AI Expertise for Consumer Insights
Welcome to AigoraCast. Conversations with industry experts on how new technologies are impacting sensory and consumer science.
Visit your favorite platform and remember to subscribe:
Apple Podcasts
Spotify
Please leave a positive review if you like what you hear!
Guest: Rasto Ivanic (Co-Founder & CEO, GroupSolver)
Host: Dr. John Ennis (CEO, Aigora)
Host Introduction:
Hi, I'm Dr. John Ennis, CEO at Aigora.
In this episode, I had the pleasure of speaking with Rasto Ivanic, Co-Founder and CEO of GroupSolver.
Rasto and I dive into where AI truly belongs in consumer research—and why the answers still have to come from real people. We explore his journey as a founder, why he halted product development for six months to tackle rampant survey fraud, and why synthetic panels and digital twins cannot replace genuine human insight.
I hope you enjoy this conversation as much as I did. Be sure to subscribe to AigoraCast to catch future discussions at the intersection of AI, consumer science, and sensory research!
Rasto Ivanic is the Co-Founder and CEO of GroupSolver, a market research technology company founded in 2012. GroupSolver provides a dynamic survey platform that functions like a natural conversation: respondents answer open-ended questions in their own words, while real-time machine learning clusters those responses and validates emerging ideas with subsequent participants. In this model, key themes emerge organically from the crowd rather than from a researcher’s pre-packaged hypotheses.
Rasto traces the foundation of GroupSolver back to his first week at McKinsey & Company, where initial training emphasized asking the right questions before attempting to supply answers. His philosophy remains simple: "Ask a question, ask why, ask what, and then shut up and listen."
He holds a firm stance on artificial intelligence and data integrity: in GroupSolver’s platform, AI serves as an analytical organizer rather than an information creator. Following an alarming study where survey responses showed clear signs of fraud, Rasto halted engineering feature development for six months to build proprietary, real-time data verification tools. Today, GroupSolver publishes monthly transparency reports on data quality and routinely filters out over 60% of fraudulent or low-attention respondents.
Rasto is a trained economist holding an MBA and a PhD in Agricultural Economics from Purdue University. Prior to founding GroupSolver, he served as a management consultant at McKinsey & Company and led business development and strategy at Mendel Biotechnology.
Dr. John Ennis: Welcome back, everyone, to another episode of AigoraCast. Today, I’m very happy to have my friend Rasto Ivanic on the show. Rasto is the Co-Founder and CEO of GroupSolver, a market research technology company that has built an intelligent, AI-powered platform to help businesses answer their burning why, how, and what questions. His mission is to build a platform for human truth in consumer research, predicting consumer behavior in an era of AI-generated data.
Before founding GroupSolver, Rasto served as a strategy consultant with McKinsey & Company and later led business development at Mendel Biotechnology. Throughout his career, he has guided companies in making strategic decisions regarding new business development, market opportunity pursuit, and complex partnership building. Rasto is a trained economist holding an MBA and a PhD in agricultural economics from Purdue University.
Rasto, welcome to the show!
Rasto Ivanic: Thank you, John. Glad to be here.
Dr. John Ennis: We share a mutual client who recommended that we connect, and it has been an absolute pleasure getting to know you. You have been on quite a journey with many interesting twists and turns. How did you go from finishing your PhD in economics to where you are today?
Rasto Ivanic: When I was studying, I really thought I would end up being an academic economist. But life throws curveballs at you. When you're young and feel like your whole life is in front of you, you think, “Okay, let's see where this left turn takes me.”
My left turn took me to McKinsey. Fresh out of grad school as an international student with no money, that sounded like a great way to start life and get ahead.
Where I am today with GroupSolver really traces back to my first week at McKinsey. They flew us to New York for orientation, and the very first training session was about how to ask questions. A typical consultant tends to immediately start giving answers because that’s what they feel they are paid for. But the training taught us: take a step back. Ask the right questions first. Ask your client; ask the market. Once you have that information, you can give them a good answer, not just an answer.
I learned the value of open-ended inquiry: ask a question, ask why, ask what, and then shut up and listen.
After three years in consulting and a stint leading strategy at Mendel Biotechnology, I never stopped believing that to solve a problem, you first need to ask the right questions to get the right data. Eventually, I realized I was young enough to take a shot at startup life, but experienced enough not to make foolish rookie mistakes.
