Eight years in, the say-do gap is no longer a hypothesis we are testing. It is the most reliable predictor of consumer response we measure across multiple use cases.

When we started building what is now Decode by Entropik, our Unified AI Human Insights Platform, the bet was that consumers do not always articulate what they actually feel. Sometimes they cannot. Sometimes they will not. And in many of the moments that matter most for a brand, whether it is the first three to five seconds of an ad, the friction point in an onboarding flow, the shelf decision in a store, or the CTA button on a landing page, the verbal answer and the user click response often arrive after the decision has already been made by the body.

Across more than 150 global brands and a panel ecosystem of over 103 million consumers, that bet has held strongest in three areas. In creative testing, attention drop off and emotional engagement predict campaign performance more consistently than recall scores. In shopper research, what a consumer looks at on a shelf rarely matches what they say drove the purchase. And in qualitative interviews, voice cues and facial emotions, whether it is hesitation, a shift in sentiment, or the pause before a polite answer, change the interpretation of the words themselves.

What has changed in eight years is not the premise. It is what brands now expect from it. Static reports delivered weeks after a study are no longer the benchmark. The expectation today is continuous, multimodal evidence that flows directly into product, marketing, and CX decisions. That is the platform we have spent the last two years rebuilding for.

Let me try to articulate this carefully because the question gets at something we think about all the time, and it is also something our customers ask us frequently.

Different modalities are validated against different ground truths. Our eye tracking is benchmarked against other webcam-based eye tracking solutions and hardware-based eye trackers, the kind used in university labs, and runs at around 96% accuracy on webcam and 95% on mobile camera. Facial coding is benchmarked against expert FACS annotation across diverse demographics and lighting conditions, and operates in the early 90s to mid 90s range depending on study conditions. Voice tonality-based emotion measurement is validated against human-rated sentiment and prosodic features. So when a single accuracy figure gets cited, it is usually referring to one of these modalities, and that is exactly where the industry needs to be more precise.

What that accuracy means in a real study is far more important than the headline number. We are not asking a brand to make a $10 million media decision based on one signal hitting 95%. We are giving them three signals, attention, emotion, and voice, that triangulate against each other and against what the consumer actually says.

If facial coding picks up confusion at second 14 of an ad, eye tracking shows attention dropping at the same moment, and the voice in the post stimulus interview hesitates when asked to recall the brand, that is not just 95% confidence. That is multiple independent signals converging, and it is far more reliable than any single number suggests.

The cost of misreading any single error is exactly why we built the platform to be multimodal from the start. A single signal system has to be right in isolation. A multimodal system can absorb noise in any one channel without weakening the conclusion.

The clearest business value shows up in use cases where the cost of getting a decision wrong is both high and visible.

Pre-launch creative testing is the strongest example. Brands often produce multiple versions of an ad, and choosing the wrong one for a large media buy is expensive. We measure attention, emotional engagement, branding recall, and drop-off moments to identify which creative is most likely to perform and which scenes need refinement before launch. In many cases, attention curves have revealed issues that traditional copy testing missed entirely.

User research is the second major area. Onboarding flows, checkout journeys, and prototype testing are all measurable revenue opportunities. If users are dropping off within the first few screens of a digital experience, webcam-based eye tracking and facial coding can pinpoint exactly where the friction is happening. Catching those issues before development or launch creates a significant ROI advantage.

Packaging and shelf research is another strong use case. What consumers say they noticed on a shelf and what they actually looked at are often very different. For CPG brands managing crowded categories, that difference directly impacts purchase decisions.

Qualitative interviews, whether moderated or AI-moderated, are equally important. Transcripts capture what was said. The platform captures what was felt while it was being said. Combining those two layers changes how teams interpret consumer feedback.

What surprised us most was not which industries adopted this technology, but which functions did. We expected research teams to lead adoption. Instead, product managers, growth marketers, and brand teams are now some of the heaviest users, especially as AI-generated creative and rapid experimentation become more common.

