Hello, everyone, and welcome to today’s webinar, AI or human research, how Vodafone is finding the right balance. My name is Kelly, and I would just like to begin with a brief introduction of today’s session as well housekeeping. So first things first, this webinar is being recorded, and the recording will be shared with all registrants tomorrow if you’d like to rewatch it or share it on demand. And if this is your first webinar with us, I’d just like to direct your attention over to the right hand side of the screen where all the controls are. Most importantly, you will see the questions and polls tab. Please feel free to ask questions, to our speakers throughout as we will be having a live q and a at towards the end of the session. Additionally, we will be pulling the audience throughout, and we welcome your participation in those polls as well. It’s always good to get a sense of where we are as an industry in addition to hearing the individual perspectives of the speakers. And speaking of speakers, I am very happy to welcome today Market Logic’s own CMO, Daniela Zwin, as well as our very special guest speaker, Sarah Emblin from Vodafone. As today’s session title suggests, the two of them will be discussing how Vodafone employs AI in their research, their goals, the guardrails, use cases, etcetera. Balancing human and AI research is certainly not a new topic, But as new applications become more commonplace, such as synthetic personas, for instance, this conversation becomes more nuanced and more important. So I’m very much looking forward to hearing what they have to share with us today. And I think that’s it from me. I will hand things over to Daniella now to kick off the interview and I will see you guys at the end for the Q and A. Thank you all for being here and Daniella you can take it away. Thank you Kelly And hello to everybody. Thank you very much for joining us on this webinar. As Kelly mentioned, my name is Daniella Zwyn. I’m the Chief Marketing Officer here at Market Logic. I have had the privilege of joining at a time when there’s been a complete explosion of innovation, both in terms of the platform that we offer, but very excitingly, and that’s what we’re gonna be talking about today. The work that our customers are doing, and how they’re applying all of this innovative technology in the real world. So, Sarah, thank you very much for joining us. It would be great if you could just give us, first of all, a little introduction as your role and where you fit in the Vodafone organization. Thank you. And thank you for inviting me to speak today. Yeah, so I’m Sarah. I work in the Vodafone Group Commercial Insight team. So our role for the organization is to serve our 15 markets with a combination of always on measurement programs, looking at CX brand and campaign measurement. Our other remit is strategic research projects, which serve the whole of our group commercial teams, looking at new product innovation, new propositions we might bring to market. And then my specific role also encompasses AI research and innovation, which I guess is why I’m here today. Great, thank you very much. Yes, and you’ve certainly been really leading and pioneering the use of AI, which we’re going to be learning more about. Now, one of those innovations in particular is the use of virtual personas, and that’s what we’re going to be exploring today in terms of the role that they are now playing in your whole research program. We had a poll, and Kelly, maybe you could bring that poll up just before we start, just to get a sense in the audience of where other people are in terms of using AI and applying that into their consumer insights programs. So, what I see there is, yeah, quite a few people who are certainly already along the field, but quite a few that have yet to begin, and that’s great. The purpose of today’s session is to give you some great best practices, so that you can follow in the good footsteps. Okay, thanks. Kelly, if you remove the poll now. Okay, so, as you mentioned, Sarah, you know, part of your re med is also that whole introduction of AI into the research process. And before we delve into personas specifically, perhaps could you share with us a little bit of where you have currently got to in terms of bringing AI into your function? Yeah, sure. So I’m sure it’s the case for many others as well. We’ve actually been on the AI journey for a little while now, at least two to three years, would say, particularly in terms of how we’re managing our research and deployment and insights of impact on the business. And it’s been quite an iterative process. I would have probably selected all of those four boxes on the poll because I think we are piloting, we’re testing, we’re adopting, we’re scaling, we’re doing lots of different… Taking different approaches with different aspects of what AI offers us today. We didn’t start with a big ambition to transform research overnight. We’ve looked at applying it where it solves real problems, if you like, with the research process, which we may not have identified before, but now AI is in front of us. We can see that it can plug some really useful gaps for us and genuinely add value. Great. I think one of the struggles that we’ve had is navigating what AI can do for us as a research organization. So we have got a structure, which I think we have a slide too. Yeah. As if by magic, thank you. So I think client side, it’s quite difficult to navigate all the agencies offerings and all the capabilities that AI gives us today. So the structure in the way that we’re looking at the world for AI for client side research, we’re looking at it with this structure. It gives us a mental model to navigate the AI landscape. So on the one hand, we’re utilizing AI to support discovery, and we think that’s where a lot of the capabilities lie. And that’s obviously where Market Logic and some other partners we work with help us. And then over on the right hand side, we’re looking at the market in terms of how AI can help us as