Marketing Week Webinar: AI Personas – A New Approach to Understanding Your Customers

Hello, and a very warm welcome to this webinar for Marketing Week brought to you in partnership with Market Logic. I’m Josh Stephenson, Trends Editor at Marketing Week, and today we’ll be exploring the topic AI personas, a new approach to understanding your consumers. Traditional buyer personas have long been central to marketing strategy, but in many organizations, they remain static, buried in reports, quickly outdated, and rarely used to guide real decisions. A new generation of AI powered personas, however, is changing that. By leveraging existing research and insights, marketers can build dynamic audience models that support earlier testing and better decision making. What’s this include? Well, we’re looking at qual based personas, which can simulate realistic customer perspectives and quant based synthetic panels, which allow teams to validate ideas at scale. Over the next hour, we’ll explore how organizations are using both spot faster innovation and more confident marketing decisions. We’ll also be taking audience questions, so please post these at any point in the q and a box, and we’ll get to as many as we can towards the end of the discussion. Now for today’s webinar, I’m thrilled to be joined by Joe Rini, director of product product management at Market Logic. Welcome, Joe. It’s great to have you with us. Why don’t you tell us a little bit about your role? Yeah. Thank you very much, Josh, and good afternoon, good morning to everyone. So my name, as Josh said, is Joe Rini. I’m director of product at Market Logic, and that means that I’m kind of involved across our product suite, developing and rolling out new offerings to our customer base. I’ll talk a little bit about Market Logic in a second to, bring some context there. And in particular, in the past year or so, I’ve been heavily, involved in our synthetic respondents offering, so really our personas, both Qual and Quant, and rolling those out, together with our customer base. So great to be here. Thanks for that. Well, I imagine it’s been a busy time. But before I start grilling you and asking you a few questions, I’m gonna hand it over to you for a short set presentation to learn more about the Market Logic kind of suite, and we’ll then dive into a deeper discussion into the topic. So, Joe, over to you. Perfect. Thank you so much. So I’ve just pulled up on screen a quick slide to explain a little bit about who Market Logic is, who we’re working with. I don’t wanna spend too much time on slides. The idea is I’m gonna you know, about five minutes here and then a quick demo and then really for the discussion. So Market Logic, you you may or may not have heard of us. We are really kind of the industry leader, providing a SaaS software solution for insights teams of, you know, some of the largest global companies out there across consumer goods, pharma and health care, retail. And in addition to all of those, insights colleagues, we are interfacing with, all of their stakeholders, so marketing, r and d, innovation teams, and so on. And we really offer kind of a suite of AI with human in the loop tools and offerings, holding all of the proprietary market insights and data that our customers have, and then allowing them to work with it, generate outputs, innovation flows, and so on. And the kind of as alluded or as discussed, the key that I wanna talk about today is our personas offering. So we’re really solving with personas, but also with some of our other offerings, this research dilemma that we see in our insights teams, but it also interfaces a lot with the the marketing colleagues, which is this trade off between having to act fast, to avoid missing opportunities, but then trying to maintain things like brand loyalty, prioritizing budgets, and not trying to waste, ensuring that we’re not wasting investments, tapping into audience or understanding of our our customers and our audience as opposed to guessing. And I think that’s really somewhere that the personas offering goes a long way to solve. And this overall offering really lets our customers in marketing, in insights, in innovation, run forward looking audience intelligence. And I think that’s a key distinction from some of the more traditional connotations that personas brings up bring up because we’re really trying to bring this intelligent layer to our customers’ understanding of their customers at scale without sacrificing all of that research rigor, but still allowing for speed. And some of the key points here, but, of course, we’ll into this throughout this hour, is our personas are built on our customers’ current data data, not just historical snapshots. These are unique personas, and I’ll talk about a couple of different ways that they’re set up to ensure that they’re really based in our customers’ understanding of their customers. They allow for hyper personalization. And as we’ll touch on, they facilitate both qual and quant audience research. So this is a bit of a wordy slide, and I really wanna just quickly go through both of our qual and quant offerings. I’m gonna cover both the qual piece, which is really that LLM based large long language model based chat experience that most of our customers and audience is a bit more familiar with, and then talk about the the quant piece, which is more of a new offering for us. So the qual piece, you’ll see this all in the demo, but we really offer a couple different flavors. The first is personas that are set up on our customer base’s existing research segmentation persona programs, often living in static decks or recorded videos, but now brought to life in this chat based experience. But we’re also able to, build personas for customers where there is no persona program internally or it doesn’t cover the full breadth of their, miss you know, their global reach by using that proprietary data that we hold on behalf of the customers. And there’s a couple different flavors there. We can also go to external data sources, for instance, usage and attitude survey data, which is really a key piece of, source of data that can be tapped into to build personas that can then interact it with. On the quant side, it’s more of a shift from a chat based experience. And here, we’re really facilitating more of a quantitative output, the traditional kind of concept testing and so on that, you know, our marketers and insights customers are very used to. And here, we’re still leveraging this persona capability, but we’re able to go to a large n number, one hundred, two hundred personas that live in the back end and really get quantitative robust outputs from them, returning scores like top two box, the, you know, Likert distributions, and so on. And I just like to show this one because I think it’s often a little bit more foreign just so people can kinda wrap their head around what they’re then gonna see in the demo. So the idea is it’s more of an interface. You can come into upload concepts, select your your panel, and really get this robust output. But enough about the slides. Let’s actually go into a demo environment, and I’m gonna really take you through how we see our customers interacting in this in this space. So I’ve come into a demo setup. This would resemble the persona component of our overall platform. So as I said, there would be additional tools in here that our customers are using to peruse their data, to