Latest / The Collaborative Canvas Podcast / Why AI Can’t Replace Human Insight (Yet) | Everyone Has AI… So Why Do Only Few Stand Out?

Transcript

Why AI Can’t Replace Human Insight (Yet) | Everyone Has AI… So Why Do Only Few Stand Out? · The Collaborative Canvas Podcast

• speaker-0: So the AI is just analyzing the words without the context, without the emotion. speaker-1: Even in this case, know, we've been always talking about hallucinations, which AI does. speaker-0: It's very dangerous to use up general LLMs. Now, that 80 % I think is still missing. You're just a messenger then, really, aren't you? There's a danger of that. I'm not going to make that bulk home. And again, just to be really clear, I think it's a wonderful innovation. adds a lot. You need to be weird in 2016. True qualitative researchers know that speaker-1: It's when I'm using of AI is definitely needed at this point of time. Everybody needs money. Everybody wanted to make some money and speaker-0: Very, very hot topic. Very divided. That's a tricky one. mean, you have this sort of sixth sense when something's a fishy. speaker-1: Post-Chart GPT era, it has a different story. speaker-0: AI now, I think, is really shaping things up. Let's not lose the craft. In this age we're going through, you're dead. So... I don't think there's an easy answer to that question. speaker-1: What advice would you give? I'll... Because my boss is after my life. speaker-0: Wow, there you go. That's that's the only sort of ⁓ mission impossible. So sometimes that's that's the reality. Let's talk about specific core projects. So in theory, speaker-1: AI is everywhere in research today. We talk about adoption, tools, prompts, speed, almost obsessively. But adoption is no longer the interesting question. The real question is this. If everyone has access to the same AI, why do we so few researchers actually stand out? Today's conversation is about advantage, not AI as automation, but AI as multiplier of judgment, cultural understanding, and qualitative courage. My guest today brings a rare and deeply grounded perspective to this discussion. Craig. Craig is a UK national with over 25 years of experience in market research and insight and has been based in Asia since 2003. Over the years, he has held senior positions in organizations like GFK, Insights Asia and Mesh Experience before founding Fuel Asia in 2016. Through Fuel Asia, Craig has focused on something many researchers talk about. but very few truly practice, else bridging the culture. With a background spanning marketing, social anthropology, semantics and decades of hands-on work across Asian markets, his work has consistently helped Western clients understand the deeper human and cultural drivers behind behaviour in Asia. Fuel Asia is known for delivering qualitative work. that is localized, rigorous and culturally attuned, supported by a trusted network of best-in-class moderators, recruiters and translators, and personally led to ensure depth and relevance. More recently, Craig has been exploring how AI and research technology can support better qualitative research, not just faster output, but smarter thinking. In 2025, he co-founded the Asia Digital Insights Summit with Insights platform, the region. first dedicated research tech conference. So today, instead of asking whether AI belongs in a qualitative research, we are going to explore something more comfortable and more important. How do researchers turn AI from a leveling force into a real advantage? So welcome, Craig. speaker-0: Hi, Ankash. Great to be here. Thanks for inviting me. speaker-1: Thank you so much. I've been following your work over qualitative research and you've been seeing over a period of time, you qualitative research has been taking its own shape. two years down the line, I've been seeing that, know, synthetic data persona, synthetic data per se, or any usage of AI has been restricting a lot many quant business. And that human element of touch and feel is coming back again, which we used to see long back when, you know, the panels were not even introduced. People used to go and talk to the respondents and they used to get the right kind of conversation going and right kind of feedback comes along, which entails to the report that goes out to the clients. But then the inception of panel came on, the technology advances and then now AI and I see a lot of qualitative moderators getting into an AI mode where they're actually introducing AI moderators. I heard your presentation about how qualitative in the age of AI is actually taking shape. So how about our audience know what was the presentation about, what was the findings behind it, and then we take on to the next one. speaker-0: Yeah, so what did we talk about yesterday? Well, I did the presentation, as you know, with my longtime ⁓ collaborator, Dave McCotland, and we were interested in in exploring AI from a slightly different angle, because I think the buzz term in the last couple of years has been call it scale. this has really been about driving efficiency by using AI for for moderation and for analysis. ⁓ So the whole conversation is a bit about increasing efficiency. That in turn brings costs down, which in turn increases accessibility. So a lot of people are saying it's more accessible. ⁓ And actually, to me, this is a good thing. I'm certainly not against it. I think it's something great to have in our toolkit because it gives us more occasions when we can use Qual when previously we might have not been able to do because of budget reasons. We wanted to just look at it from a different perspective. How can we use AI to actually enhance our value as consultants? Because the minute we start focusing on efficiency, we're in this race to the bottom, and we might lose some of the qualities that made us good consultants. We wanted to flip that and ask this question. What's a different way to look at AI? We talked a lot about being about being weird. That was one of the terms that Dave likes. We need to be weird in 2026. Weird is the word of 2026. What we really meant by this, actually, it's marketing 101. You need to stand out and be different. You need to have your own personal brand. You need to understand what makes you unique as a consultant and then use AI to actually amplify that. ⁓ not dial it down, not rely too much on AI so you lose yourself in the process, but use it to ramp up that uniqueness. And just to make this tangible, let me give you an example of ⁓ myself. So as a consultant, I generally position myself, my background, educational background, social anthropology. So I'm interested in culture. and cultural insight and how cultural insight and qualitative work together. So that's kind of been my focus. So now what I'm doing with AI is going back to the work of a lot of great social anthropologists and AI has this great ability to