The signal moved from being a research output to an operational input much faster than we expected.

Building Entropik has taught us that AI delivers real value when it helps brands make faster, clearer, and more confident decisions, not when it simply generates more data.

In consumer research, the biggest challenge has never been a lack of information. It has been the ability to understand meaningful human behavior at scale. AI becomes valuable when it can reduce research timelines, identify patterns humans may miss, improve consistency, and help teams act on insights faster.

Where AI creates the illusion of value is when it prioritizes automation without context or interpretation. Just because a system can generate dashboards, summaries, or large volumes of outputs does not automatically make the insight useful. Consumer understanding still requires human judgment, business context, and the ability to ask the right questions.

That is also why we strongly believe in multimodal intelligence and human-AI collaboration rather than AI operating in isolation. The goal is not to replace human understanding, but to strengthen and scale it.

Yes, definitely. While adoption has grown significantly across industries like CPG, retail, media, entertainment, and BFSI, we still believe emotion AI remains underutilized across several industries and functions. Healthcare and education are at the top of that list, and for good reason.

Most early adoption happened in industries where consumer interaction and business impact are easier to measure, especially in advertising, shopper behavior, customer experience, and product engagement. That is why sectors like B2C, retail, and BFSI became some of the earliest use cases for us.

Healthcare is one of the most important untapped areas. Patient trust, treatment comprehension, digital health onboarding, and teleconsultations are all deeply emotional experiences. The challenge is not the technology itself. It is consent, regulation, and clinical responsibility, and the bar should absolutely be high. We are beginning to see early adoption in patient communication and pharma messaging, but the opportunity ahead is still massive.

Education is another major area. Online learning platforms generate enormous amounts of behavioral data, yet most still rely on completion rates and quiz scores to measure engagement. Whether a learner is truly understanding a concept or simply staying logged in while disengaged is exactly the kind of gap multimodal signals can help uncover.

The functional gaps are equally interesting. Customer service teams handle thousands of emotional interactions every day, but rarely analyze them meaningfully. SaaS companies optimize onboarding funnels constantly, yet very few measure the emotional state of a user when they encounter friction for the first time.

The next phase of adoption will be about behavioral intelligence becoming an embedded layer across product, marketing, and customer experience, rather than something used only for periodic research studies.

Richer data only leads to better decisions if it leads to clearer action. That is the real test, and it is not automatic.

NPS became widely adopted because it is simple. It gives teams a single number that is easy to track and report. But it tells you very little about why a creative is failing, where an onboarding flow is losing users, or which moment in an ad is causing disengagement. It answers one specific question well, but not necessarily the questions product and marketing teams need to act on.

Where emotion AI can create paralysis is when platforms surface raw signals like gaze plots, emotion curves, and attention heatmaps without translating them into decisions. That is a real failure mode, and we saw it ourselves in some of our earlier products.

The platform has to convert behavioral signals into actionable outcomes. If attention drops at second 14 of an ad, that is the scene to rework. If emotional engagement spikes during the brand reveal but weakens during the value proposition, that is a messaging problem. If hesitation increases when pricing is discussed, that may indicate a comprehension issue rather than a pricing issue itself.

These are not NPS style insights. They are operational decisions that teams can act on immediately.

The shift happening right now is not simply from less data to more data. It is from observation to action, and that is where richer behavioral data either proves its value or becomes noise.

I want to answer this with some humility because the question points to a very real limitation.

The platform is most predictive of the aspects that the creative itself actually controls. Attention and retention in the first few seconds of an ad. Emotional engagement during the brand moment. Branding visibility and recall after the stimulus ends. These are signals we measure before launch, often weeks before media spend goes live, and we have seen strong directional alignment with post-launch performance metrics, especially view-through rates and aided recall.