rework it, to communicate it internally, and so on. And the persona’s environment, imagine that, you know, typically, a classic case would be our customers might have a set of personas set up across a couple of different, categories or regions. So in this particular case, I’m gonna take a look at a couple hair care, expiration routes in a second. This customer has got some personas built around hair care, around personal care, and so on. And if we just click into one of them, let’s take a look at Martha Gomez. So very classic standard case. She probably lived in a deck. She might have been based on some u and a serve survey research or just an internal brainstorm from an insights innovation team. And, you know, this is a quick explanation of who this kind of fictional persona is. And the challenge has always been, okay. But how do we actualize this person, bring them to life in our marketing innovation, etcetera, flows? I’m a show you a couple different ways we could chat with a persona like this. So just for the sake of speed, I’ve actually precreated a chat here. So the very first chat I wanna take you through is a a typical exploration that we see our customers doing. Early stage, they you know, maybe you’re a marketer and you wanna come into the system and just understand, like, who is this particular segment? Let’s figure out a little about them and then go down a route of, for instance, creating a product idea with them. So you can see I’ve simply asked Martha to describe herself, and she’s given me a kind of nuance, very realistic sort of feedback on what she does in day to day and so on. And at this point, this would typically live almost verbatim in those decks that I’m talking about, but now it’s being brought to life in this very interactive way as you can see. And here’s where we see it start to get really creative and interactive. So I wanna now investigate some work around children’s hair care needs because I’m gonna gonna communicate internally to some stakeholders. So I simply asked Martha, okay. What’s the biggest pain point you are facing in providing for children’s hair care needs? And then you can see this this answer. This is now gonna be, of course, based in that persona description. And you can see that she’s, you know, struggling with time, but also trying to get the preventability of the hair each morning. Different kids have different hair types and so on. So, you know, it really talks to anyone who understands what it’s like to have kids with these hair issues. And you can see what I would then be able to do is, for instance I’m gonna just paste in a follow-up question here. But, you know, what job would this new product that we’re, you know, talking about considering here need to solve, and how would that be best communicated to you? So something very clear that, you know, marketing colleague might be doing in the environment. And now you can see the power as Martha is reflecting what I’ve just asked. And, you know, a couple seconds later, she’s coming back with core needs that this product would have to solve for her. So for instance, work with multiple hair types. Right? And then just a little bit on the the way that it would would need to be communicated to her. And none of this is now existing in the underlying, let’s say, deck or presentation that the persona would have been built on. It’s now really working with my questions and the persona’s understanding of its overall kind of, yeah, profile and giving me this kind of creative ad hoc responses back. So suppose for this particular journey, I’m more or less done, and I wanna communicate this internally, I could summarize that chat, and you’ll send that off, downstream. What we see a lot of customers doing is leveraging our image, generation capabilities. So I’m simply gonna create an infographic here. And what it’s gonna do is now come back in a second with this with a high quality image, capturing a little bit about Martha, capturing a little bit what we spoke about, the product that you suggested, the needs that you suggested. I can then take that and send that downstream along with what I’ve understood here. So, you know, very quickly, of course, I hadn’t seen this before. It was just generated, but this is Martha, some of the issues that she’s facing, some of the product ideas she’s looking for, how that might be communicated. So everything we talked about in that chat, and I can simply download that, copy that, and send that off to a colleague, and off we go. So coming back to the environment, I now wanna show you a kind of, let’s say, faster or higher impact way that I could have a very similar exploration with Martha, and that’s with our AI moderated discussion. So now rather than having insights or marketers needing to lead that conversation that you saw me have, I can simply let’s go back to Martha. Kick off this AI moderation piece, and I’m gonna say, could you please just paste it in here. Understand the response daily routine, pain points around their hair care needs, and some suggest a new product direction. So kind of close to what you just saw me manually carry out with the persona. Gonna hit start chat there. The AI moderator is now reflecting on what the the the kind of task is here, and you can see the just the, you know, the quality of the question and the length of what’s being investigated is a little bit higher than I as just a person quickly trying to type there was doing. And we see this back and forth now happen with Martha feeding back her inputs and the moderator then considering what’s been stated and coming back with questions. It’s so fast that, actually, you have to go back later to to take a look, but let’s give it a second. Typically, three to five questions would be needed to kind of nail down the task that I’ve just asked the moderator to carry out. And the conversation is now concluded. Of course, we can scroll back up, go through it, could export it, and a summarization of this chat is gonna pop up in a second, auto queued up so I can then see, you know, what was discussed, what was the idea of that chat, what was the outcome, key findings, and so on. So kind of a juxtaposition of that manually led investigation that we can do with personas and now this newer AI moderation where you’ve really got an AI moderator, of course, enabled by your input and this AI respondent or persona in this interactive back and forth. So those were all personas that would, as I said, have typically have been set up in the system on behalf of our customers. So these would always, you know, live in a PowerPoint presentation and so on. What we’re seeing, though, the flip side of that is more and more customers maybe don’t have a fully built out set of personas, don’t have any, maybe they’re old, maybe they’re not being updated or pulling on the most recent content. And that’s where our on the fly persona building builder capabilities can factor in. So with this capability, let me just click here, marketers insights can now simply come to the system and prompt it to build them with persona based on repository of data that we hold. So all those insights and data that we’re holding in our repository. And I’m simply gonna ask for a UK so I think you saw there was no UK based hair care personas previously based hair care persona. And let’s go with splurge is Alright. So this might be a kind of niche persona that I wanna talk to. It doesn’t exist, but I still wanna