Here is a problem that market research has never quite solved, and mostly just agreed to ignore: people are unreliable narrators of their own minds. You put a product in front of someone, you ask them what they think, and they tell you something. The something is shaped by what they think you want to hear, by whether they had lunch, by a vague desire to seem like a reasonable person. It is data, technically. It is just not necessarily data about the product.
Entropik's founding bet is that you can route around this. Instead of asking a consumer to describe a feeling and then trusting the description, you measure the feeling directly - in the flicker of a facial muscle, in where the eyes land and for how long, in the tonal wobble of a voice. The pitch, in the words of founder and CEO Ranjan Kumar, is that "95% of decisions made by a consumer are emotional in nature." If that is true, then the survey has been measuring the wrong 5% for about a century.
"95% of decisions made by a consumer are emotional in nature. We are the 2nd largest repository of human emotion data globally."
- Ranjan Kumar, Founder & CEOWhere it started
The company did not spring from a McKinsey deck. It started around 2013 as a college project at IIT Kharagpur, where Ranjan Kumar was poking at emotion tech and cognitive science - the unglamorous question of whether a machine could look at a person and infer a state of mind. That question turned out to have a large commercial shadow, and in 2016 it became Entropik, headquartered in Bangalore.
The founding team pairs the science with the plumbing. Kumar runs the company. Lava Kumar leads product, bringing a background in affective computing - the academic name for machines that reason about human feeling. Bharat Singh Shekhawat handles engineering and the data-intensive systems that emotion measurement demands, because reading one face is a demo and reading a million faces across cultures and lighting conditions is an infrastructure problem.
Three signals
The technology rests on three modalities that Entropik has wrapped in 17 global patents: facial coding, which reads micro-expressions; eye tracking, which measures attention and gaze; and voice AI, which handles transcription, translation, and - the interesting part - tonality. Individually these are known techniques. Entropik's claim is in the stitching: combining all three into what it calls a Multi-Modal Emotion AI, so a smile that the face reports can be checked against where the eyes were actually looking and how the voice actually sounded.
The strategic asset underneath all this is boring and enormous: data. Entropik says it holds the second-largest repository of human emotion data in the world. In machine learning, the model is only as good as what it trained on, and whoever holds the largest labeled dataset of human expression has a moat that is genuinely hard to dig around. You cannot buy last year's faces.
What you actually do with it
All of this would be a lab curiosity if it did not do something a marketing team could expense. Entropik's answer is Decode, launched in 2023 as a DIY - do-it-yourself - unified insights platform. The idea is to collapse the traditional research pipeline. Old way: recruit respondents, schedule sessions, hire a moderator, run interviews, transcribe, code, analyze, wait several weeks, present. New way: run quantitative, qualitative, media, and shopper studies through one platform and get insight back in something closer to real time.
Speed, here, is not a vanity metric. If a brand can test a campaign, read the emotional response, and adjust before the media spend goes out, the research has moved from autopsy to steering wheel. Alongside Decode sits Qatalyst for user research and UX testing, and Affect Lab, the older consumer-research tool for putting creatives and product designs in front of measured emotion.
The bottleneck in research was never collecting the data. It was making sense of it fast enough to matter.
- The GenAI thesis, in one lineWhich brings us to 2024, when Entropik did the thing every software company did in 2024: it added generative AI. In its case the move is unusually on-point. The company launched AI-Generated Reports across Decode and Qatalyst - GenAI that mines the collected signals and turns them into presentation-ready reports with visualizations. It also introduced Mira, an AI moderator that can conduct research interviews without a human in the room. If the moderator's job is to ask consistent questions and probe follow-ups, an AI that never gets tired and never leads the witness is either a labor-saving device or a slightly unsettling glimpse of the future, depending on your seat.
The customers, and the caveat
The client list is the kind that makes a Series B deck write itself: P&G and Nestle on the consumer-goods side, ICICI and JP Morgan on the finance side, with 150-plus brands across CPG, healthcare, media, and retail. These are companies that spend real money on research and have real reasons to want it faster.
The honest caveat is that scaling a deep-tech category is hard, and public headcount trackers suggest Entropik went through a meaningful workforce reduction in 2025 - a reminder that being early to a category and being comfortably profitable in it are different achievements. Emotion AI is a genuinely novel product built on genuinely defensible patents. It is also a bet that the research industry will rewire itself around measurement rather than self-report, and that rewiring happens on the industry's schedule, not the startup's.
Still, the underlying observation is hard to argue with. If most decisions are emotional, and emotion is measurable, then someone is going to build the instrument that measures it at scale. Entropik got there early, patented the approach, and stacked up the data. Whether that becomes a large independent business or a widely licensed technology, the direction it is pointing - away from the survey, toward the signal - looks less like a fad than a correction.