allow us to look at data in many different ways by we just change the lens on our camera. and let's look at it from the lens of this anthropologist. If this person was looking at this data, what would that tell us about what's going on? So AI now gives me many different lens to look at the data in my area of interest, which is this kind of culture. So this is just one example. There are many different examples of the presentation, but again, we're trying to flip the narrative and not being this race to the bottom and quicker, cheaper, faster, you know. but actually how can we increase our impact and our influence as consultants using AM. So that was basically what we were talking about. speaker-1: So we can say that it's more about adaption. speaker-0: Yes. speaker-1: Yeah, and if that is the case, again, the one thing, the two things, fact, one is ⁓ the narrative was to get weird. But wasn't that the same every year because most of the qualitative researchers do that. now with the advent of AI, ⁓ which is my second point that you mentioned that, you know, we can actually check multiple LLMs to see, you know, if the narratives are coming the same way or not, or, know, but the data gets feed into that as well. And if you're to a particular industry when there is always an asterisk saying, disclaimer saying that, know, do not trust that the chat, it could be different. How do you synchronize with the data that you have actually got firsthand with the conversation with the respondent on the conversation with the respondent is exactly identical with what you're actually getting from AI. speaker-0: Well, yeah, maybe I'll flip that question slightly. So, and you'll have to tell me if this kind of answers the question or if I've sort of missed the mark. But the danger I see with, and we talked about this in our presentation as well, the danger that is there with AI. And I think we all know that it has a tendency to average everything out so that everything comes very vanilla. So if you over rely on using AI, think you're going to get a very sort of average vanilla type kind of view of the world or the language that's used to express the findings. True qualitative researchers know that insight lives on the edges, on the fringes. So if you get overly reliant on AI, you're going to lose that ability and option to see where the insight lives. And it tends to be on the fringes and the edges. So that's very much a danger. And why... Even the quality scale, it's certainly in my toolkit for use cases, but it's not there yet in terms of giving the depth that we can get from traditional methodologies. you know, if you compare... speaker-1: Let's say, mean, you know, there's when I was growing up with MR in the in this world, I always learned that, you know, to validate you do call to quant and then quant to call, or probably like quant, call quant. Similarly, when I'm developing a product right now with my company, I realized that, whatever information I'm getting from AI at this point of time is the data that we fed it. So even if I fed like 2 million points, data points into it, it might change the very next day, everything gets uploaded because the dynamics are changing instantly, very, very soon. And the behavioral patterns are also changing. By the time I am uploading the data, something actually triggered the intention of a consumer and he, other than buying X item, bought a Y item, whereas I have the information of an X item. I cannot severely rely on AI that it's going to give me the right information. And as a qualitative moderator, when I'm interviewing you, I'm really finding that difference. ⁓ you mentioned about, know, the most AI conversations in the research are about adoption now. And then why do you think that, you know, very few have an advantage? And what is that the cost of that gap which is happening? speaker-0: Yeah, I think it highlights to some points that I mentioned that if there's this over-reliance on, again, it's a drive to efficiency and I think a lot of practitioners ⁓ almost feel forced that, you know, this is the future, I need to be on board, I need to use AI. So I think the community feels this pressure to adapt. And I think now, and I sense it at this conference that we've both just been to, There's a little bit of pushback and a little bit more discernment in, you we're starting to find out what makes us uniquely human. You know, what AI is good at and what human, you know, we've got two very different intelligences and how much you can rely on one and how much you can rely on the other. The metaphor that I used at the yesterday's presentation was art. So, And I showed a famous painting on stage and then went through this process of saying, actually, if you describe what's in this painting, it's actually quite straightforward. There were four people sat in the diner and blah, blah, blah, blah, blah. But it only becomes interesting when you understand context. You needed to understand when it was painted and what was going on at time. then it only also you need to look at the apparent contradictions that are in the painting. And in doing that, that is telling us something about the culture at that point in time. I, in using art as a metaphor, I was trying to show that meaning making requires, you know, the ability to understand context, to resolve contradictions, ⁓ you know, and make meaning from time and place and context and contradictions. And that's still quite a uniquely human skill. My advice to people was to really double down. Don't let, make sure that you as a human are the guardian of meaning making. You're in charge. You reclaim the role of meaning making. Don't pass that over to AI. It's not like don't use AI, but use it in a way where you're speaker-1: Of course madam, using of AI is definitely needed at this point of time. Whether you're articulating a certain fact or you're trying to find out or researching a certain fact, it is definitely needed. ⁓ Most of those researchers definitely do that. Do you recommend that you know, if you do something like that, so for example, you research a particular product, you wanted to know what people think about it. Should you validate it by actually doing a qualitative interview? If yes, then what should be that number? that they can decide on to validate it? Because if I really have to compare the data which is coming from AI, it is massive. is probably in trillions and what not it would be there. Will that two interviews, three interviews qualitative will validate what I actually got from or will there be a certain number? speaker-0: So there's a couple of issues in here. And I think the question is, you know, what is good enough for a project? think that's the underlying question. And that depends on a number of things. So if I'm talking with a small medium sized business and they want to do something and