run all these ideas by them. I’m simply gonna create the persona. And what now happens is the system’s gonna go away, search through all of those market insights, UNA content, and so on, and build that persona for me. Typically takes about forty five seconds. So I’ve just pre prepared one, and you can see systems come back to me with, like, Charlotte Whitmore here. And I can click on view details to really see what was done to create this persona. So it took a look at a bunch of different market research, you know, kind of across both European and UK sources and so on, and built that persona on the fly for me for this particular next exploration that I wanna do. I’m now gonna chat with Charlotte. Well, let’s even sorry. Let me pull her up here. Here she is. So I’m gonna go ahead and chat with Charlotte. But what I really wanna do is, of course, I could ask her a bit about herself. Tell me. Just to get that intro and very similar to what you saw me doing with Martha previously. Of course, I could do this with a group of personas as well. So now I can really start see a bit more about this person. But what I wanna do is actually imagine that, together with, marketing, I’ve been working on some hair care concepts. So here, you can just see two fictional hair care concepts at this point. And I want to get the personas early feedback on these. So I’m simply gonna drag that in and ask for concepts and ask that she reflect on both these concepts. So she’s now gonna take a look at both of these, both the copy, but also the images and so on, the pricing, and give me back her reflection on both of them. Likely, she’s gonna actually pick a a favorite as well, all based on that understanding of this UK based persona that I just built, you know, seconds ago, essentially. So let’s just give it a second as she’s reflecting on those. And once she answers, I’m gonna, you know, probably likely take away those inputs. I might also communicate them downstream. I might rework them here in the chat. So first of all, she’s provided feedback on both, starting with the nourish concept I was showing you. It’s not particularly appealing. The price point’s also potentially, let’s say, off here. Yeah. It’s not fuss enough. I or it’s something about not being it’s be being too price too low for her kind of splurging attitude. There’s one that’s a little bit more in her wheelhouse. Right? So you can see very easily how I’m able to start tapping into this type of realistic feedback from a persona that essentially didn’t exist in the repository seconds ago. And I might go, okay. The favorite, please suggest a few slogans. So here, we can really start to tap into the creative aspects of the large language models and the persona, and let’s go a little bit down the road of the, repair and shine, concept. And she’s come back with a couple different slogans that might appeal to her if we’re gonna be working that into a campaign, for instance, with reasoning for why she’s chosen those. I hope that gets across the power of the qual personas, the different ways that you can interact with them on early stage concepts by them, get their feedback, get their creative outputs, and so on. And now I’m gonna switch gears into that quant environment that we spoke about. So what I’ve come in here of course, you can see it’s a bit of a different look and feel. And I’ve uploaded the very same concept, and I’ve now selected a panel of personas. So there’s twenty of them. You know, quickly preview them if we needed to. But who they really all are and their in their details is not as important in this type of offering. And I’m now gonna ask each of them. Actually, I’m gonna apply a gender split. So quite typical that we’ll see something like a seventy thirty female to male split if it’s a consumption household consumption item and simply have them all now reflect on the concepts. And then we’re gonna, behind the scenes, convert what they’re saying into a quantitative output. So I think this is quite a step change from what you saw with that more qualitative chat based investigation into the personas into now really replicating what’s available via the the the quant offering here, the panel offering. So as you saw, it’s now essentially extracted both of the images, and it’s coming back with mean scores, top two box, bottom two box around both of the concepts. So you can see that neither of these were particularly favored by the by the panel. I can then take a look at kind of a deeper breakdown of where along the the the five scale Likert scoring this came in. I can even inspect the individual responses that each of the panelists made, and I can run, you know, an analysis on top of that. So I’m gonna ask a large language model to take a look at all those responses and let me know, like, what were the top reasons for the those that actually fell into the top two box. So it’s summarized the top reasons. It’s also pulled a quote from one of the personas for each. It’s also giving the top reasons for rejection. So I hope that demonstrated kind of the the step change there in terms of being able to take those early chat based persona investigations over into more of a real robust quant type of output. Great. So I think with that, I wanted to stop demoing. Let me stop sharing my screen here and come back to you, Josh. Fantastic. Thank you, Joe. That was a fantastic introduction, and I always admire anyone brave enough to do a live product demonstration on a webinar. It always takes some courage. Before, I dive into some of the questions that I have, I’d love to remind everybody watching at home that you can still fire questions in for Joe. We’ll try and get through as many as we can in the end of a panel kind of q and a, so do keep firing them across. But before we get into that, guess we can kinda dig more into the the kind of the granular into this sort of topic, Joe. And let’s try and think about what’s actually really different here compared to the personas that most marketers already have and that they are using. Perhaps you can give that little bit of distinction between call based AI personas, quant based synthetic panels, and and the advantages that they could perhaps have. Yep. Yeah. Good question. So I think there’s couple different axes or different points to to cover. I think let’s split out the qual and the quant first because, in fact, I think many marketers are indeed able to go to these, you know, paid quant respondent data in a way that, know, exists out there in the market and has existed. And so we’ll cover that in a sec. But, also, in terms of the chat based personas, I think we see a kind of a a mixed bag. So most marketers will have access to some kind of static materials, which I kind of been alluding to or mentioning. So they’ll have PowerPoint presentations. Sometimes we see, you know, live recordings or, yeah, live presentations that are then recorded and available internally in ninety minutes on, I don’t know, the the hair care personas in the UK. Take a look at this video, enable yourself around it, and then try to actualize what you know about these personas in your day to day work and your campaigns and so on. So that might be kind of in a best case if they don’t have access to one of these Gen AI large language model backed chat experiences yet. That’s more or less the baseline. Of course, some customers that we have and, you know, prospects out there that we see actually don’t have