the respondent. And we’ll come and talk, I guess, about whether they should replace or not. But yeah, what’s their role in that part of the landscape? Okay, okay, great, thank you. And as you’ve started to incorporate that, how has that changed the way the actual team around you works, the insights team specifically? Yeah, I think the biggest area for us as a team has been so far on the left hand side of that slide. So insight generation side and obviously our platform that we have with Market Logic, which we call Vic, it’s Vodafone Insight Centre. Even before it was AI enabled, became a huge valuable source, pulling together all of our research into one place so that we could really sweat that knowledge asset. More recently, it’s enabled us as a team to have greater impact and reach across the organisation. So typically as a client side team, quite a lot of your resource sometimes can be taken up asking quite fly by or ad hoc questions by stakeholders. That’s completely gone now. We don’t have to do that because they have that at their fingertips. So we’ve really democratized the access to insight at Vodafone. There’s less duplication, I would say as well. So it’s really freeing up a lot of time for us increasing our efficiency, think for the business as a team. Is that change, that democratization, has that really changed if you like expectations around your stakeholders in the business in terms of how quickly they can get to everything that they need? Yeah, I think AI is doing that in general for all of us, isn’t it? We expect the answer immediately. And I wonder what situation we would be in if we didn’t have our platform AI enabled, because I think it came along at exactly the right time when we’re all now in that mental model of asking a question and getting an answer immediately. And of course, what our platform and similar ones that many others are using. Make sure that the information that’s being provided is generated from your own proprietary research, is relevant for your category, is traceable. And the important thing is that we’re giving insights that the business leaders can make decisions on. So it’s quite a brave move to make, to put the insight in the hands of the business, because you never quite know how it’s going to be used, but we’ve not had any problems so far. That’s good. So, yeah. That’s good, great. And now onto digital personas, which is one of the areas here so that we can drill into it. And Kelly, I think you can remove the slide now. The exploration into digital personas, you know, as you started to look at that area, because it’s a very new area, what was, you know, what were your goals? What were you trying to, you know, establish whether or not they would be useful? I’d say initially with the AI in general, well as digital personas, it was more curiosity than anything else. I think sometimes even with that, in our industry at Vodafone and in the tech sector, sometimes you’re asking yourself this question, is tech leading or is it consumer leading? And I think this is also something that you can evidence with AI in our research industry as well. I think the tech came along and then it was more a case of like, well, how can we actually, I’m curious about this. What can it give us? How quickly AI was developing and what that could actually mean for us. Okay. So the actual building of the personas, it established you wanted to explore them. Can you take us through a little bit of how you got started in terms of how are you going to create them? What you’ve put into them? Yeah. So we’ve used a cross market segmentation study to build our personas. So at the moment, we don’t have a huge number of personas across the business. We have six that we have recently made accessible outside of the insight community. And three of those were built from an existing segmentation study. And in a group sort of role, one of the challenges we have with anything synthetic is that we’re trying to represent the world or at least the 15 countries that Vodafone operates across. And the three, think we spent probably about a year validating and testing to see whether they were giving us the kind of responses that we would expect from the human seed studies that we had used to see the personas. So yes, from our own proprietary research. And then the other three, actually created dynamically from all of the information that’s held on our insight platform. So, which was one of the, I thought really exciting facilities that the personas gave us. Two very different approaches then. With very established data and one, well, is still using the same data, but in a very different model. What have you seen in terms of the difference between perhaps, how dependable, what is the difference in the answers? Have you seen much disparity between the two groups? Between, yeah, we’ve done a lot of human versus digital persona testing, sometimes deliberately and sometimes by stealth. So if there is qualitative research happening, we will behind the scenes test whether we get the same result from personas. It’s surprisingly close, which gives us a lot of trust in what we’re doing, but not universally. So, and I think that’s one of the key learnings is that I think it’s what we’ve learned from doing this is that we can’t say personas are a good tool, full stop. They’re a good tool for certain questions. And I think that’s the key distinguishing aspect that we’ve learned. So what we see where the differences occur are when we’re talking about recent customer experience, emotions, things that like CX and brand and those kinds of questions that we might want to evaluate. They’re a little bit more limited than that because there’s not recent context, emotion and recency aren’t necessarily evident. They’re not quite as reliable for that, but they’re excellent for deep diving into certain propositions, thought processes around them, believe it or not, we’ve been really surprised that where we’ve seen some themes and direction and the richness that the personas give