you know, they might not have, you know, they might have two or $3,000 to spend on a project and you know, then they're not going to have budget to do qualitative research and they would never want to spend that. That might be like a quarter of their marketing budget for the whole year. So there's no way they're going to commission primary research. in a case like that, think AI and the deep research capabilities with the right guidance and the right questions is a wonderful way to get some learnings to ⁓ improve your ability to make decisions. ⁓ Sometimes at the start of a quality project, I will do the deep research just to unpack it, help with forming hypotheses, what have you. The second part of question, ⁓ I'll try and answer it directly that I, you I think to get a feel for a topic if you're in depth interviews, you know, you know, I think a minimum of around eight or 10. As a moderator, I've been through this process that you start seeing that you start forming viewpoints around. speaker-1: I mean, you're saying that there could not be a one static number that we can go off the context. speaker-0: Not really, It's always a dance with what's the cost of getting the decision wrong and how much, even though QAL is not about statistics, still the underlying question is how certain are the direction you, do you need to be or the direction you're going, what's the cost of getting it wrong? The higher the cost, the more research obviously you want to do to make sure you understand it at the level you need to understand it. speaker-1: Let's take it as an example. I would probably love to hear from you on this because you've seen AI shaping in the qualitative work in different parts of its own methodology. ⁓ Post-Chat GPT era, it has a different story. Well, it's the same story ⁓ with old methodology, wearing new clothes. What do you think with an example? speaker-0: In this sort post-GBT area, definitely everything is changing. mean, think Qual's relationship with tech is generally been quite positive. So I think if we just take a step back the last 20 years, know, pre-digital, we just had, you know, our focus groups and what have you. Tech has given us a lot of things. So it's given us mobile ethnography. You know, so it's given us real quote, but delivered in a way that's much more viable and cost effective. It's given us online communities, the ability to engage with people over time at a time and place which suits the people doing the research. So it's given us a lot. AI now, I think, is really shaking things up. ⁓ It's giving us quality scale, so it's given us something new in the toolkit. But then that then has started a kind of very introspective ⁓ mode of, what is quality? Because then there's this question, quality scale, is it quality? Or is it conversational surveys? And is it actually then stealing share from what might have been commissioned as QAM. It is genuinely new. AI moderation and analysis and everything is radically new. It's a radically different alternative. But as I said before, the focus has been on how can we make QAM more efficient, more accessible and affordable? think now, just this last maybe speaker-1: Nah speaker-0: few months, even at this event, there's a slight counter movement in hang on a sec. Let's remember what quality is and what we're good at. Let's not lose the craft. So I think that that feels like where we're at. speaker-1: And I think that's a very good point that you raised about radical moderation. I had this presentation seen in the conference and Shelley was talking about how the radical moderation should be done. ⁓ Let's say in the space where AI has, in fact, you also mentioned about tools and techniques about online communities. Now AI has actually flattened that out. You have all everything available into the system. ⁓ producing good enough qualitative interviews. I still believe there's a certain differentiation between a researcher and a core agency should have. agency has all the access to do these tools. And researcher, if being alone or part of the agency also get some access to these. So what do you think is exactly actually creating that differentiation between a researcher and an agency today? if we talk about tools and techniques which are already being available at AI and then everyone can produce good enough qualitative work. speaker-0: So is the question kind of, if we've all got access to the same tools, how do we be different? speaker-1: Yes, and considering that, you know, there's also a part of radicality that you mentioned, know, understanding from the perspective of the respondent itself. And when we have everything available into the system and then validating it, we have been speaking about that. So what do you think the differentiation, you know, to produce a good qualitative work will actually ⁓ create between probably a research agency or a researcher? speaker-0: So. We need to remind ourselves that we still have agency. We still have the choice about how... speaker-1: And when I say, to add one more thing, ⁓ researcher from the brand, because a lot of people are actually moving into the brand side and they're doing their own research. Researchers from the brand versus agency, both have the techniques and tools available to AI. So what is that differentiation that is creating them to work versus them to work? speaker-0: Yeah, the way I see this is, and I understand where the question is coming from, because I think there is a sense that, we've all got the same tools and everything. So how do we stand out as researchers and do things differently? We still have a lot of agency in how we use those tools. For example, call it scale. I have a very clear point of view about when I would use it and in what use cases, what examples, what it is and what it isn't. I'd have a clear point of view on a lot of things. So that means that I'm not just kind of sucked into this momentum of AI, but it's more considered. I'm like, okay, this is my viewpoint. This is what I think is good. So I think... I think that's one thing. ⁓ It goes back to what I said before, that I think you have to come back. to what is it that's unique about yourself and what's different in how you do the work and then double down on that. again, the example for me would be culture. There's someone at the conference yesterday, they're really interested in behavioral science. So whenever they do something, it's always through a behavioral science lens. So I think it's... It's really a dubbing down on what makes you unique and then using AI in a way that amplifies that and not just using AI in very generic ways. Okay, we're to do a project with AI moderation, AI analysis, and then you're just going to produce something very vanilla and the same as everything else. But if you use it in a way and then you add that behavioral change lens, you can actually