any persona work really done. And there, I think it’s even more of a a stretch to say that marketers have access to personas. They’re more or less forming their own opinion about their, you know, potentially, the key customer base for their given product brand, etcetera. They don’t really have this kind of robust persona work that’s been done by an insights team or the marketing teams and so on. And that kinda varies by, as I said, maturity size and so on. And then if there is an LLM based, so something like the chat experience that I’ve I’ve showed you, We typically see kind of a spectrum, but we don’t often see that there’s a company wide accepted uniform environment that’s using these vetted personas to interact with. Rather, it’s more piecemeal. There might be something that’s been spun up in house by you know, around a couple personas, or there might be some pilots with a couple, providers. And I think that’s something that we also provide, which is this uniform experience with, you know, all of the personas built out, the ability to then additionally build them on the underlying repository and so on. Absolutely. Go ahead. Yeah. Yeah. No. I I think it’s interesting. It’s also a thing you can sometimes the kind of discussion on this can be it’s either qual versus quant. So market has got the budget to do both. I imagine something like this actually is quite complementary, and they can actually get more bang for the buck as it were than they’ve ever been able to do before. Yeah. Definitely. So I think we started with, but also that’s what we see generally the customer base customer base doing. We started with the l the the chat based personas, and we’ve moved into the quant base. And we know from our customer base as well that they’ve introduced flows where they’ll come to the environment, speak to these LLM based, chat based personas to receive qualitative feedback, rework ideas, concepts, products, and then go to those paid quant tools. But we’re seeing more and more of a shift towards starting to also go to the synthetic quant offering in that flow, and they and they really do complement one another. So you can yeah. You even saw me do it in that chat. That that wasn’t just, a kind of environment, but you can really first get qualitative feedback from a couple key personas in the chat experience, rework it with them, take it away, rework it in house, then go out, get quant feedback, still look at the responses there, potentially rework it, etcetera, and then finally go to the the pay providers, Nielsen’s, the Qualtrics, etcetera for the, you know, for the final robust paid testing. It’s it’s fascinating stuff. I guess one of the kind of questions I imagine that many people have at home is what kind of data goes into these AI personas and panels? And I guess, moreover, how do we make sure that things like this are grounded in real insight rather than kind of AI guesswork and hallucinations, which people speak so much about with this kind of new technology? Yep. Yeah. Good question. So there’s a, I mean, there’s a huge breadth of data that goes into these personas that we see across customers, and and it’s not just one per customer. You know, it mixes and matches and so on. I think for the the standard personas, when we have a customer where their their persona is built out, right, and they exist in these decks and so on, they’re typically built on maybe three or four different sources. Market research documents, and so on, raw transcripts, so interviews with the customers themselves, usage and attitude survey work, so really, you know, large numbers of responses around, how customers have, consumed or enjoyed products and so on. And this is not always specific to our customer. In that case, it’s more of a broad based kind of surveying that’s done. And that may or may not have a demand spaces flavor to it. And then those are all then already mixed and matched and built by teams and finalized into these polished decks that we then bring to life in the persona environment. That’s the one piece there. And then we add this architecture to it where the large language model understands that kind of signed off on persona description, brings it to life with a degree of creativity. Because I think you you kinda see how I was chatting with the system there. Not everything I was asking it, like, reflect on this latest concept, gonna be in the underlying research from the past. So the persona needs to be or the system needs to be able to take the persona’s view and apply it to new novel ideas with the power of the large language model without just bringing in and hallucinating content, you know, from the wide Internet, let’s say. And And that’s kind of a key distinction that we built into the flow. When it comes to the quant piece that you saw there, there were I’d I’d say we more strongly are of of the position that it really needs to be built on more of a structured kind of representative understanding of the customer’s customer base. So what we’ve typically done to date is use usage and attitude survey data. Our customers will often have this, and we can then ensure that there’s representativeness along key demographic attributes like age, gender, etcetera, and that’s then captured in that panel. And then when we’re asking that panel or sampling from that panel and asking it to reflect on concepts, we’re ensuring that it actually matches what our customers would expect to see when they’re getting that large end response rate. And there’s the same idea. Once this panel is built and built on our customers’ proprietary data, it’s about getting the large the power of the large language model to reflect on all of the novel concepts and so on in a way that, is both somewhat creative but still locked in, if you will. And that’s just more of an art than a science that we’ve been working on for year plus. Yeah. Absolutely. I mean, that this entire AI revolution seems to have been a a a message to marketers and anyone to make sure the data is AI ready as it were, ready to go, ready to be pumped into these sort of different models. It’s such an opportunity, isn’t it, nowadays for marketers to really get their datasets in order, get their house in order so that these tools can be as effective as possible. Indeed. Indeed. Absolutely. I mean, you mentioned it kind of, a little early in one of your answers, the idea that, sort of traditional sodas kinda end up on use, that kind of old message of a a deck in a drawer somewhere that the insights team did. It was everyone was excited about it for ten minutes, and then it was forgotten to history afterwards. Why do so many personas end up in that state, and how can this tool start to really change that? Yeah. So that that’s definitely the case. So as we kind of started to talk to our customers about the chat based personas over the past year, a large percentage of the calls we’d have with, you know, existing trusted customers would be well, it looks great. We had personas. We built them in twenty twenty one, but there’s been no refresh. I I can’t remember where they are. Let me, see if I can find the email container. So you you just get the picture that, of course, they their time was spent on them, but they’re then not leveraged going forward. Of course, that’s not always the case. We also had customers where there was a kind of more disciplined regimented approach to ensuring that they’re indeed looking at their static