Where prediction becomes less reliable is when the outcome is driven by factors outside the creative itself. Media spend in a competitive category. Timing of a competitor campaign. A cultural moment that changes how the message is interpreted. Distribution decisions that place the ad in front of the wrong audience. We have seen campaigns that tested strongly before launch but underperformed later, and in most cases, the reason came down to variables the creative could not control.

That is why we are careful about how we position predictive testing. It is a decision support layer, not a replacement for brand judgment or media strategy. A stronger attention score between two creative variants is a useful signal, but it is not a guarantee of campaign success.

In practice, the most valuable use of predictive testing is often not choosing winners. It is identifying weak creative early enough to avoid expensive mistakes. That pattern has repeated itself consistently across categories.

That distinction is exactly the right one to make, and where the chain breaks is often more interesting than where it works.

The translation from signal to recommendation happens through pattern recognition. When the platform detects attention dropping at a specific scene in an ad, it can connect that drop to the visuals, audio cues, and script at that exact moment. Across enough campaigns and categories, clear patterns begin to emerge. Close attention with low recall often points to a branding issue rather than a creative issue. Strong emotional engagement in the first few seconds tends to correlate with higher completion rates. Voice hesitation during post-stimulus responses usually signals a comprehension gap. These are the kinds of patterns a system can surface quickly enough for teams to act on immediately.

Where the chain breaks is equally important. First is intent. The platform can detect that attention increased on a specific element, but it cannot always determine whether that attention came from attraction, confusion, or cognitive overload. We have seen cases where high attention was initially interpreted as positive engagement, when consumers were actually trying to make sense of what they were seeing.

Second is cultural context. Expressions and behavioral cues do not always translate uniformly across markets, which is why local calibration matters.

Third is strategy. A system can identify that an ad is underperforming emotionally, but it cannot decide whether the campaign still makes strategic sense within a broader brand narrative.

The right way to think about this is augmentation, not autonomy. The platform handles pattern recognition at a scale and speed humans cannot. Strategic interpretation, brand judgment, and the final decision on what to do next remain fundamentally human.

The honest answer is that off-the-shelf global models do not perform reliably on Indian consumers, and anyone claiming otherwise is overstating the technology.

The bias shows up in measurable ways. Models trained primarily on Western datasets can miss or misread emotional cues that are expressed differently across South Asian demographics. Facial coding accuracy drops, certain emotional categories get misclassified, and confidence intervals widen. That is a documented industry problem, not a theoretical one.

India is also not a single consumer profile. What works in a Tier 1 metro on a high-end smartphone with stable lighting may not work in a Tier 3 town on a low-bandwidth device with inconsistent connectivity. Most published benchmarks do not account for that level of variance.

Our approach has been to treat localization as an ongoing engineering discipline rather than a one-time training exercise. That means calibrating models across Indian demographics, device conditions, lighting environments, and language contexts. It also means designing panels that represent both urban and non-urban consumers based on the research objective, not simply based on who is easiest to recruit.

Representation in India goes beyond facial diversity. It includes language comfort during interviews, digital familiarity with the testing environment, and even behavioral differences between webcam-based studies and real-world shopping environments. All of those factors need to be designed intentionally.

India is one of the most important markets for this category because of its scale and diversity. But that opportunity only matters if the technology is built responsibly for the market, not adapted from Western models and expected to work by default.

Oh, wow. That is a very important conversation, and honestly, one that the industry needs to take very seriously as AI capabilities continue to evolve.

At Entropik, we have always believed that trust, transparency, and consent have to be foundational to emotion AI. There is a very clear difference between using AI to responsibly understand consumer experiences and using technology in ways that feel invasive or manipulative.

For us, the line is crossed when transparency, consent, and user control are removed from the process. Consumers should always know when data is being collected, why it is being collected, and how it will be used. Ethical AI cannot function without awareness, accountability, and strong governance frameworks behind it.

Our approach has always been research-led and consent-led. Participants knowingly opt into studies, and the purpose is to help brands better understand engagement, usability, emotional resonance, and friction points, not to exploit individuals or influence them unfairly.