do very unique work. it's using your, it's deciding how you want to be weird to use the term we used in the conference. What makes you different and unique and really doubling down on that and making sure you use AI in a way that amplifies that and you don't just become in the sea of sameness. You want to be raspberry ripple, not vanilla. That's what I... really wants. speaker-1: Tell me something about that weird ⁓ context that you guys brought up onto the stage. Is it like thinking out of the box or is it like something else? speaker-0: Well, was Dave's clever creative idea because everyone was talking about AI being weird because of that Harvard paper of Western, Educate, Industrialize, Rich, Democratic. And then Dave noticed that actually he was starting to see stuff being written about weird in a different way that actually in an AI era where everything gets flattens. and we live in the sea of sameness, actually you need to be weird as a consultant. You need to dial down your new difference because if you're not different in this kind of age we're going through, you're dead. it was just ⁓ a very clever creative sort of flip to flip the weird, the weirdness, AI bias into be weird, be different and stand out. not just as your company, but as your individual, as your personal brand, as a researcher, stand out and make it clear why you should be hired, whether it's your job. speaker-1: I asked this question to Dave as well. Being getting weird is a context that has been coming from ages. It's not very new. Out of the box thinking, out of the regular thinking mode, we always have to wear that hat, you know, how we can work things out. I bet 90 % of the people who were in the conference were not even concentrating on what is actually getting delivered. But even if they have slightest of the thought about what they've mentioned and about what he actually meant by saying weird, and if they have even implemented it, it would be a different story. So I ask you, since it came from him, what advice would you give? to the young researcher who really wanted to go weird or probably are not even thinking about, know, they just wanted to get things done. So what's your advice in this case? speaker-0: Yeah, that's a good question because I think it's, I think when you're, you know, several years into your career, you're quite you're almost fully formed and you have so much lived experience that you're quite clear who you are in a market research context and you're quite clear what you stand for and what you believe in, what your passion is. But I think when you're younger, yeah, that is, speaking, a journey. ⁓ I don't think there's an easy answer to that question because speaker-1: But you actually mentioned about the patterns. I believe you also mentioned about the patterns. So if you see 90 % of the people or the audience who are sitting in the conference, they follow the same pattern. And the conference delivered a good learning for them that you really need to get weird. speaker-0: I think maybe the answer is just experiment using AI in really creative and quirky ways. Everyone can have access to the quality scale platforms and they'll help you write a discussion guide and do it and produce the outputs. That's just working to be a project manager. I think to do really stand out work, you have to do something, you have to go against the crowd and use it a little differently. it's how you use AI. And again, I'll come back to that example of just study, you know, look at the in qualitative research, it's qualitative research is based on ⁓ many aspects of social science, know, psychology, sociology, social. biology, behavioral science. There are so many theories and models and everything that you're very useful and interesting to look at a lens of what you're learning. I would say play around what's interesting for you in the social sciences. Are you interested in behavioral change theory? Study that, learn about that, see how you can apply that to the world. Be an expert in insight. that is framed in a way that delivers behavioral change. That would be one example. Are you interested in culture? Follow your interests, follow your passions, see what's interesting for you and then experiment using AI because what AI does is it sort of democratizes expertise. So all this expertise from the social sciences is available. You can just pick a, let's pick a famous. speaker-1: Yep. speaker-0: anthropologist, Clifford Gitts or Mary Mead or someone. Looking at these learnings, what would Mary Mead say about what's going on about this? That's just a simple prompt. You can build custom GPTs or gems or what have you, but you can build your own frameworks or thinking lenses based on who you are as researcher. So think when you're young in your career, you want to experiment and play around and see what's fun. and see what makes you a little bit different. Just be known for something, just, okay, a qualitative researcher, be known for, stand for something beyond that. Play around, experiment with AI, use it in non-obvious ways, I would say. speaker-1: So as a researcher, would ask you this question. My boss is after my life and he needs to get the report done ASAP. And if I really look into this project, at least I need a week to do that. But my boss is after my life because the client is after our lives. What would you recommend? Because now the work to get them faster, speed becomes an element of it. ⁓ Data is available, which we used to in a week's time with AI Advent, it is like a couple of hours. So how would I, how would I being weird and solve the purpose and still not take a week and still, you know, delight my boss. speaker-0: Wow, there you go, that's thrown the of mission impossible. I said before, we need to reclaim the role of being the interpreter of ⁓ human meaning making, and that takes time. If that were me in that situation, first of all, might find, okay, well, why does the client need it at this time? They've probably got some meeting or something, so I might dig a bit further and find out, what meeting have they got? you know, what needs to fit into that. And I might find that actually, they need something very specific for that meaning. So maybe then there's a bit of negotiation, say, okay, let me give you a one pager just of what I think the key themes are, or something like that. And maybe that's good enough for what they're meaning. And then buy a little bit more time for deep thinking. I think Our role as trusted advisors is to make people aware of the trade-offs of what they're asking for. So if someone is saying, okay, I know that you normally need one week to really think and do it, but I've got one day, fine, I can give you something in one day. ⁓ We can use AI in this process for summation on that, and we can get some themes in that, but it won't be. in my mind, the finished