personas, if you will, regularly and so on in their flows. But the general rule that they end up not really being overused that at least is is certainly true. I think there’s a couple of reasons. The first is they aren’t being refreshed, you know, just by the nature of the fact that they are in static presentations and so on. It takes a lot of work to manually rework, update them, and so on. And often, the research that underpins what goes into the personas is done only every three years, so a big survey effort or something. And it’s just simply not easy to be constantly updating them if they’re not digitalized, if you will, if they’re not living in an environment where new data can be pushed into them regularly. So that that’s point one. The other thing is just the very I mean, personas are kind of fun to play with. Unlike many knowledge management type tools, I would say, are just general feedback from customers as they enjoy working with them. It gives you this realistic feedback. So there’s just a something about the usability of them that people are much more likely to start using interactive chat based interaction with customer than they are to just read a static deck and try to loop it into their thinking. And that’s even if it’s not built into a workflow. So that’s the next thing. I think once these live in a system like ours that also contains other insights and marketing activities and use cases, what we see is just like this muscular memory of going to the personas in your marketing campaign flow, going to them in your innovation flows becomes more of a natural easy thing, and it starts to just reinforce itself. And we see this kind of positive feedback loop that it then really leverage much more. And that leads to them then being updated because there’s this kind of internal buzz around them. People wanna know, hey, insights team or whoever owned them originally. What are they being updated with? Are they being updated with, I don’t know, shopper data or some foresight data? And this conversation kicks off, and we can then help to ensure that they are updated. So it’s kind of this positive loop that they get more of a centralized place in all of these flows, and that’s what really reinforces them being used versus that more static living in a presentation somewhere. Yeah. Absolutely. I suppose as well, You obviously, you’ve worked quite a few different brands. It’d be great if we could get some kinda, like, some practical examples from you of sort of brand work on where we’ve seen decisions sped up or real positive outcomes at the back of this. What what can you maybe share with us that isn’t too confidential kind of work you’ve been doing? Yeah. Sure. So I think we we we have a wide, let’s say, swath of customers using them from for, I would say, all sorts of different use cases. We have a couple published case studies out there with, for instance, with Swiss Rail, with Philips, and other customers, and we see different different kind of activities. So somebody like a Swiss Rail has a smaller number of personas, but they are, really enabling the entire organization to go to these personas and test their, test their thinking with them down to things like, you know, considering how particular particular campaigns or activities at a given, like, train station, for instance, might work with this sort of set of idolized personas. So that’s that’s one case. In other customers, we’re seeing much more of a building in of the personas into innovation workflows or marketing workflows. And that can happen both in our tool with our innovation offering, which I haven’t really spoken about today, but also via, for instance, API calls. So our personas are also available via API calls. In in house innovation flows, we have more than one customer where they’ve got this full flow generating, new concepts, new ideas, new marketing, campaigns, and then they’re getting inputs from our personas in every single one of those in this routine way where a human user doesn’t even need to go over to the personas and chat with them. So they’re really looping this directly into their flows. And then we I think we have everything kind of in between. So we have cases where just not in a forced or not in a systematic way, customers are taking early stage concepts, bringing them over to the personas, reworking them, discarding, and then picking the winners from, you know, essentially via our tool that they’re then investing more and more time into. I think the quant piece is still earlier for us, but we are seeing very good feedback in terms of I would and it’s not the case that they are replacing real quant level respondent feedback. I think that’s also the case for the the persona chats. It’s rather that they suddenly allow just much more and much often and also much earlier interaction in those more traditional ways with these synthetic customers. So what we’re just seeing, instead of doing a final quant test, you can actually test all of your concepts and what you typically be paying for with these synthetics and just getting back the the feedback, reworking, and so on. Absolutely. I mean, sticking a little bit more into kind of practical adoption, where are you finding teams that kind of using these sort of tools first? Is it the kind of campaign planning stage, the testing stage, the strategy stage, a mix of all three perhaps? And what kind of changes in workflow are you starting to see once these tools are adopted and maybe some more practical examples, from your kind of case studies? Yeah. Sure. Yeah. So it’s definitely a mix. I wouldn’t say it’s just in, you know, first in campaigns or, you know, just in product early stage product ideas. It’s sort of across the board, and it could be at a particular customer for different teams in there. You know, marketers might be doing more of their campaign planning interaction in in our environment, and insights colleagues are doing more of an understanding of the underlying segmentations and so on. So we can take a look at the chats and see that kind of historical interaction. Yeah. So, yeah, I think that would kind of cover the the the different ways it’s being used. Yeah. Absolutely. And you can kinda mention there’s all these kind of synthetic panels for validation. I mean, testing and kind of getting things out there for feedback is so big now at the end of kind of campaign ideation. How should marketers think about this as perhaps another tool they can use to kinda have that confidence that what they’re doing is working? Well, I think in in a in an ideal world, I would love to see, and I we are seeing this more and more, but that marketers are enabled to themselves go to the synthetic quant outputs and qual outputs, but let’s talk about quant for a second, and actually feed in their their concepts or their packs or their anything that they would otherwise take to to the more robust, paid testing and get that feedback from our environment, and potentially start to understand comparing it with some of the traditional, like, historical outputs they’ve gotten back from the pay providers. Get that feel. I I said it before, that muscular memory of of this actually looks like it’s matching up with what we would likely get in a paid way. We can actually operationalize this without having to go out and wait for the test results back. We can start to shift, if you will, internally or shift to the tool some of the cases where we’d otherwise