us when we’re looking at new positioning areas for certain products, for example. And I was just going say sorry, that we’ve also seen where we’ve tested things, which I think I’ll come and talk to a bit more, like just preference in terms of naming, where we have gone into human testing, we’ve seen actually the same signals. And when you first started to test them, was that very much, you know, one on one conversations that you and your team were having, or did you test them, you know, with group conversations? How did you approach the testing before you put them into some of the use cases that we’re going to be taking a look at? Yeah, I think that’s one of the learns actually, it’s quite difficult to just go into a test phase. We found it anyway, maybe others don’t, but we found it quite difficult to go into a test phase of digital personas without a real case to test them with. So we looked at past research, posed them questions that we knew the answers from past research. We did that in a group setting, also one to one on the platform. And yeah, that was where we started to build some evidence that, they’re actually giving us what we would sort of expect in terms of their output. I think one thing I would say is that, the depth is very system to the kind of you get great depth. But I think we recognize, and this is what we’re all learning, guess, about AI and how we can use it in research. In terms of acting like humans, they’re not, thank goodness, they’re not doing that. I think that’s a strange world. And we may live in that world at some point. But in terms of predicting like how someone feels on that day, what mood they’re in, their emotion and what might be their decision that day. That’s a stretch too far, I think, for digital personas. So at the moment, that’s the testing that we’ve done really against concepts that we tested in the past. We know not to stretch that too far. Right, right. Yeah, so it’s the kind of questions that you think they’re reliable on is something that is emerging part your rollout. Should we maybe take a look at some of those use cases? I know that you’ve got some information. It’s, yeah. Next slide. That’s it. Yeah. So yeah, I think it’s nice to give some tangible examples of how these things are being used. Otherwise they feel very abstract, don’t they? But yeah, so as a research team, I mean, so far we’ve talked more about how the stakeholder and how the business gets value out of it. Obviously, as a research team, we’ve really sort of only fairly recently, to be honest with you, got quite excited about how we can use them to refine stimulus. So we have a loyalty app. We were looking recently at some new potential sort of gaming options for our app, which create a lot of stickiness in our My Vodafone app. And we’re going into human research to look at the preference and some of the journeys, the suitability of certain journeys across personas. But what we’re able to do as a research team is identify any frictions in stimulus before I even went into human research. So because we can put creative in front of the digital personas, which is, I don’t think something we expect to be able do so soon, but we can. And again, the feedback they gave us were feedback as the research team could see in the stimulus. So it gave us the opportunity to prioritise the questions for human research. We could make sure that we’re sort of ironing out those difficulties in the stimulus that might have taken the conversation off in a direction we didn’t want to get to in the human research. And in some cases, it’s even brought up some questions that we hadn’t thought of that we can validate when we go into the human research study. I think that’s a really interesting use case. It’s not one in particular that I’d heard of before. So is that, you know, time saving, I guess, in terms of the end to end research process, because you’re getting closer to how you want the stimulus to be put together? It does save time. It also saves and it can cut through those internal conversations that can happen sometimes about whether we as a business think something works well or makes sense to us, is not something that naturally could make sense to a consumer. And actually it cuts down those discussion times for sometimes on client side, can have conversations around, well, I think this works, I think this doesn’t. And it really helps. Even though as researchers, we can have that input, it has far more weight if we’re able to say we put it through the digital personas and there’s this consistent tension point coming out from the stimulus that we need to resolve before we put it in front of humans, as we now say. So yeah, it’s really super helpful for that. So it helps to make the whole process more efficient get through. Okay, great. And you’ve got some other ones as well. Yeah, the second slide. Yeah, so the second use case is I touched on earlier is shortlisting names for human testing. So we’re often testing new names for propositions or new products. And quite often we’re faced with a very long list of names to test, which is very expensive. And sometimes a short list can be created with quite subjective input from the business. So what we’ve been using the digital personas for, and I would say this is one of the sort of most initial uses for it when we were first rolling it out, was putting a longer list of names through the personas to get a clear rationale for the shortlist. So, and it wasn’t just what’s your preferred name, it was about attributes that that name like communicated to you, whether it fit with what we were trying to achieve. And then we can obviously look at differences across personas as well. So we got a much more transparent and objective shortlisting process. We also iterated a few times, so we didn’t get a shortlist go straight into human research. It threw up some questions for us where we were like, actually, let’s go back and think about how we can change that name. Can we make it more lifestyle orientated or does it