Because we work with several large global brands across luxury, CPG, retail, and other industries, confidentiality and data protection are extremely critical for us. In many cases, we intentionally do not disclose certain brand engagements publicly because consumer response data, behavioral insights, and market signals can reveal strategic direction, campaign effectiveness, or emerging trends that competitors could potentially interpret or act upon. When you are operating with respondent panels and large-scale consumer intelligence, protecting that ecosystem becomes just as important as generating the insights themselves.

As emotion AI becomes more capable, I think the responsibility on both platforms and brands becomes even greater. It is not just about what technology can do, but what it should do. The brands that will succeed long term are the ones that use these technologies responsibly, transparently, and with the genuine intent of creating better and more human-centered customer experiences.

Ultimately, AI should help brands understand people better, not make people feel surveilled.

The bottleneck has already shifted. Content creation is no longer the constraint. Content validation is.

Marketing teams can now generate dozens of creative variations in minutes using AI. The real challenge is identifying which of those creatives are actually worth investing media spend behind, which need refinement, and which should never go live at all. Traditional research systems were not built for that speed. A four-week study cycle cannot support content that may only stay relevant for two weeks.

That is the environment Decode by Entropik was built for. Multimodal validation combining facial coding, eye tracking, voice tonality measurement, AI-moderated interviews, and behavioral intelligence, operating at the same speed as content is being created. Brands can test new creative variants with targeted audiences from a panel ecosystem of over 100 million consumers and receive actionable insights in hours instead of weeks.

The bigger shift is not just faster research. It is a fundamentally different relationship between content and consumer understanding. Brands will move from testing a handful of creative ideas to validating large volumes of AI-generated variations continuously.

As that happens, the role of research changes as well. It stops being a gatekeeping function and becomes an operational layer that helps teams make decisions across the entire content lifecycle.

Traditional research is not disappearing. It is being asked to do something it was never designed for, which is to keep up with the speed of modern decision-making.

Surveys, interviews, and focus groups still matter because they capture what consumers consciously think, their intent, reasoning, and language. That layer remains essential. What has changed is the understanding that conscious feedback alone is no longer enough, because many consumer decisions happen before they are fully articulated. Combining stated feedback with implicit behavioral signals like attention, emotion, and voice creates a much more complete picture.

The larger shift is around speed and accessibility. Research was once a quarterly output, usually delivered as a report or tracking study. Today, insights are expected to flow continuously into product, marketing, and customer experience decisions. The audience has also expanded beyond research teams to include product managers, growth marketers, brand teams, and CX leaders who need answers in days, not quarters.

What survives this transition is the part of traditional research that has always mattered most: asking the right questions, designing studies rigorously, and interpreting results with context. What AI changes is the friction between asking the question and getting the answer.

The future is not traditional research versus AI-driven research. It is the convergence of human understanding, behavioral intelligence, and continuous evidence, operating together at the speed modern businesses now require.

The organizations that will win over the next five years are the ones that learn to understand what consumers do not explicitly say.

Today, most enterprise decision-making still relies heavily on stated feedback such as surveys, NPS, reviews, social listening, and focus groups. That data is valuable, but it is incomplete. It captures what consumers articulate, not always what they feel, notice, hesitate over, or remember. The brands that close that gap will understand their customers very differently from those that do not.

The real differentiator will not be who adopted AI first. In a few years, every consumer business will use AI in some form of research. The difference will come from how they use it. Some companies will simply use AI to run existing research processes faster. Others will use it to build entirely new forms of consumer understanding through continuous, multimodal insight embedded directly into product, marketing, and customer experience decisions.

Those companies will build stronger products, more relevant communication, and deeper customer relationships, not because they have better AI alone, but because they recognized that the larger shift is about understanding human behavior more completely.

For us, that has always been the core belief behind the company. Five years from now, I think that shift will feel obvious.