thing, it won't have had the level of thinking that will make it as good as it could be. So I think our job is just to make people aware of the trade-offs when they're asking for something, ⁓ you know, that's all we can do. If they really need something for me, then we have to deliver that, right? So sometimes that's the reality. ⁓ speaker-1: Getting weird takes time, but considering we are in a very dynamic environment, everything runs at a very high speed. So there's always quietly something that we are losing when it comes to meaning making or sense making. What exactly are we losing? speaker-0: I think we're losing, if we're relying on, know, I obviously can, can, okay, let's talk about a specific core project. So in theory, even with normal moderation, let's say we've got a transcripts, AI generated transcripts, the transcripts are fed into the analysis, the, you know, all the technology is there to do something very, very quickly. And that will give something that, you know, we'll have some some usefulness. So yes, the speed can be met, but what's the trade off? Well, in qualitative research, analysis is so much more than about analyzing words. You're probably aware of this idea of low context and high context languages. ⁓ know, famously, Japan is a very high contact language, Thailand, where we are now is fairly high context language that you A lot of the meaning can't be taken just from the words. So if the AI is just analyzing the words without the context, without the emotion, without the felt sense of, particularly if you've done face to face, the felt sense that you get when you, like, we're talking to each other. You know, our brain's working a million miles an hour to get a felt sense of what you're feeling and thinking, and, you know, this is all going on. So we've got to remember that. Core analysis is not just analyzing words, it's meaning making and it's not just what's said, it's the unsaid and it's also how it's said and it's also the relationship between the two people that are having the conversation, the power dynamics and it's a whole range of stuff that a qualitative researcher will be taking into account when they're making meaning. Now, AI. I'm going to say can't yet do that. I've had arguments with people that say AI can. I don't think it can. ⁓ Maybe when we get to AGI, it would be better this. at the moment, from what I see is, I think some of the AI tools they have, emotion decoding and stuff like that, maybe that works part of it. But there's no way in a high context language ⁓ that an AI tool can replicate. the ability of an experienced ball type researcher. No way. speaker-1: But even in this case, we've been always talking about hallucinations, which AI does a quite number of times. Low-context language may be part of it, highs definitely. And when it comes to ⁓ cultural nuances, every word has a different meaning altogether. So what I'm trying to extract from AI versus what I'm getting has sometimes a difference. And before our shoot, we were talking about how I'm actually generating my own thoughts. And I have to constantly look into whether this makes sense or not. Because it's more western centric. So everything comes in say dollars, in the names of Americans. Not necessarily citing exactly where I am. So probably if I'm Thailand, the name should come as Thai names or the currency should be Thai or the food should be. So the cultural nuances are missing unless and until we feed into it. So those hallucinations are always there. know, mismatch is always there. And it's bigger risk. pre-weturing certainty at this point of time. How should call teams handle this uncertainty when working with AI tools? speaker-0: Yeah, it depends what tools they're using. I think a lot of the platforms that have been developed are addressing... I think if you're using a general LLM, that could be quite dangerous in call analysis because you would definitely get hallucinations. And if you ask it for supporting quotes based on the transcripts, it will give it to you. But sometimes it just made it up. Yes. So, it's very dangerous to use that. general LLMs. Now, I wouldn't say I'm an expert on some of the core analysis tools, but I think my understanding would be that they've addressed, you won't get hallucinations, so it will be much more grounded in the data. And certainly the difference between using a chat GBT and a notebook LM, which is based on that, I think the technical term, it rag system or something, but it's only taking from the information that you're giving it. I think I think the kind of products have been built to avoid hallucinations, but that still doesn't mean, it's still an analysis of written words and quality. That's 20 % of what quality is. The other 80 % I think is still missing. And that's why we talk about human in the loop and everything. So, AI summation is very good for getting themes, unpacking something as part of your analysis and thinking process. But you shouldn't be 80, 90%. You shouldn't be relying 80, 90 % of your final report and analysis on that. It should just be part of it. Some of the grunt work, the content analysis, but the real thinking still needs to. speaker-1: I think that's a very good point that you mentioned about 80 % of, you know, relying on your monelaces because client don't buy AI. buy the confidence, they buy the clarity, they buy the trust that you're giving them along with the, you know, report. When does AI is strengthening a researcher's authority in this case? And when does it accidentally weakenings it? Where do you think it happens? speaker-0: I think it's, I maybe start on the weak in it because, and this is from my own experience as 30 years as a researcher. So if you've done a qualitative project and you've done interviews yourself, you've done moderation, and you've done the analysis and what have you, you you've lived and breathed that project. So when you're delivering that presentation, you know, that is... part of you, you've been there, you've had that experience of interacting with people. And that comes through in the way that's delivered your confidence, your certainty. If you've delegated that to a lot of the thinking and process to AI, if you haven't done the interviews, or you haven't even watched them, you just kind of used a platform and it's gone through the process and you've got the outputs. I think you've lost a lot of your value. The value has come from the platform, it's not come from you. You haven't done those interviews, you haven't done the analysis. You're just a messenger then, really, aren't you? speaker-1: The point is, Craig, I if I have not watched it, sorry, I have not done the interview, but I've watched it. And still there could be certain things which the moderator who has taken the interview then would have missed a certain point by, say, by large. ⁓ For example, I'm