prediction to Suggestion: Databrain AI Objectives

Objectives to Guide Your Assessment

Setting up objectives in Databrain AI is important because they determine how your creative is evaluated and which benchmark data it is compared against. The selected objectives ensure that your predictions are benchmarked against similar assets, allowing you to receive meaningful AI-generated insights. This also ensures that your Databrain Impact Score (NIS) is calculated in a fair and contextually accurate way.

Purpose Definition

Brand Building: This strategy focuses on increasing brand awareness and recognition, not immediate sales. Brand Building campaigns create a trustworthy image of your brand through storytelling content like educational infographics or engaging videos. These campaigns cultivate familiarity and trust, laying the foundation for a sustainable relationship with customers.

Conversion: Conversion campaigns leverage the brand equity built through BB efforts. They are targeted initiatives designed to convert brand awareness into action, such as purchasing a product, signing up for a newsletter, or following social media. Conversion campaigns are direct and purpose-driven, featuring dynamic and urgent visuals and texts to prompt immediate responses.

Industry Benchmarks

We benchmark creatives across 5 core industry categories and 12 specialized subcategories, giving you access to insights tailored to your exact market. Whether you're in FMCG, Finance, Automotive, Telecom, Travel, or Health, your ads aren't measured against generic averages  they're compared to the right competitive set. That means sharper benchmarks, mo re relevant impact scores, and decisions you can stand behind .

Read more about industry categories here.

Use Case Benchmarks

We provide benchmarks for 15 distinct use cases spanning Digital Advertising (including display, social platforms like Facebook, Instagram, TikTok, and YouTube), Traditional Advertising (print, OOH, TV), Product assets (packaging), and Websites (brand and e- commerce), plus an overall

Read more about use case categories here.

• Brand building

  Focuses on raising brand awareness and recognition in the market rather than driving immediate sales. It builds a solid corporate image through narrative content such as educational infographics or high-engagement video. It cultivates familiarity and consumer trust, and lays the groundwork for a sustainable commercial relationship.

• Conversion

  Capitalizes on the brand equity built earlier. These are strictly targeted initiatives designed to turn awareness into a measurable action: buying a product, subscribing to a newsletter, following on social platforms. They are direct, transactional campaigns that use visuals and copy with a sense of urgency to provoke an immediate response.

From Prediction to Suggestion: Databrain AI Benchmarks

Selecting the Right Yardstick for Assessment

Databrain AI benchmarks are the recommended performance ranges for each key metric evaluated by our platform. They indicate what an optimized creative asset should score across metrics like Focus, Engagement, Memory, and more

Data collected from tens of thousands of ads & AOIs

providing you with an industry-standard measure against which to compare your creative's predicted performance. Our benchmarks derive from an ever-growing dataset of more than 12,000 images and over 10,000 videos (with more than 100,000 distinct Areas of Interest or AOIs), gathered from various regions and across diverse industries and use cases. This extensive dataset ensures that our benchmarks are not only representative but also continuously refined as more data is collected. Read more here.

AI Recommendations are based on different AOI types for images and videos

Image AOI Types Video AOI Types

The Calculation Process

Here's the simplified process:

Assets are categorized: Creatives are grouped by factors like industry, use case (e.g., social, display, OOH), and visual elements such as branding or product placement. Metric scores are aggregated: AI predictions for metrics like Focus, Engagement, and Memory are collected across all assets within the same category. Performance ranges are created: The full score distribution is divided into five performance buckets (from low to very high) to show how assets typically perform. Recommended benchmark ranges are set: The top-performing range (usually the top 20%) becomes the recommended benchmark marketers should aim for when optimizing creatives.

These benchmarks are continuously validated with real campaign performance data and decades of neuromarketing research, ensuring they reflect what actually drives results.

Want to see the full methodology? Read the detailed process breakdown here.

• +12,000 images

  Static advertising assets analyzed.

• +10,000 videos

  Audiovisual pieces included in the set.

• +100,000 AOIs

  Distinct areas of interest identified and classified.

From Prediction to Suggestion: The Databrain Impact Score 7.1

Databrain Impact Score

The Databrain Impact Score (NIS) is a single 1-10 score that shows how effective your ad is, before it goes live.

It distills key neuroscience metrics (like attention, engagement, cognition, and memory) into a single score, tailored to your campaign goal: branding or conversion.

Instead of crunching the data yourself, you get one score that shows your ad's impact and tells you what to do next. All backed by 20+ years of neuroscience research.