My wife is a market researcher, so I saw the surveys she ran and had conducted many myself. Twelve years ago—long before AI was on corporate agendas—I began building technology with rudimentary models to blend crowdsourced human truth with machine intelligence. The goal was to pull out the nuggets of what people are saying that actually matter.
Dr. John Ennis: Around 2020, I did an interview with Professor Betina Piqueras-Fiszman at Wageningen about intelligent surveys. Even before modern generative models, it felt inevitable that dynamic surveys were the future. GroupSolver began in NLP and traditional text analysis. How has your product evolved over the past twelve years?
Rasto Ivanic: The foundational architecture that connects the dots hasn't changed; the underlying models that make it perform have improved dramatically.
Take a concrete example: ready-to-drink chilled coffee in grocery coolers. A brand wants to know: What is the job that ready-to-drink chilled coffee does? For what reason do you buy it?
From the beginning, our system has opened with an unaided open-ended question. Respondents arrive asynchronously and write whatever is on their mind.
Suppose Respondent A says: "I drink ready-to-drink coffee because I need a quick caffeine hit when I'm driving, but I sometimes find it too sweet."
There are three distinct ideas there:
A decade ago, we had linguists on staff building custom rule-based splitters using early tools like BERT. Language models struggled tremendously back then to cleanly isolate distinct semantic clauses. Today, modern NLP and LLMs handle that syntactic segmentation effortlessly.
Once an answer is split, our system processes the statements without generating any artificial data. AI is strictly an organizing tool.
Now suppose you come along as Respondent B and say: "It’s a treat after a good workout when I deserve something sweet."
Our system evaluates: Has anyone mentioned this being a "treat" yet? If not, it creates a new semantic cluster. At that moment, we don't know if "a treat" is an idiosyncratic response or a broad trend.
When Respondent C enters, after they give their own open-ended response, the system presents your statement: "Other people mentioned they drink this as a treat. Does that apply to you?"
If Respondent C says no, the idea stays quiet. If multiple subsequent respondents say, "Yes, that's true for me too," the treat concept builds statistical momentum and earns its place in the idea pool.
Because people express ideas differently—some say "dessert," some say "sweet reward"—the system groups them semantically, identifies the most representative statement, and provides statistical validation. Because ideas are tested across respondents in real time, researchers can cross-tabulate open themes against demographics or Net Promoter Scores (NPS) just like quantitative data. It bridges qualitative depth with quantitative rigor.
Dr. John Ennis: When you onboard a new client, do you encourage them to seed the survey space with their pre-existing hypotheses, or do you start with a blank slate?
Rasto Ivanic: Ninety percent of our clients start without pre-seeding anything.
Our platform allows clients to seed hypotheses if they have specific themes they want to test. But almost every time a client insists on pre-seeding ten statements, those exact themes emerge naturally from respondents anyway.
More importantly, when a pre-seeded client hypothesis falls flat because real consumers don’t care about it, clients accept the finding because the open-ended crowd didn't validate it. If an idea is genuinely important to consumers, it surfaces on its own. I am a big believer in: Ask the question, and let the data tell you where to go.
Rasto Ivanic: There is another major area that has changed dramatically: data quality.
When I started in consumer insights, panel recruitment was straightforward. You worked with trusted panel partners and used basic attention checks: removing straight-liners or participants who failed simple traps like "Select option D if you are paying attention." You’d scrub 10% of the sample, write it off as noise, and move on.
Over the last several years, the research industry underwent heavy private equity consolidation. Programmatic sampling took over. At the same time, survey completion became a lucrative side business for click farms and bad actors in low-wage markets. If someone can complete dozens of surveys a day by pretending to be targeted buyer personas, they make a substantial living.
A few years ago, I ran a study where the numbers simply didn't make sense. I began investigating raw telemetry. I couldn't sleep for weeks because I realized the industry's house was on fire, and companies were busy watering the lawn.
I halted all feature engineering at GroupSolver for six months. No new features, no platform upgrades—nothing. I told my team: until we can guarantee that our data comes from real human beings paying attention, all the advanced analytics, AI, and visualizations in the world are just junk in, junk out.
Because we own our codebase, we built forensic fraud detection directly into the survey engine: tracking mouse dynamics, keystroke rhythms, timing anomalies, and cross-checking behavioral cues.