interviewing you right now and I'm actually thinking what else I can ask, you know, what is my inquisitiveness about this piece. And I can immediately ask you, I can in fact probe you to understand, if I'm getting things right or wrong. ⁓ What if I've not interviewed you, but I've watched the interview and then dragging down the transcript and then writing a report, but still know that, you know, this element was not in the questionnaire, which is not being asked by the, doesn't that weaken the report by large? Because I think that this question should be there. I'm just thinking aloud. speaker-0: Yeah, I'm not quite sure I fully understand the question. speaker-1: So it's more like, you know, we're interviewing and somebody's observing us. And then the reason why I'm saying is because when we were in this radical moderation presentation yesterday, she gave us a task. So two of them were talking. One was a moderator, the other was the interviewee. And the third one was actually observing. So he was asking all the questions he should to me. And then we switched roles and then I was asking all the questions to him. But the third one was actually observing us. speaker-0: Yeah. Yeah. speaker-1: So let's say, see the same as an example. So interview is being done, but none of us conducted somebody else conducted. Now I'm looking at that interview and I'm trying to observe that interview by the conversation that happened and then extracted that. But then I realized that this point was missing there. speaker-0: Okay, I think I'm clear on both. Yeah, so I think the speaker-1: more radical moderation, then that's exactly where the point is coming speaker-0: The process we go through in sense making. So, you know, if I'm moderating, I've internalized what I'm trying to do in this project. So I'm always trying to get to not even the questions that I want to ask, but the underlying business question. it's, you know, that's why moderating is so tiring because you're constantly trying to answer that underlying question and pivot and you're asking something. ⁓ You're getting a response and you're figuring out what does that mean and what's the... So the process of moderation is this very intensive way in which you're answering this business question. And it's not as intense as observing. If moderation is dialed up to eight or nine, then observing is four or five. So you don't get the same internalization. And also when you're in a room with someone as well, you feel it. It's a felt sense of a truth. When you're watching something, it's definitely better than not watching and you definitely learn a lot, but there's nothing more powerful than sat with someone, especially if you're sat in their home or in their own environment and you're taking in that information as well. So you definitely lose something. I think the more you step away, ⁓ ⁓ If dialing up to nine is moderating and then watching is dialing it down to seven, you're not in the room, you don't quite have the same. And then actually you don't even observe and you just feed it through a platform, then you're not close to the source, right? You're further and further further away. Then you've lost your authority to... You just become a medium. speaker-1: That's my point. speaker-0: know, pass on the message, but you're not the, if you've moderated, you know, you are the message, your experience is what you are. speaker-1: Can we easily say that AI is not strengthening the capability or it's basically weakening the researchers capability to think beyond? speaker-0: a danger. I'm not going to make that bold claim because it all depends how it's being used. I think there is a danger and a concern that if a lot of projects are done through quality scale, and again, just to be really clear, I think it's a wonderful innovation. adds a lot to industry. But if 80-90 % of the quality is process driven through that, then there is a danger that the craft of qualitative research will get lost because you learn the craft from speaking with people. speaker-1: And... speaker-0: even better if you're doing it face to face. ⁓ That's how you that's how you learn the skills of Quo and thinking on your feet. And, and then, you know, you've been moderating all day, and then you go home, have a shower, and then you're thinking about what you've learned, and you get this aha moment. And, you know, that that's how Quo works. Yeah. And if you take away that closeness, and it's all done at a distance, and it's all automated, then some of the craft may get lost and that's the danger. I think there'll always be a need for, ⁓ know, but that's could be what happens. speaker-1: And when you were actually talking about that aha moment, was realizing AI is strengthening to a certain extent. We are all seeing it. And again, ⁓ no further questions on that. But when we do quant, there has to be an email address people use to answer surveys when they receive those surveys. And over a period of time, we see that bots are answering those surveys. data fudging. existed at certain point of time, but this is not data first. This is actually duplicating a certain respondent basis, the behavior that they've been getting. So what's been trained onto that. Now that's one part of the quality thing, which is by large being stopped by a client. They don't accept that kind of a data. And I'm yet to explore this. So this is more of an out of the syllabus kind of a question. Now you've been doing qualitative for a while now, and I'm sure that sometimes, you know, you're overwhelmed with the work. So you distribute the work to somebody else at the age of AI now. Do you see people actually doing that data capturing work from AI and you're able to catch them up and they're not talking to people? the qualitative space? speaker-0: quality I I'll be honest that, while I have a point of view on quality scale and it's part of my toolkit, most of... Because my company Fuel Asia, obviously, is focused on doing research in Asia and quality research in Asia. And most of what we still do is traditional quality. And in doing traditional quality, in most Asian markets, we still use actually quite traditional recruitment methods. it's recruiters recruiting real people. So, of what we're doing, there's not an issue with QC or fake people. I could certainly see why you're asking this and how it can be an issue. There's obviously a massive issue in QAN at the moment because of the ability of AI to you know, to behave like a panelist. speaker-1: Honestly, I don't want to get into the deep sea of quantitative research, but I really wanted to understand from the qualitative part because in a traditional way, you can easily QC it. Yeah. But now the advent of AI where people are getting AI moderators, I'm sure we're not very far and probably by the before the end of this year, we'll see synthetic respondents