need to be more technical, for example? And then the human research becomes the validation stage. And as I said there as well, shortens the internal discussion again. And I would say overall, we’ve tested more than 50 What this is really doing for us is bringing research into the conversation where it didn’t even exist before really. So, Sarah, you dropped ever so slightly for me. You said we’ve tested 50, is it 50 concepts that you’ve tested? Yeah, 50 more than we would have done before. Right. So yeah. Okay. And do you feel that the names and, you know, the ideas that you’ve incorporated into them, you’ve taken forward, do you feel there’s been a bit of a shift? You’ve been learning something new, exploring maybe new avenues because of this input? Yes, definitely. Yeah, I can’t go into loads of detail because I’ll give too much away, obviously. But I guess the best example I can give is that sometimes in our industry, we can be quite technical and functional in our naming and in our descriptions. And we’ve seen the personas actually come back and make suggestions, not name suggestions, obviously, but a different positioning for the names, I said earlier, being more lifestyle and more emotional perhaps. And it just raises that conversation internally as to whether that’s something we should consider and go forward with. Okay, great. And then I think you’ve got a third use case, which I was really interested in, if you could take us through that one. Yeah, we have one of the things actually, just as a sidebar a little bit, when we were initially testing the digital personas is how agreeable they would be. We all know, I think when we ask AI a question, even Claude, I think, who’s known for being maybe not so much now, but a month ago, was known for being a little bit more discerning. One of the concerns we had is that they would be, or the AI kind of augmentation behind it would still give us a more agreeable output. And we were quite keen to make sure, and for anyone that works in research and sees actual consumer feedback, they can be super negative sometimes. And we were a bit concerned whether the digital personas would do that. I mean, we know if we ask any question to AI, it tells us, first of all, how amazing our question was. So it’s very agreeable in general. So we’ve seen evidence in what we’ve done in terms of early stage concept testing that it has actually killed a concept or two. So we have had real tangible action where we have not continued with a concept, because there’s been unanimous feedback from the digital personas. Now we only have six, we have slightly more than that in testing in the research team. And when we put it across all of them, no one disagreed with the fact that one or two of them was such a terrible idea. So I think that’s great. That means that we’ve got much lower investment risk. It means for our product teams, the concept is killed earlier in the cycle rather than further down the process when maybe there’s been more commitment to it. There’s been a business case drawn up, etcetera. The other aspect to it, is really good, is people are more creative with their ideas. They’re a bit braver with their ideas, because there’s less risk. Yeah. But did you also have to be very brave though, because in terms of persuading the stakeholders, look, we’ve run this test, and this is really not hitting the points that we need to, in terms of us backing, taking it forwards. That must have been quite a tricky conversation to have the very first time. Yeah, yeah. But I think if you work in research, get used to those conversations. But I think when it comes from a digital persona, yeah, was quite surprised. In those two cases, there was no particular pushback, even though it was one of the first few times we’d used the digital personas. Perhaps it was just, it was so unanimous. Perhaps if we’d had some agree, some disagree, we might have had a more difficult conversation. But yeah, I think the fact that it was so unanimous was pretty complete, I think, in terms of the discussion going forward. But really, really interesting to see that you’re able to use them to find out where it says no, not just, let’s say yes, and to what degree are you going to say yes to something. So, quite a different range, different spread of use cases that you’ve talked us through there. Just to, you know, step back. When you look across them all, just to look at, you know, those early learnings. You’ve been deploying them, you’ve got them in these different modes. What would you step back and say has changed in terms of how you’ve learned to use them and deploy them? We’ve definitely been able to, like I say, bring the consumer voice earlier in the development process, into more decisions that are being made by the business. We are aiming for a greater speed and rhythm so that we can, in more cases than we’re maybe currently doing going forward, that when we’re creating the human research projects, we’re in the development stage of actually putting something into what we could now call human research. We are arriving with sort of stronger stimulus, less assumptions, better developed questions. So that’s… And I think that’s an important point that what we’re discovering is that… I wouldn’t say AI in general, but definitely with digital personas, it’s not really replacing what we would have done before. It’s making what we did before run better, more efficiently. Yeah, that we’re getting more out of it so far. And how do you see that changing, you know, how your business stakeholders start to plan their processes, and when they would be engaging you across these cycles. Yeah, think that’s interesting. I think people take, probably AI is one of the, I guess, most recent examples of where human behavior has changed quite dramatically in a short period of time. But I think in general, there’s still a more traditional way of approaching research. What we’re having to make sure is that we are in the conversation earlier on in the phase, making people