go outside and just become much faster, much more, yeah, agile in terms of being able to yourself serve these kind of tests, take the inputs from them, rework, and go from there. I think, though, that this will not replace the the the robust paid respondent work that can be done that is done with real respondents, absolutely, for a couple of reasons. I mean, also for us, we we we we’d love to see our customers continue doing it so that we’re getting the validation and getting the continual set of new tests. We’re then also able to compare against and keep that flow going, but also just because, you know, we are not in any way saying that it actually can replace that final human input at the end of the process before making those big decisions. It’s really about earlier on being able to ideate much, much more with the synthetics. Yeah. That kind of brings me on to my next question because, of course, many watching me, this sounds fantastic. It’s save me a lot of money. It’s gonna solve a lot of my research problems. But what are the kind of risks here of over rely on AI personas or synthetic panels instead of real research? Because it certainly seems that there still needs to be a balance between the two. Yep. I think now qual and quant aside, so just gen as a general point, many of the best practices around how we interact with or how insights and marketers interact with respondents still apply. And that is, first of all, with, you know, not making a decision based on the inputs of one or two people or personas. Right? So we don’t want to see this, and we’re seeing our customers do a lot of work towards preventing that this becomes an environment where you can just go get some validation quickly and then go make a decision. It’s really about expanding the new to new ways of thinking, new ways of testing, and so on. So I think that’s just a key point generally that we need to keep driving home, and we work with our customers also drive that home internally in their enablements and so on. Some of the other risks, yeah, certainly, I think that we we are freeing up the the end user to, so, you know, maybe the marketer who previously doesn’t get that, often experience speaking to respondents to interact with them. It’s a learning experience, for everyone, but I think the space is going there, with synthetic respondents, either way. So it’s rather about in a controlled way starting to build up that capability internally to, exactly, to prevent the risk. And I think the personas and the panel also go a large way towards, providing that versus what we know in any case is happening just in, like, large language models in ChatGPT. People go in there, chat with ChatGPT and ask it, you know, like, how it react to a concept, and you’re getting back much less control, just kind of wild inputs, whereas we can have a much more controlled environment, avoiding those hallucinations locked into the customer’s persona understanding or their panel, understanding and ensuring that those large language model based outputs are of this controlled nature. Absolutely. And before we move on to the, audience q and a, which we’ve had a flood of questions in, but do please keep sending them in. If you just had to kinda summarize the biggest shift here for marketers with everything you’ve spoken about so far, what would you say it is? I I yeah. I I would say that it’s, one, that this I mean, I spoke about the conference two weeks ago that I was at. Couple years ago, maybe two years ago, it was, are we gonna be using these AI tools or not? It was still in the air. That that ship has sailed. Right? So the these new tools are here. The synthetic capabilities are are certainly here. Insights teams are starting to use them more and more. And what they allow is essentially the the ability for you to interact with customers, albeit synthetic, but really based on what your organization knows about its customer base in a way that just simply wasn’t available a couple years ago. And I think that’s gonna unlock all sorts of new new inputs, new ways to early test concepts, campaigns, etcetera. And it’s about embracing it cautiously as a kind of stressed a few times, building up that capability as a team, but also as an individual to work with these new capabilities. And but, ultimately, you know, it’s it’s something that really offers a lot of new opportunities. It’s something right there, so I would, you know, look to it and embrace it. Absolutely. Well, thank you, Joe, for your time so far. Let’s delve into some audience questions. Like I say, we’ve got a lot to get through. So here’s one here, which I think is a very fair point. This sounds, like, a very powerful tool, but is it realistic for smaller organizations who perhaps don’t have large insight libraries or dedicated research teams? How will they get the best out of something like this? Yeah. That’s a good question. So it is a fair enough, I wouldn’t say challenge, but at least, consideration that, indeed, the large the larger organizations that we work with, they have this you know, typically, they have way too much paid market research data, survey results, and so on. And there, it’s more about, okay. How can we put the the key pieces, the and and weed out some of the still valuable research, but maybe not very applicable for mocking up what a respondent would say in front of our marketers and insights people. For for small organizations, the challenge is kind of what’s been summarized there, which is do we have enough data to build, robust personas? And I think there, that’s where our expert services team works with, customers because often, it’s more at the level of the the granularity of the persona that can be built that makes sense. So maybe you don’t have personas per market, per category, and then, like, five of each within that because you simply don’t have that level of data and understanding. But you you do actually have a very good understanding at the overall category market level. And there, we can, you know, work with you to build either a set of personas or the the builder capability that I showed to let your users prompt and build personas that are are only summarizing to the level of data that you actually contain. So ensuring that they’re actually then still data backed when you turn them over to users. Absolutely. Another one here. How do you validate that an AI persona genuinely reflects customer behavior? And isn’t just summarizing kinda historical biases that are already present in data. Seems a a core kind of issue to address. Yep. Again, a fair question. I think, we we’re seeing customers themselves do it with us, so we really are doing it with real, data. A classic way is to actually take new fresh interviews that literally don’t underpin anything in the persona and compare how the persona is responding versus how real real people are. And, actually, they were seeing very good results. And I think the fear that they they would just regurgitate, for lack of a better word, what’s in the underlying, research is on the surface. Yes. I understand it. But, actually, when you start to play with these guys and use them, there’s a a strong degree of creativity and novelness to what they can generate due to the power of the large language model. So we are actually unlocking all of that large language model capability, but we’re ensuring that it’s tied to the underlying persona description or understanding. That’s what