One Score, Clear Direction

The Databrain impact score is your asset's "pre-flight check." A score of 7 or above means your asset is optimized and a low score is an opportunity to optimize your creative before spending your media budget.

Objective-based prioritization

Each Databrain Impact Score is built on a selection of prioritized metrics tailored to the goal (brand building vs. conversion) and format (image vs. video) of your creative. These metrics were selected because they consistently align with success in real-world campaigns. Each metric is included because it has either: strong theoretical support, or demonstrated data correlation with real KPIs. Metrics that didn't show a consistent impact (e.g., attention on generic body text) were excluded to ensure clarit y and relevance .

Calculation Methodology

To build this comprehensive metric, we first evaluate each prioritized element of your creative individually to assign them their own Impact Scores. Every relevant metric receives a score between 1 and 10, illustrating exactly how well the creative performs against benchmark expectations for its specific industry and use case.

We calculate this by comparing the asset's actual metric results against our recommended benchmark ranges. Assets that hit the recommended range receive a performance bonus, while those falling below receive a proportionate penalty. By factoring in the exact distance from the recommended benchmark range, this dynamic adjustment system guarantees highly detailed scoring, preventing distinct assets from being lumped into the same broad categories.

Once these precise individual scores are finalized, the overall Databrain Impact Score is calculated by simply averaging all of the available Impact Scores. If the creative naturally lacks a specific element, such as an ad that does not feature a physical product, the system intelligently skips that metric. The creative is not penalized for this absence; the system averages only the available metrics to provide a fair and reliable reflection of the elements that are actually present.

How it is calculated

1. 01

   Measure the raw result

   Take the asset's raw value for the metric being audited (for example, Focus = 55).

2. 02

   Compare to the benchmark

   Contrast it with the range specific to the commercial goal and the category.

3. 03

   Place it in a bucket

   Assign it to a scoring band: low, medium or high.

4. 04

   Adjust and convert

   Apply a bonus or penalty depending on how close it sits to the ideal range, then convert to a 1-10 score.

From Prediction to Suggestion: Databrain AI Recommendation Engine

Selecting, Adopting, and Leveraging Large Language Models

Databrain AI's recommendation engine uses API integration with the latest state-of-the-art large language models (LLMs). Databrain AI uses Claude (by Anthropic) for image analysis and Gemini (by Google) for video analysis. Before selecting these models, we conducted a blind evaluation with 200 marketers, insights managers, and designers who assessed the quality and usefulness of LLM-generated responses. Their feedback helped validate our choice of Claude and Gemini as the best-performing models for delivering actionable creative insights.

As we continue to develop our AI insights and recommendations, we aim to always use the highest-performing models available, meaning that our AI insights will continually improve. The current models already support multiple languages and can process audio input.

Our Method of Teaching LLMs About Marketing

We ground the LLM using Databrain's predictive scores and benchmarks so its insights are contextual and aligned with the asset's industry, use case, and purpose. To further improve relevance, we provide AOI-based assessments (Areas of Interest) and brand kit information, helping the model understand key brand elements such as logos, products, and messaging.

LLMs

The LLM analyzes the asset together with the predicted heatmap, benchmarks, and brand context to generate summaries and recommended actions tailored to the marketing scenario. Soon, the model will also incorporate structured insight topics, further improving how recommendations are organized and interpreted.

Using LLMs for Insights Summaries & Topical Analysis

The Insights Page turns Databrain's AI predictions into clear, easy-to-understand creative feedback. Instead of only showing scores and heatmaps, LLMs help explain what is happening in the creative and why it matters. The model analyzes the asset together with predicted attention data, benchmarks, and visual elements. It then organizes the analysis into structured insight topics such as branding visibility, focal points, visual hierarchy, and message clarity.

Each topic highlights what works well, what could be improved, and provides practical recommendations to help marketers quickly optimize their creative.

Client Data Processing Explained

The model uses the client asset and the attention heatmap in the analysis and generation of insights and recommendations. The client retains all rights to their data and the generated AI insights and recommendations are part of this data. None of the data is used for training purposes.

From Suggestion to Generation: Databrain AI Visual Recommendations

Visual Recommendations turn written insights and recommendations into concrete new visual examples: mockups that show what the proposed improvements to an ad would actually look like.

They draw on four inputs: the original creative, the scores for attention, engagement, cognitive demand and memory, the diagnosed insights and recommendations, and an automated workflow that translates those recommendations into concrete generation and curation steps.