aware that we can do this, making the business more comfortable with learning and iterating earlier, that they can take more risks with some of their ideation. And they don’t have to wait till they have a fully formed proposition to come and talk to the research team about doing some testing. So that’s really about us making the business aware. There are different levels of comfort with AI for research, and that depends very much on the stakeholder. And I think that’s what we need to use our people skills to understand that we’re not salespeople for AI, but you know, if we really think AI has a role to play for a stakeholder that we recognise and tune into how we need to talk to it, talk to that stakeholder about how we can add impact. So there are some stakeholders that are more comfortable than others. And we just need to recognise where they are on that journey and sort of help them along, like AI counselors. Not quite. And in terms of those teams, it about getting them to know when to engage you, or are you also giving some of these other teams direct access to the personas? Yes, some of the teams we are. Haven’t gone out and done a big launch across the business. We are doing it in baby steps and we’re acting as kind of a consultant to that process. We’re finding that some stakeholders are more comfortable than others, not just with the idea of it, but the actual doing of it as well themselves. So yeah, they’re the main things really. Which groups then have started to get access? Where did you start to let them out? So in a group context, we’re a really flat structure. So we’ve got the brand, the global brand team, some people who work in CX, and we are launching next week to products, marketing and digital. And that sounds like huge parts of the organisation, but a group that’s maybe 30 to 40 people. And I think that’s manageable, that’s just about manageable, so yeah, at the moment. Do you have like a group training session to tell them exactly how to use them, or how do you actually roll it out to those people? So far we haven’t got that like really structured. There’s not like a certain journey people go on with. Like I say, we’re adapting ourselves to the teams. Across our insight team, we all work with different stakeholders and we’re kind of working out what approach we need with each of them. But what I am finding out is they do appreciate some help. So I think maybe six months ago and we were like, everybody wants these, they’re all going to be researched because they’re very seductive things, digital personas. It’s like AI, you can ask it anything and they will give you an answer. So we’ve been really cognizant of the fact that we need to give really clear guardrails to what you should be using in their output. And there are, yes, so there are some that… But what we found actually, sorry, since we’ve launched them is that people are a little bit nervous about using them, which I don’t think we expected. And they’re coming to us for some help, how they should ask questions, what order they should do things, all the kind of research methodology approaches that any of us have worked in research for years. It’s actually nice to see that there’s some respect for that. And actually amongst our stakeholder groups, they’re asking us these questions. Right. So they’re actually learning, not just digitally, where is it, but how do I get the most out of it? What about the local teams? Because for Vodafone, the regional teams are really important. Have you gone as far as taking it out into any regions? Yeah, again, baby steps. We haven’t gone across our 15 countries and done a big launch yet, but we have little hotspots of usage, I think I would refer to it as. So we have users in Germany and the UX researchers in our German team have been using, they have largely been side by side testing at the moment. They want to test for a little bit longer before they’re really sure to use that as their sort of central sort of tool. And we’re talking to our team in Czech at the moment who are really interested in using them. So it’s quite a thing to manage rolling out digital personas. So we’re kind of reacting to who’s coming to us at the moment rather than doing a big push out. But I think we will next financial year, think that’s probably where we’ll go. Right. And when you do take it out into the markets, said you’ve got some central personas. So presumably, you know, part of the attraction is being able to localize that. What’s involved in creating that localized persona? Yeah, that’s some of the challenges we have ahead really. I haven’t got kind of a model for how we do that at the moment. The challenges for us are having the right seed data in the local markets, what’s available, what use they’re going to have for them, because the use for UX versus a broader digital persona for Albania versus Czech Republic. They’re very different kind of animals, if you like. So yeah, I can’t give you a sort of a best practice for that at all at the moment. We’re to find out, I think fairly soon, but we will look to build bottom up rather than top down, I think, with our local markets. Right, right. Yeah, no, it’s a fascinating opportunity. Vodafone is a huge enterprise. I think everything that you’ve been speaking about at the moment has been very much on the consumer side. But you also have very substantial B2B business. Do personas have a role to play there? Yeah, we’re starting to explore that. I can’t take any credit for that side of the work because I have a… There’s a B2B insight team, colleagues in that team. I think what we recognize now that it does bring quite a bit of additional complexity. So with consumers, can build personas around attitudes, technology engagement. With B2B, it’s more about who that persona represents. A small business owner can be very different for someone who’s buying in an enterprise level, depends on the category they’re in. And you can have buyers versus decision makers versus influencers. So yeah, that’s not on my lap, it’s