lets it reflect on new novel concepts, ideas, trends, etcetera, that simply are not part of the underlying data. But it you know, for instance, events or trends hadn’t even occurred yet when the persona was built and put in that deck, but we’re still able to let the digital persona reflect on those concepts and so on. For the for the quant piece, it’s, yeah, it’s it’s a similar case, I would say, but there it’s even more so that what we’re really doing is letting these synthetics reflect on new concepts, new ideas. So there’s actually not gonna be necessarily past concepts that look like the thing that’s being shown now. It’s more the thinking process, for lack of a better word, that we’re trying to capture, but then setting it loose on new novel stuff. Absolutely. I mean, I suppose kinda going back to that point of research being kinda lost in a drawer. I’m interested as to and a question here based on what’s the kind of clearest business impact you’ve seen so far? Because the business impact is the most important thing. Is it the kind of faster decision making, better campaigns, reduced research costs and flow on those lines? Yeah. I think so it’s a good question. I think the most quantifiable one that we then are getting fed back to us from our customer base is around speed. So, I mean, besides the quality of in response to us or input back to us that, well, these actually work like our real customers think and talk. It’s like and now instead of something that would have taken three months, we literally get it in three three minutes from chatting with one of these personas. So it’s really that kind of time turnaround that’s been a huge impact. And then as a function of that, we can now so, you we can now use this in flows in a way that we previously couldn’t because we previously wouldn’t wait three months to get the input to loop it into something at the middle of our innovation or campaign thinking. Now we are doing that. And so it’s a function of the fact that there’s the quality there and then the speed, and then, you know, again, being told that it’s now part of processes that it just simply wasn’t before. Absolutely. A few questions here, which can maybe go through quickly rapid fire. What has gone into the building of synthetic personas? What is the real foundation, real world data? Yeah. So, again, it it it varies by customer, but the real world data would typically be raw respondent interviews with customers, usage attitude survey data, either raw or already summarized by, like, a customer team for us, so pulling out key trends and so on, and or additional content from the overarching insights repository, so all of the market research that a customer might have. We also have some integrations with our persona going out to providers like Next Atlas who are able to pull in, like, future thinking or foresight kind of oriented trend oriented data. But by and large, it’s the customer’s underlying survey and customer understanding data. Absolutely. And how long would one of these quantitative studies take to run on average? So I should stress that we we ourselves do not perform the quantitative studies. Rather, we are tapping into those that the customers already hold. I think, actually, it can be quite lengthy, you know, at least several months to to perform on the customer’s end. But then when once we get ahold of those materials, we can, you know, typically spin them into the persona set and turn them back over to the customer within a couple weeks, essentially, if not faster. Perfect. Another one here. Great product demo. What is the evidence that this is correct, high fidelity, useful, and what is that impact on efficiency or effectiveness kinda going back to that days ago? So yeah. I mean, good question. So, indeed, it’s it’s typically at a customer by customer level. We we have the process where first, some initial team of of users is vetting the personas that we’re building on behalf of their data by actually comparing it with real studies. And I I mentioned a couple different ways that we can do this. So they may have interviews, and they we can subject the synthetic personas to those very same interviews and then, you know, subjectively compare how well it actually matches up, and we’re getting, you know, very close feedback there. Same thing would go for having it. So if you as you saw in that chat reflect on a concept, we’re actually getting very similar outputs from other personas to real respondents. The quant piece is actually even easier to compare because there, you can really take past concept tests, run them by the quant, and then go look how they compare to what was brought back from the real paid tests, top two box percent, for instance, and see how well it stays within the bands. And there, again, we’re having, you know, very strong feedback. And I think that one’s even easier to, let’s say, prove out and tweak than the subjective personas. By and large, that’s our our end users. Our customer users are verifying that they talk and feel like real people. Absolutely. A question here. Have you worked with retail banks? And I’m asking an underlying part of that is the security concerns around this. How do we make sure that things stay within the platform and not leaking out into AI models? I personally so on the product side, I I don’t have an oversight of all of our customer base who are using the personas. I’m not sure if any banks are using the personas, but we do have, for instance, like, pharma customers with with with comparable, I would say, levels of oversight in terms of the evidence backing, the realism of the of all of our offerings, not just the personas, but also our more, you know, our our answer capabilities based on AI. And, yes, so we have multiples using personas and set up in a number of different ways, both built built on their existing persona offerings. Also, we’ve built some for, some pharma customers on raw transcripts, for instance, and, you know, very good, feedback from them as well. Absolutely. Another one here. How work intensive is it to build up the tool with data insights, and how often do you need to kinda feed it new insights to make sure it’s staying up to date and on trend? Yeah. So that’s a, that’s a good question. It goes beyond just the personas offering. So as I kind of alluded at the beginning, typically, customers will be using us not just for the personas, but also as a knowledge repository in a way to work through, and work with, their insights. We have a couple I mean, kind of a sliding scale of how long it takes to implement the overall, offering with us. And then we typically have customers, like, just regularly in an ongoing way adding in new research to the tool as they go. But if it’s more of a as we we kinda call it a stand alone use of the personas offering, then it can be set up very, very fast, I would say. So, like, I mentioned, like, two weeks there. Typically, it’s some consultation around which data makes sense to underpin the personas, and then we do much of the heavy lifting to kinda get it set up for you. Absolutely. A good question here. How do you work in those kind of more unhomogeneous markets like South Africa? We find they find at least all these AI solutions can be quite US or UK focused in their outlook, maybe not as broad as they’d like. Yep. So that’s I mean, it’s another good question. I I would say one key differentiator of of our personas, of the quant