The closed loop

What separates this from a general-purpose generative tool is that the output goes back into the predictive model. Variants are not generated and displayed — they are generated, measured again by the same engine that evaluated the original, and only what projects an improvement survives.

1. 01

   Translation

   The process starts from the written recommendations, which are grounded in the predictive metrics. Those recommendations are converted into a structured format that image models can interpret consistently.

2. 02

   Generation

   Multiple visual versions are produced from those directives. Each one is automatically checked to confirm that a visible change was made and that it reflects the intended recommendation.

3. 03

   Scoring

   The remaining versions are evaluated with the prediction engine. The original objectives and areas of interest are applied and a new Databrain Impact Score is calculated. Versions that do not meet the criteria are discarded.

4. 04

   Curation

   From the set of versions that do show a projected improvement, a selection of the strongest and most visually distinct examples is shown.

Each run produces a curated set of examples that shortens the distance between diagnosis and decision: instead of a written suggestion to interpret, the practitioner gets concrete alternatives with their projected impact already calculated.

Scientific References

Bagdziunaite, D., Nasan, K., Clement, J. & Ramsøy, T.Z. (2014). An added value of neuroscientific tools to understand consumers in-store behaviour. European Marketing Academy Conference, June 8.

Bundesen, C. (1999). A new look at functions of attention. Visual Cognition, 6, 89-91.

Bundesen, C., Habekost, T. & Kyllingsbæk, S. (2005). A neural theory of visual attention: bridging cognition and neurophysiology. Psychological Review, 112(2), 291-328.

Christensen, M. S., Ramsøy, T. Z., Lund, T. E., Madsen, K. H., & Rowe, J. B. (2006). An fMRI study of the neural correlates of graded visual perception. NeuroImage, 31(4), 1711-25.

Harel, J., Koch, C., & Perona, P. (2006). Graph-based visual saliency. Proceedings of Neural Information Processing Systems (NIPS).

Hou, X., Harel, J., & Koch, C. (2011). Image signature: Highlighting sparse salient regions. IEEE Transactions on Pattern Analysis and Machine Intelligence.

Itti, L. & Koch, C. (2000). A saliency-based search mechanism for overt and covert shifts of visual attention. Vision Research, 40(10-12), 1489-1506.

Kyllingsbæk, S., Valla, C., Vanrie, J. & Bundesen, C. (2007). Effects of spatial separation between stimuli in whole report from brief visual displays. Perception and Psychophysics, 69(6), 1040-1050.

Kyllingsbæk, S., Markussen, B. & Bundesen, C. (2012). Testing a poisson counter model for visual identification of briefly presented, mutually confusable single stimuli in pure accuracy tasks. Journal of Experimental Psychology: Human Perception and Performance, 38(3), 628-642.

Milosavljevic, M., & Cerf, M. (2008). First attention then intention. International Journal of Advertising, 27(3), 381-398.

Milosavljevic, M., Navalpakkam, V., Koch, C., & Rangel, A. (2012). Relative visual saliency differences induce sizable bias in consumer choice. Journal of Consumer Psychology, 22(1), 67-74.

Plassmann, H., Ramsøy, T. Z., & Milosavljevic, M. (2012). Branding the brain: A critical review and outlook. Journal of Consumer Psychology, 22(1), 18-36.

Ramsøy, T. Z. (2015). Introduction to neuromarketing & consumer neuroscience. Neurons Inc Publisher.

Ramsøy, T. Z. (2019). Building a Foundation for Neuromarketing And Consumer Neuroscience Research. Journal of Advertising Research, 59(3), 281-294.

Ramsøy, T. Z., Jacobsen, C., Friis-Olivarius, M., Bagdziunaite, D., & Skov, M. (2017). Predictive value of body posture and pupil dilation in assessing consumer preference and choice. Journal of Neuroscience, Psychology, and Economics, 10(2-3), 95-110.

Ramsøy, T. Z., & Overgaard, M. (2004). Introspection and subliminal perception. Phenomenology and the Cognitive Sciences, 3, 1-23.

Ramsøy, T. Z., & Skov, M. (2014). Brand preference affects the threshold for perceptual awareness. Journal of Consumer Behaviour, 13(1), 1-8.

Trabulsi, J., Norouzi, K., Suurmets, S., Storm, M., Nieuwenhuis, I., & Ramsøy, T.Z. (2020). Optimal gaze filters for eye- tracking on small screens. Frontiers in Neuroscience.