in my colleague’s lap at the moment, but we’re working together and we have a really close working relationship in terms of what we’re learning. Guardrails are the same, etcetera. But yeah, they’re just entering into that world right now. But definitely a space for them to contribute. Although they need to be modelled differently. We’re going to take a look at that balance, the balance between human and the digital persona. I know you’ve got a slide on that, but just before we get to that, Kelly, would you just throw up the poll there around the audience and where they feel the balance is? So, essentially, is it is it going to be more human dominated with some persona on top, fiftyfifty, or more synthetic with shrunk down portion of human research? Interesting to see where the the poll lands, I think. Mostly people who want to use it as a top up. Oh, we’ve got a few few people maybe coming on board at the bottom. Okay, so mostly still a top up area. Obviously, it’s a very fast evolving balance, but now maybe we can go to your analysis and you can talk us through how you see the two combining, Sarah. Yeah, sure. So I think I would have voted with the majority there. So I think, as I said at the beginning, that the way we see AI impacting our, the work that we do at Vodafone is very much at the beginning stage of the process. So as I’ve said, we’re able to leverage what we already know using AI with the platform that we have, obviously. We can do faster discovery, exploration, early guidance, advance or not concepts using AI. Our advice to our business at the moment, because the senior level stakeholder pressure for moving towards AI is great in our business. We are at the moment making the case very heavily that human evidence is absolutely essential, particularly the closer you get to launching something, evaluating and feeding back and measuring your brand equity and tracking. So anything where context, as it says on slide motion, memory or precise measurement matters in all of our testing so far, and we’re going to continue to test this actively. We absolutely need human evidence and the AI will only ever be as good as the human evidence that’s seeding it. So I don’t think humans are defunct anytime soon, particularly not in the world of research. So, and in fairness, this should probably be more of a 30, 70% split, but just for the design of the slide, it looks a little bit more fiftyfifty. I think that’s an important point. If you could elaborate on it in terms of how, you know, how the human research actually is a direct way of fueling your personas and how they connect together. Yeah. So we want to see obviously what we say AI gives us an excellent picture of yesterday, really. And I think any of us in research teams feel very strongly that if we’re not feeding that with what we see happening in the market today, the wider world, the wider context, news, events, anything that happens over the last, you know, if we think about what’s been going on in the world in the last month or two, how that might change people’s perceptions, choices, which may not be reflected if you had an entirely AI source for that. There’s all the statistical reasons for not doing that as well, which I’m less versed on, but it’s pretty critical that we do that. And that’s part, I think, of keeping these digital personas fresh is having that broader context, which I think we’ll be looking at more going forward. It’s not just our research, but also, as you touched on, Daniella, the wider landscape and what signals are being sort of reacted to in the marketplace that might change perceptions and feelings about things going forward. So how often will that refresh need to take place? Is it going to be kind of a systematic annual or is it just iterative as new things come on board? Again, we don’t have hard and fast rules. It so happens with the personas that we’ve developed, there is an annual segmentation study that is done to support that. So that will get refreshed every year. Those personas will get refreshed every year. I think, yeah, so that’s the ideal, we would do that at least every year. Think that the difficulty is managing the expectations from the business that AI will save money. And I think it does in some ways, but those big cost heavy projects, UNA studies segmentation, which typically are the ones that are good for seeding either synthetic panels or digital personas are still a huge cost to the business, but are still required. So I think that’s the balancing act at the moment that I think for any of us that are under some budget constraints in the world of AI and the expectation that AI can save lots of money, I don’t really see that as a direct budget saving just yet. So I think that’s one of the key things to manage this human AI interaction. Do you have an opportunity though to perhaps look at how you’re using all of the AI to then illustrate how much more you’re able to use the segmentation studies? Is that a viable way of calculating value or showing value? Yeah, and that’s one of the things we’re trying to evaluate at the moment, that it means doing more rather than replacing what we’re doing today. Not in every case, but in a lot of cases. And that’s the conversation we need to have with the people who hold the budgets and want to see decreases in research spend. Great. It’s a bit early for that. And I guess, just to wrap up before we, because I can see we’ve got stacks of questions here, I don’t want to prevent us from answering those. You were stepping back now, Kelly, you can take the slide off, I think. If you’re, if you’re speaking to somebody, you know, a peer who is going down this path, do you have, you know, a few kind of like key takeaway points of advice around how to approach introducing personas, you know, in any of the areas that we’ve discussed. Yeah, and I guess these are learnings from… I’m not saying things we did wrong, but things we might go back and do differently. And I think that’s where you learn the most from people