Sebastian (00:02)

You are listening to the Insightful Connections podcast. Our guest today is Andrej Zov, a cofounder of unSurvey, founded in 2024. unSurvey is a platform for running qualitative interviews using AI that helps researchers speak to far more people than ever before. Unlike other tools in the AI qual at scale space, unSurvey can conduct rich, humanlike voice to voice conversations thanks to its conversational flow builder, which allows clients to steer discussions in complex ways and capture more nuanced insights.

Prior to founding unSurvey, Andrej worked at BlackRock as a researcher, and before that he completed his PhD in Natural Language Processing at the University of London. Andrej, thank you so much for being on the podcast today.

Andrej (00:41)

Thanks, Sebastian. My pleasure.

Sebastian (00:44)

Hey everyone, Seb here

Sebastian (01:29)

Now I like to ask everyone who comes on the show about their professional journey and how it led them into market research. From what I know, part of that path for you happened at BlackRock. Can you talk about how you arrived here, what influenced you, and how it explains where you are now and where you are headed?

Andrej (02:19)

That is a great question. I have always tried to build an eclectic career. I think life is too short to climb a long ladder in a big corporation, though I did spend time in one.

Here is the short version. I studied computer science and went straight into a PhD. At first, I explored probability theory, but there was a big wave of work in natural language processing back then. We often just call it AI now, although at the time it was nowhere near what it can do today.

I did my PhD in that field, focusing on entity recognition and entity linking. Basically, that means highlighting important words in long pieces of text. Not very groundbreaking in hindsight, and I joke that my PhD is now obsolete because we have moved on to modern AI. Still, it let me see the beginnings of neural network approaches to these problems.

Then I joined a small startup in London called DigitalGenius around 2015, which was trying to automate customer service. At the time, that was almost impossible, though they have since made huge strides. After that, I moved to BlackRock, where I worked on natural language processing. One of my projects was reading all the news every day using algorithms to extract information for investment decisions. It was pretty rudimentary compared to what is possible now, and the limitations were clear.

When the pandemic hit, I got bored. My cofounder and I spent two weeks building a demo and decided to apply to Y Combinator. They told us our idea was terrible but let us join anyway. We started out working on pricing, which ties into market research because pricing optimization relies on surveying people in various ways. We tried automating pricing by showing different prices to different groups, but quickly realized the industry heavily depends on direct market research. Eventually, we figured we could do that at scale using more advanced AI. That is how we got here.

Sebastian (05:42)

Why did Y Combinator say your idea was terrible, and why did they invite you despite that?

Andrej (05:50)

I am a little embarrassed, but we applied with a name inspired by the TV show Silicon Valley. In that show, the team builds a compression algorithm and calls it middle out. We called our idea Middle Out, thinking startup people would find it funny.

The concept was a dislike of middle management and replacing it with AI, which would keep leadership and frontline employees aligned. This was during the height of the pandemic, so future of work ideas were everywhere. YC called it a COVID idea and said it was terrible, but they still liked us as a founding team. So in we went.

Sebastian (07:01)

What do you think drew them to you as a team?

Andrej (07:04)

In general, investors look for builders, people with a track record of quickly creating something real. They also want to see curiosity, speed, and the ability to iterate. We threw something together in two weeks, showed it to a few early clients, and got a little traction. We checked enough of the right boxes.

Sebastian (07:51)

Let us talk about AI’s role in market research and how it is collapsing the boundaries between qualitative and quantitative methods. Can you expand on that?

Andrej (08:19)

Sure. In the past, there has been a firm line between qual and quant. You could scale quant easily by sending a survey to thousands of respondents, but it was hard to scale qual. One thing people often do is run a big quant survey, then follow up with a small qual study to get richer insights. You end up with descriptive quotes that give you a better picture of your sample.

With modern AI, you can replace the human interviewer in some contexts and scale up qualitative work. It is also easier to pull quantitative insights out of open ended responses, thanks to more advanced natural language processing. So the usual barrier between qual and quant dissolves, since you can scale qual and also turn that data into more structured findings.