coming in. So where are we heading with the qualitative research? Where as of now, when I speak to people, I say that's the speaker-0: Yeah, yeah. speaker-1: Qual era is coming back. lot many people are getting into the qualitative approach because they don't want to lose the human touch. But if tomorrow people see that everybody needs money, everybody wanted to make some money and when they see a value there, they'll probably create synthetic respondents and synthetic moderators and both of them are talking to each other and we are finding extractions out of it. speaker-0: think there's two parts to this question. One is about sort of fraud and then the other part is about synthetic respondents. So two quite different questions. I mean, I'm not a sort of tech expert in the sense of, you know, I don't know. with the research tech space, we could make sure that, ⁓ know, presumably, all this tech that's being built has guardrails to make sure that, you know, even people that are doing the interviews from my AI on moderators are real people, I'm assuming they've got methods in place. But I don't know, I'm not totally up to date with how they do that. But yeah, I get that that could happen. I think the other question, maybe the more To me, anyway, interesting question is, is the value in synthetic respondents in the qualitative space? Very, very hot topic. Very divided research community on this space, very polarized. In my mind, from what I'm reading and hearing about, I see there's much more potential in the quantum space because there's... studies that, you know, there's 70, 80 % accuracy in predicting, you know, I think the devil is in the detail about how these respondents are built. I don't at the moment see much value in a synthetic respondent for Qualitave, unless it's maybe going into a project, you know, maybe it's developing some stimulus, let's just test and or create some hypothesis. you know, maybe at the beginning of a research project, synthetic could add. speaker-1: ⁓ That's why I don't want to go that far. because I am not too sure. I've been hearing a lot many people talking about synthetic respondents answering the human moderator. That's by far, I mean, that's future. I'm not too sure whether that will happen or not. you know, this has come up over a period of time. My question again was to how do you think, you know, people that are doing these online surveys, sorry, qualitative surveys, Are they in any ways creating a data which is not quality-wise correct or they are fudging the data? Is it possible in the age of AI that they can do that in Coal? speaker-0: in Quo. It's definitely possible. it all comes down to... speaker-1: And if it is possible then how you know guys like you or clients find that out speaker-0: Well, I'd probably go back to what I said earlier that because in the market side working, we've been working with the recruiters we've been using for like 10-15 speaker-1: I again, we consider that. I I consider that because, you know, we still, this whole Asian belt is very traditional in almost every approach, whether you talk about culture, whether you, the Western influence is still limited, whether, know, if you talk to somebody, you can still find some Western influences in clothes, but traditional, the way they think, there's a very traditional approach to it. Tomorrow, if, and not very far. let's say things changed and then you really have to look into it. I'm just trying to understand your perspective. Let's not say what fuel Asia. I'm just trying to understand your perspective. If certain bad quality work happened on qualitatively, you probably gave it to somebody and he just fed in the data without even talking to you. How would you know that this is bad quality? speaker-0: ⁓ Yeah, that's a tricky one. ⁓ I mean, generally the way good quality in quality starts talking with real people. That's the first principle. You know, having watched Quar over so many years and, you know, either as moderating or behind the mirror or whatever, you have this sort of sixth sense when something's a bit fishy. I'm not sure about this personally, you know. And it's a very human thing. I'm not sure there's necessarily a tech solution to it. I mean, my... I'm just thinking of an example. speaker-1: Because A will not be able to defend its own creation. So if you actually go and use a tech and you know say can you tell me if these are fake answers and it always politely agrees. So in looking at the response it might say yes it's speaker-0: So there's a lot we can do in the way we design studies. ⁓ you can you can design if this becomes a real problem in the industry, you can design a discussion guide in a way that will highlight contradictions. So you can ask them different questions in different I mean, you can do this in quantum as well. But when you know, if they're coming in and they're doing role playing, and they're fraud, and they've been told that, okay, this is who you are in this thing. So you need to behave in this way. Good questioning normally can you. You know, so I think if this becomes a problem in the industry, I think we can be one step ahead and we can, we can be smart in the way we ask questions and we can show that, this person has said this, but then they said this and they're all over the place. There's something not right here. Normally, if you, with AI moderation, it's going to be tricky because the AI might speaker-1: Hmm. ⁓ Yeah, I mean, one of the questions I actually asked somebody at the conference and she was promoting, she was presenting a case that they feed in lot of data to let AI probe on a certain question where respondents like probably juggling with words or something like that. So I said, if AI is probing, it means it is creating the questionnaire to be biased. That's what AI wants, not the human. Whereas when it goes natural, we're talking, we can dig into those probing words coming from our brains. And I'm again, will not say that it becomes bias, but it more is a conversational kind of a discussion that happens. Whereas AI is a tool, it's not human. So probing would not work. You can only feed in the data as much as you can. But in order to put that data down or feed that data down into the system, you really have to do real interviews. You need to get those words so that, your AI can also understand that I am not limited by different, ⁓ nuances or different thoughts that a consumer or a respondent or people you're interviewing would think about. So I think that's where the challenge lies. ⁓ just randomly, I thought, know, when we look at quant, there's this massive bot inflow happens and then lot of data gets cut down. Same happens with qualitative or not, I was trying to understand. But there's one thing which I really wanted to ask you in the conference, but I didn't get time and I have to leave early. And the question