probably. Is think about the seed data that you take into your personas. Like, do you have the right information to be able to develop the personas? Sometimes you use what’s available rather than what would be ideal. I’m probably revealing way too much there. So yeah, the suitability of the seed data, think, is something that would be critical to think about. Obviously test before you trust. Our approach with testing was not to test to get… To see if they were good, it was actually to stress test them, to see what they couldn’t answer, to make sure they gave us negative responses. And we deliberately look for failure as well as success to establish where they are or aren’t reliable. So yeah, I would definitely go into it with a healthy sort of injection of scepticism. And I think the other aspect I touched on earlier really is like not to confuse the fact that they can answer every question that they should be asking every question, that they should answer, sorry, that they should be answering every question. So… And I think of that as, don’t confuse access with their capability. And I think it’s really important you invest in the training guardrails and make sure people really understand where they add value and what decisions they can support. Right, right. So those limitations are really important in terms of making sure you’re comfortable with how they’re being implied. Yeah, I mean, they’re very seductive things to have on your things, sorry, personas to have on your laptop. Yeah. And you could ask them anything, but it’s just making sure that you’re asking them what they’re capable of answering. Right, right, okay. Good, that’s helpful. We’ll certainly be writing up some notes with your, with your learnings here, which we’ve, we’ve really enjoyed hearing about. Kelly, I can see we’ve got a lot of questions, and I do wanna make sure that we give Sarah some time to answer them. So if you would perhaps… Yeah. I will go through the q and a now. We’ll try to get through as much as we can. I will say there are some questions that touch on the same topic. So if I inelegantly mash two questions together, I will try to, you know, communicate that as best as possible. I’m gonna start from the top. So you talked about, you know, annually refreshing some of the data behind the personas, but I’m wondering if you’re seeing any… If you have any concerns so far about the aging of synthetic personas or if you see them changing over time. I know you are… You know, this is a fairly new product, but have you witnessed anything that, you know, is either a concern or does it seem to be keeping up with the new data you’re putting in at this stage? That, I mean, in the human studies, we are seeing our personas change, make some subtle changes over time. In particular, like for example, in their use of content and streaming across devices, let’s say that’s a topic that’s definitely changed over the last two or three years. And because we feed the personas with that data, then that comes through in the persona, the digital persona. So it’s critical that as that new information comes through, you have a process for updating them. So yeah, I’ve only got one example of doing that, but yeah, that’s not been a concern so far, but we’re watching it. I’m not assuming it won’t be. Yeah. Yeah. Again, it’s all very new. We… You just were talking about the guardrails, but I’m wondering if you are able to go into any more detail about that, specifically when your non research and insights colleagues are using personas? Who within your team is responsible for communicating how to use them? What is a good question for a persona? Which one is it? Etcetera. Again, we don’t know to go into a lot of detail, but if there’s anything more you can share on that. Yeah. So I’m largely responsible for that. And my colleagues on the b two b team, they will be responsible for that. I mean, put simply, we’ve kind of got a list of things you should do and you shouldn’t do. We’re making people, making sure, we’re hoping that we’re making sure people understand that this persona isn’t a customer, that it’s a consumer perspective. It doesn’t mean that everything should be treated as fact. And the other thing we’re asking stakeholders to do is in that they declare that this is a digital persona output when they are taking it into their internal presentations. So there’s absolute transparency. So, yeah. I think that nicely encapsulates the evolving role of the insights team that you’re not just guardians of the knowledge, but also you know, guardians of the… Of how to use it and communicators of those guardrails. So that’s very cool to see. Okay. We did have a question about… I I know you’re not fully on the b two b side, but are you seeing any difference in the reliability so far between AI for b to b versus b to c research? I can speak to that a little bit. My colleagues have tested two personas so far and the feedback from them is that they actually had good reliable responses. So there’s no indications of that, but we are in early stage testing. But so far with two personas, it looks quite positive. Like I say, the big complication there is representing the vast array of B2B segments. But yeah, alignment so far. Great. Yeah. I mean, that might be a whole another webinar that we do, you know, sometime next year when there’s there’s more to to share on that topic, but it’s very cool to see you testing it in the b to b space. I think that’s… You’re at the forefront of that. Okay. There are some questions. I think this goes more into… I’m gonna add this to the stage. Purpose built AI like our platform DeepSights versus the more general LLMs like Claude or Copilot or whatnot. I don’t actually know if Vodafone has any of these generalized LLMs deployed within insights or if you’re using that alongside a more bespoke tool like DeepSights. Is that part of your strategy so far, or can you speak more on that? Yeah, we’re active users of Copilot at Vodafone. I mean, we have over 70,000 users of Copilot. I’m a