Sebastian (10:09)

What is the main benefit in collecting quantitative insights from open ended data, or from scaling up qualitative methods?

Andrej (10:19)

I would say there are two big advantages. The first is cost. I recently talked with someone who wanted to interview 1200 people in a representative sample but only had the budget to do a 30 person study if they used a traditional approach. With an AI platform, you can run 1200 interviews for a cost closer to just running software.

Second, there is the respondent experience. Surveys have not evolved much for quite some time. Respondents click radio buttons and rate how strongly they agree or disagree. Conversational AI might allow them to speak naturally, with the AI playing the role of a skilled qual moderator who asks clarifying or probing questions. That is a different and possibly more comfortable experience for many participants.

Sebastian (21:06)

You mentioned AI might transform the entire experience of research across three areas. You listed the respondent experience, the builder experience, and the analysis experience. Could you walk us through each of these?

Andrej (22:00)

Of course. First, the respondent experience. We are focused on voice AI, but there are plenty of other ways this could happen. In real human conversations, there is a lot of subtlety. People avoid interrupting each other, they pause, they show empathy, they rephrase questions, and so on. Skilled qual researchers spend years honing those techniques, and replicating them with AI is challenging, but we are getting closer every day.

On the builder side, AI is already being used to write questionnaires or brainstorm discussion guides. Even something as simple as ChatGPT can help with that cold start problem, where you do not know how to begin. You can also see a future where AI handles back and forth between stakeholders. For example, a client wants one thing, the agency suggests something else, and AI helps reconcile priorities.

Finally, the analysis side. Turning unstructured text into something you can measure used to require a team of specialists. Now large language models can do a lot of that very quickly, like tagging responses for sentiment or categorizing themes. Another frontier is synthetic data, which is a whole other story. Some are experimenting with using large language models to generate consumer insights. It might work for widely discussed topics, but it might not be great for something more niche. Still, it is an area to watch.

Sebastian (24:38)

Let us think five or ten years ahead. Where do you see AI taking us, especially in terms of our work?

Andrej (25:32)

One thing is certain. AI will become more human in small but meaningful ways, especially with voice. We are going to see better timing, better emotional cues, accents, laughter, and so on. The cost of AI will also drop. GPT4 today is not cheap, but imagine something that capable at a fraction of the price. That will unleash ideas we have not even considered yet.

Another interesting angle is that genuine human interaction could become more valuable. As AI takes over more routine interactions, people will learn to recognize the difference between an AI conversation and a real human one. Real human contact might be seen as something special, which could change how we do business or build relationships.

Looking at software development is a good way to predict the future. ChatGPT and similar tools have given a massive productivity boost to the average developer. Junior developers see less benefit because they are still learning the basics, while elite developers may not trust AI enough to write perfect code for them. That pattern could repeat in other fields. The average performer jumps ahead thanks to AI assistance, but top experts still shine, and the entry level gets squeezed.

In the end, we all have to keep pushing ourselves to become true experts, not just rely on AI to do everything for us. It is fine to use AI to speed things up, but we have to move beyond the average and keep learning. For me, it is exciting to work in a period where the pace of change is so high and where there are big opportunities to create something meaningful.

Sebastian (26:23)

My last question is about motivation. What keeps you motivated?

Andrej (27:26)

This might sound basic, but life is short and I see this as a seismic moment. Every generation has big shifts to deal with. Ours includes AI. Big changes shake up the status quo, which makes it easier to learn quickly and build something new. If you think of it the way BlackRock might, high volatility is where you find alpha. It is not just the money, but also the thrill of moving faster, learning faster, and shaping the future. That is what gets me out of bed in the morning.

Sebastian (29:10)

Andrej, thank you so much for your time today.

Andrej (29:10)

Thank you, Sebastian. It has been a pleasure.

‍

‍

‍