was, I wrote it down. looking ahead to 2026, you both said, Dave and you that, you know, the world that wins will feel a little uncomfortable or weird that we started the conversation again. So, speaker-0: Yeah. speaker-1: How do you know you're pushing in the right direction? And that's basically a question to summarize the entire podcast. speaker-0: Well, Dave did a really interesting exercise at the beginning, which was he got everyone to get out of their phone and type into the LLM of their choice. Yeah, well, actually, he says to do it by themselves, what makes and then actually put your own name in, you know, original and unique and see what comes back from, you know, do research on yourself. speaker-1: Who knows Dave? speaker-0: The point being is, you know, what's our, what's our point of difference in the market? Increasingly, Google searches dying, people are actually finding out from language models. So what does the language models think is unique and different about us? then, okay, so it's spell, I've done this, I do this exercise regularly. I think it's a great one. What does it tell me about me? And then, okay, so what, what do I need to do to to a different answer to shape it. this is marketing 101, right? Just differentiation. How do I want to position and what do I need to do to position? ⁓ that means using social media, doing things like this, putting out content in there. ⁓ But not just that, you've got to live that brand as well. You've got to show up and be that with clients and in the way you deliver. you know, how, what you offer clients and everything. So it's a total brand perspective. So I can't remember the question now and I can't remember if I've gone off track or not. speaker-1: You're actually right on track. that differentiation is what, in fact, that was my last question with, I interviewed Dave and being weird is something of course, you know, we, are talking about it in this podcast, but it wasn't there, but definitely, you know, when we're talking about AI and its, you know, origin and how it is being now, we all, we also talked about, know, where do we stand in terms of as a human, you where does our brand exist, whether it exists or it doesn't. So go and check on. And in fact asked me to do as well. And I'm going to definitely do that. My thoughts to him was that I've been feeding a lot of data about myself to AI. So whatever comes out is me. And you know, the things which I mentioned about, I created an agent on Gemini, that's all me. So if I type, give me a next story about me since I've already fed on, it gives me a story which is not even related to me. It creates a story. And this is very Western centric, which is why it stays. It says that even if I say I met say an Indian name Ashok and it would not give me a show. It would give me say Mark. Yeah. And it becomes completely Western. And that's where the story has to be rewritten again. I have to go edit it multiple times until unless I feel that this is exactly my story that now it's ready to be published. speaker-0: Yeah. speaker-1: And that's exactly what he also defended saying that, know, unless until there's a specific that you're treating AI to say exactly like you, you really need to feed in a lot of data. But it is not the right way of doing it. Probably things change every time it changes. You know what we are right now. Ten minutes ago, we were somebody else and ten minutes after we will be somebody else. speaker-0: Yeah, mean language models always evolving. It's bit like search engine optimization. That was always evolving. The algorithm is always changing. you always had to kind of react to that. And I suppose that's the same that you constantly need to check and make sure that what's coming out is broadly in line with how you want to be seen in the marketplace. But yeah, it's just marketing 101. Stand out. Be different. Be conscious. speaker-1: How evolving, speaker-0: about your personal brand. I know we had a chat about personal brand versus company brands and I shared my view on this that I think personal brand is becoming more more important because of the ability to ⁓ speak authentically and humanly as a person. But when you're representing a country, a company, it's a bit more abstract and you know, but when you're talking from yourself, you can share personal stories, you can be very human. speaker-1: Exactly. speaker-0: So I think focusing on personal brand is going to be huge in the world we're moving into. speaker-1: Thank you so much, Craig, for making time out for this conversation. It was definitely very enlightening. I mean, think ⁓ back to back, I just had two conversations. It was full of stories. It was full of knowing how ⁓ the world is looking at market research these days, how AI is participating into it. And it's easy to simply comment and then actually be part of that comment. So just thank you so much. ⁓ speaker-0: Thank I've really enjoyed it. It's been a pleasure talking about something I'm passionate about. Yeah. speaker-1: All the reasons why I really wanted to have you as a guest on my podcast. Thank you so much. Thank you. So one thing that's come through clearly in this conversation is that AI hasn't removed the need for qualitative skill. It has exposed how rare it actually is. When everyone can summarize faster, synthesize quicker and sound more Polish, the real differentiator becomes harder to fake. Judgments, cultural sensitivity, confide with ambiguity and the courage to say something that isn't obvious. What's striking is that the future they both describe doesn't reward playing it safe. It rewards work that feels a little uncomfortable, a little surprising, sometimes even a little weird. And maybe that's the real shift we need to sit on or sit with. AI hasn't made qualitative researchers less human. It's raised the bar on what being human in research actually means. Craig, thank you for bringing such a grounded and culturally... ⁓ rich perspective and forward-looking perspective to the conversation. This wasn't about chasing tools or trends. It's about sharpening the thinking. And for everyone listening, the question to carry forward is simple. If AI is available to everyone, what will you do with it that makes your thinking harder to replace and not easier? So thank you for listening, and I request you to subscribe the channel because such conversations are rare. They happen in certain closed ⁓ conference rooms and not everybody is able to access it. I am really thankful for the team to let me be there and hear them firsthand and helping me bring this out conversation for you guys. please subscribe, share this with your colleagues and friends in the market research industry and the consumer insight industry. And I will be bringing more such conversations to you guys. Thank you for listening and I'll see you in the next conversation.