Copilot champion for the organization. That’s been a key part of our watching brief is to I mean, we’ve actually created digital personas in Copilot. We’ve done comparisons between what we see in the Vic digital personas, what we create in Copilot. We’ve had many, many discussions internally about both about the governance of personas, where they should sit, who should be creating them, how they’re kind of nurtured and refreshed and rehydrated versus Copilot. So that’s another webinar entirely, and I’m happy to have a conversation one to one. I think that might be Andy Dexter. But… Yeah. Let’s… But but… Yeah. It’s not off our radar. It’s on our radar and a very lively discussion we’re having internally at the moment. Yeah. I think as as we’re moving into the phase where it’s more about how the AIs are talking to each other and and how you can ensure that the correct information is shared between them. So I agree that that is its own webinar for another day. Okay, I think we might be running out of time, Flash. I’m having trouble keeping up with all the questions as they are continually being added through. Is there anyone that you want to make sure is answered, didn’t you? I did actually just wanna come back to what you were talking about in terms of seed data, and just maybe if you can just elaborate on that a bit more. And especially as, you know, technologies are evolving, you know, what kind of data you think is gonna be really important to consider getting into the virtual personas, and to keep them as forward looking as possible as you move forward. With digital personas, I think it comes down to, obviously, it’d be a very different kind of I think in general with… Did I cut out then? Sorry. I think in general, we’ve been looking at this more from a synthetic panel perspective, and we know that usage and attitude or segmentation studies are… I say I know, I shouldn’t say that in the world of AI because what do we really all know is sort of developing every day, but it looks to us right now that those are the best sources of seed data to create a synthetic panel or persona. And I think if you’re in a global role and you don’t have a centralized segmentation study, which we don’t, and I’m sure other businesses don’t either, That’s a particular challenge because in some cases you would have to do that human study first to be able to create the AI, a reliable AI source. So yeah, there’s an assumption that you have the information to be able to create either a persona or a synth panel, and sometimes then that may not be the case. So that requires an extra level of investment. Okay. But yeah. Good to know. And any last words on value and advice on how to build the case for yourself in terms of measuring and tracking the value because they’re going to, they’re being used in so many different areas. Yeah, it’s not straightforward that, I mean, we’re able to identify the, we’ve worked with you guys actually to identify how much time we’re saving and make estimates of less research duplication. And then we can obviously cost up where we’re able to deliver research where we couldn’t before. So what a typical research study would have cost, wouldn’t which have been invested in at that time. So it’s definitely filling a gap. So we’re able to do it in a way that’s quite tangible for the business. And then the other aspect is the efficiency of us as the team in terms of we’re roughly the same size, but we are delivering huge amounts of not just volume of insights, but we’re moving ourselves more towards consultants in the business alongside our stakeholders, which I guess is always the sort of ideal when you work in research. So yeah, the value is clearly there. We are being asked to evidence it all the time, but I think it’s fairly clear. I’m sure, great. Well, we’ve gained enormous value by having this opportunity to discuss how you’re leading Vodafone through this, and especially what you’re learning about virtual persona. So, thank you very much for having made the time and sharing very generously with our whole community your learnings. We really appreciate that. Thank you to everybody who joined. I’d very much hope that you found it very valuable. As Kelly said, we’ll make sure that we offline get to any questions that perhaps came in or that you send in afterwards. And, Sarah, if we can connect on those, hopefully we can get the response back to everybody. But I’ve really enjoyed that. Thank you very much. And thank you Thank you. Kelly for for sharing us. Yeah. Thank you, guys. Have a good good evening. Thanks. Bye bye. Thank you. Bye.
What this webinar covers
What should AI take on, what should human research validate, and how should the two work together? There’s no single formula. As with qual and quant, the right mix depends on the decision you need to inform.
At Vodafone, synthetic personas are being woven into new processes that balance AI and human research inputs.
Sarah Emblin, Senior Insight Manager at Vodafone, helps shape the Group’s approach to AI in research. Her team uses synthetic personas at the earliest stage of the process: testing ideas, messaging and propositions, and narrowing long lists before they move into human research.
What you will learn
In conversation with Daniela Zuin, CMO of Market Logic Software, Sarah shares:
- Where synthetic personas add the most value, and where human judgment matters most
- How side-by-side testing with human research is helping build confidence in personas within an analytical, data-driven organization
- What changes as personas move beyond a small core team into wider use
- Why keeping insight in one central place matters as AI tools and agents multiply across the business
Whether you’re new to synthetic personas or already using them across your team, stream this webinar to learn how one of the world’s leading brands is building this AI into its research processes.
Speakers
- Sarah Emblin, Senior Insight Manager, Global Consumer Insights, Vodafone
- Daniela Zuin, CMO, Market Logic Software