We now publish monthly data quality transparency reports. In any given study, we routinely reject and eliminate over 60% of incoming respondents.
Dr. John Ennis: That matches what we see. My wife is an academic psychologist who runs online studies, and respondent fraud is an enormous problem. Catching it requires deep forensic rigor.
Dr. John Ennis: I can relate to the entrepreneurial grind. I spent three years running an AI startup where we raised $3 million, built great technology, but struggled to find the exact initial product-market fit before shutting it down. What has your founder journey been like over the past twelve years?
Rasto Ivanic: Last summer, I took time on Mondays and Fridays to write a book about this journey—partly as an outlet for my own mental health, and partly for young founders considering jumping into these choppy waters.
Being an entrepreneur has been deeply rewarding: I will never look back in old age and wonder what it would have been like to build something of my own. I’ve mentored dozens of people and launched young careers.
At the same time, it takes a massive toll. At our peak, we were over 50 people, well-capitalized, working with clients like Google, Amazon, Microsoft, and Adidas. But we hit a fork in the road: investors wanted pure SaaS subscription metrics because that’s what drove frothy valuations.
I told them our enterprise clients valued our service and results; they didn't want self-service SaaS seats where they had to do everything themselves. But we bowed to investor pressure and poured millions into building a DIY SaaS platform—competing directly with commoditized tools.
When macroeconomic headwinds hit, clients froze budgets and reduced headcounts. We lost more than half of our revenue in a few months. Our venture backers told us: "You ran the experiment, it didn't pan out, you’re on your own."
Almost overnight, I had to lay off two-thirds of the company. To keep our core engineers and protect our team, I went without a salary for two years.
That level of stress changes you. Early on, you wake up at 5:00 AM excited to solve problems without needing coffee. Under existential stress, you wake up at 4:00 AM with a knot in your stomach because your nervous system won't let you sleep. That is the hidden price of entrepreneurship.
Dr. John Ennis: My father has run the Institute for Perception for over 35 years, and he always says you have to have an iron stomach for this work. At big corporations, you have the illusion of stability, but you're always one reorg away from your role being cut. In your own company, you carry the stress, but you also own the freedom and the successes.
Rasto Ivanic: There's widespread speculation that AI agents and synthetic personas will render human consumer research obsolete. Some clients even say, "I don't need to interview consumers anymore; I'll just query a digital twin."
That thinking is fundamentally flawed. Even if personal AI agents end up doing our online grocery shopping, humans still decide whether they like the taste of the coffee. You have to instruct that agent on what you enjoy and what you dislike.
AI models are trained on historical corpora. They are bound within an $N$-dimensional data space of past text. But human culture is unpredictable. Nobody's historical dataset predicted the cultural shockwave when the Beatles landed in America in 1964, or when Nirvana released Nevermind in 1991, or when sudden geopolitical conflicts erupt overnight.
If something happens outside the training corpus, an LLM has zero basis for predicting how human sentiment will react. Synthetic data is a helpful tool for exploratory bootstrapping or connecting disparate studies, but it is not a primary source of ground truth.
Dr. John Ennis: I completely agree. Machine intelligence is extraordinary at synthesizing large amounts of text, but humans own their subjective lived experiences. A language model contains statistical weights; it lacks a personal point of reference to assign genuine emotional weight or personal meaning. If you want to understand human experience, you must talk to human beings.
Rasto Ivanic: Exactly. I am on "Team AI"—we've built with it since 2015. But working closely with AI teaches you its boundaries: what to trust, and what to verify.
Because of fraud, the total volume of careless, cheap surveys is going to decline, while the value of verified human truth is going to rise. When brands realize they must pay for authenticated human insights, the science of our industry will only get stronger.
Dr. John Ennis: Rasto, this has been an incredible conversation. We could easily keep going, but as we wrap up, what is the best way for listeners and potential partners to connect with you?
Rasto Ivanic: I pride myself on maintaining a zero-junk inbox. If you send me a note, I will read it.
The easiest places to reach me are:
Dr. John Ennis: Fantastic. Thank you so much for joining us, Rasto.
Rasto Ivanic: Thank you, John. Great conversation!
Aigora is a contributor to the Aigora blog, sharing insights on AI-powered sensory science and product development.