piece I showed you as well is we are largely built on our customers’ data. In other words, we’re not, for instance, giving you a drop down list of personas for, I don’t know, the US or Europe that we’ve built based on some research we’ve done. We’re actually taking your data, your understanding of your market, the research you’ve had, you have interviews you have, and turning those into specific personas for you. So we have very like, again, to to the question, quite niche persona sets built for certain customers in, like, you know, niche geographies and so on because it’s built on the data that they have and can provide to us. Actually, back to that health care provider that I spoke about, we actually set up a number of personas in South Africa, Indonesia, Brazil, Canada, and then the more, typical US, UK kind of Europe global splits. So it’s it’s more a function of the data you have as opposed to worrying that the the tool itself is gonna favor, like, the US or Europe. Absolutely. And kind of following on from that, another question here. You said there was a few hundred personas in the background of the quantitative studies. How are these personas created, and are they tailored to the company, or did they generally represent a more, like, general populace? Yeah. So, again, that’s correct. They are yes. They would they would typically be at least several hundred so that we have some kind of robustness there from a respondent level. And they’re set up on the customer’s data, in particular, usage and attitude survey data. So most of our customers have some survey work they’ve done, but that’s then captured in a structured way. We turn that into those hundreds of personas. Essentially, they then live in our system for you to reference them. We have not done this with a customer that, for instance, doesn’t have that survey data or some other kind of demographic understanding of their customer base. Something we could explore, but then we’d, indeed, as the person mentioned, need to ensure that it’s, like, representative of the target market and so on. Another question here. How well does the tool handle b to b scenarios where you’ve got those kinda multiple decision makers, very difficult priorities, and quite difficult to reach as well? Yep. So let’s say the tool handles it. The tool as the architecture and the interface handles it. That that’s not the the the major issue. In fact, we have customers again in the pharma space, also in, like, renovation and kind of home care space who have b two b persona set up. So you can imagine things like in the b two b setting, buyers or sellers, you know, at a at a large scale of, like, materials and so on. Or in the pharma health care space, medical groups, doctor groups, pharmacists, and so on who need to be targeted in in terms of, like, drug adoption. Right? So they can be set up. It’s more about the data that that you have. I think in those cases, we’re often seeing it’s based on interviews. So customers will have ten interviews with the back to the case I just mentioned, like, with the South African pharmacists, and we can then spin those up into a representative one or more personas that can be spoken to. Sometimes there’s also presentations and work actually done in a more formalized way in the b to b setting, and then we just bring those to life as well. Absolutely. A question here around the kind of tech reliability of the outputs and testing on them. How common are hallucinations? And is there a procedure in place where you can check the responses against data kinda, like, easily and excessively? Yeah. So that’s a good question, and this kind of comes up every time, I would say, with the the customer base. The first thing is I always like to stress the difference between, like, what you would term a hallucination in a in a more of a fact based or purely fact based answer bot. So we have our DeepSights environment, which I only alluded to very early on, that provides completely fact based answers based on our customers’ documents, integrated source, and so on. You can go there as something like, you know, what do we know about a given market in in a time period? And what happens is the system combs through all of the research and gives you back the synthesized answer tightly sourced. So, like, literally every statement made in that answer is sourced to the page, to a document. The personas are something, a little more free flowing than than that, as I kind of mentioned a few times, because they’re able to or they’re even built to provide that kind of creative feedback on things like concepts, ideas, and so on. So, indeed, they are tied to the underlying understanding of the persona, but it’s not like every statement that the persona makes can be found in the underlying data, just by by by nature of how they’re built. Right? So when it reflects on a new concept, it can’t say on page seventeen of, you know, this market research report, that’s why I say this I like this, color of this new concept. It it just doesn’t work that way. But we do, nonetheless, have, some capabilities. In particular, we’re releasing, a compare with research capability where what we do is once the persona chat has taken place, you can just click a button, and, essentially, that chat is now compared against the real market researches out there. So not just what under like, the persona per se, but the overall market research repository. And you get back this report about how much of what the persona said is at least found in the underlying data. So the same way you might take a respondent interview that’s really been done and go check, okay. What do we actually know about that in the real research? We’re just bringing that, you know, forward and automating that as part of the process. Fantastic. And we might just be able to squeeze in one final question in the air. One panel research. You had twenty personas there. How many segments did they represent? How much I imagine they’re asking here is flexibility is there to tailor this on how you need it? So, mean, that was just a mocked up, set of twenty, that I showed. Tip so what we’re seeing is customers are typically building this, let’s say, per geography and per category or kinda like mega category for them based on, like, ten thousand u and a survey results. And then we’re boiling that down into something like five, six hundred panelists that can then be sampled from. So they’re not per se tied to a formalized persona that’s exist exist in the organization. They’re really individuals that are based on the customer’s underlying u and a survey data. Of course, they would fit into, let’s say, I don’t know, ten different segmentation cuts could be assigned to each of the the six hundred in the panel, but that’s less important about how they respond than their individual attributes and what we know about them, I would say. Absolutely. Well, that is about all we have time for. We still had plenty more questions which came in, so clearly, you’ve struck a chord. I’m you hope everyone found this discussion useful. It will be available on demand once the session is over. So I want to thank you again, Joe, for your insights today and to everyone who has joined us. I’d also like to thank Market Logic once again for partnering with us. Thank you for watching. I do hope you’ll join us again soon for another marketing week webinar. Thank you so much. Thank you so much. Great. And thanks to the audience.