AI Personas: Helping Brands Make Smarter, Faster Decisions with Segmentation
Editor's note: This article is an automated speech-to-text transcription, edited lightly for clarity.
C+R Research sponsored a session during the January 20, 2025, Quirk’s Virtual Sessions – AI and Innovation series. Two speakers from C+R Research and one from Signoi Ltd. presented on using AI personas for smarter and faster decisions.
Kathleen Blum, VP of shopper insights, and Paul Metz, CEO, of C+R Research partnered with Andy Dexter, founder director of Signoi Ltd. The trio walked through the process of how Signoi has used AI personas with it’s clients and the effect it has had on their business.
Session transcript
Joe Rydholm
Hi everybody and welcome to our presentation, “AI Personas: Helping Brands Make Smarter, Faster Decisions with Segmentation.” I’m Quirk’s editor Joe Rydholm – thanks for joining us today.
Just a quick reminder that you can use the chat tab if you’d like to interact with other attendees during today’s discussion. And you can use the Q&A tab to submit questions to the presenters and we will get to as many as we have time for at the end.
Our session today is presented by C+R Research. Enjoy!
Paul Metz
Hello everyone. Thank you for joining us today. We're excited to share the work we've been doing to leverage AI and segmentation research. I'm Paul Metz, CEO of C+R Research, and I'm here today with two brilliant minds who I'd like to introduce themselves.
Andy Dexter
Hello, I'm Andy Dexter and I'm one of the founders and directors of Signoi Ltd., a U.K.-based tech firm.
Kathleen Blum
I'm Kathleen Blum and I head up the shopper Insights practice for C+R Research.
Paul Metz
Thank you, Andy. Thank you, Kathleen.
Before we get started with the webinar, for those of you who are unfamiliar with us, I want to share a bit about C+R Research.
C+R is a full-service market insights firm based in Chicago, founded over 65 years ago.
We have deep expertise in custom research, including quantitative, qualitative and communities and panels. We have extensive experience helping clients find answers in B2B consumer and shopper spaces.
We also have in-depth knowledge of key audiences including generational cohorts, LGBTQ+ consumers, youth and family and our CulturalBeat practice, which focuses on multicultural audiences including Latinos.
Andy, could you share a bit about Signoi?
Andy Dexter
Sure thing, Paul.
Well, we don't quite have the long history of C+R Research. We’re a recently formed AI tech and analytics firm serving the insights industry, and that could be end-clients, insight agencies and ad agencies.
Essentially, the software we've built runs end-to-end analysis from large scale techs, EG, half a million NPS comments or massive social media trolls to decoding semiotically images and video at scale through to AI audience simulations, which is very much the focus of today's webinar.
And on the latter, we found any audience is fair game with enough data with simulate Gen Z, Millennials, Boomers, barbecue enthusiasts, Pinterest users, first-time homebuyers, fashionistas, drinkers of whiskey, tequila, beer and wine, department store shoppers, high-net-worth yacht owners and even Middle Eastern soccer fans, and it's all very interesting indeed.
Paul Metz
It sounds really interesting, Andy. Thanks for that overview.
Shortly after ChatGPT burst onto the scene in late 2022. We knew that we were going to work on integrating AI into our segmentation research.
Segmentation is at the core of what CNR does, and we've been on the forefront of innovations and segmentation methods for a long time. Dating back to one of our founders who is credited with originating the idea of occasion-based segmentation methods.
Now we've turned our attention to advancing segmentation research with AI.
As we do this, we're intent on doing it in a deeply thoughtful and human-centric way. In fact, before we started, we developed guiding principles for how we would use and apply AI in our work.
At the core of our approach is the belief that great research begins with human ideas and that we use AI as a research assistant and a sounding board for generating ideas that help fuel our analyst's creativity rather than replace them. We brought this experience and our AI principles into conversations with Signoi and started building our partnership.
Andy, do you want to describe your role and view in the partnership we're forging here?
Andy Dexter
For sure. I think what's very interesting, actually I'll come to this in a minute, in a little bit more detail, is my personal background is not as a tech bro. It is as a research and strategy consultant that's basically set up and run various research insight agencies over the years, but Signoi itself kind of builds on those principles.
Our core principle as a software firm is to basically develop and deploy innovative AI platforms for analysis and interaction, but in such a way that it is oriented towards what market research people need, whether it's client-side or agency-side.
That leads into AI bots, which is essentially about curated realistic, intelligent persona simulations based, as in this case, on segmented target audience research data. Which allows you to query that info to obtain insights or have in-depth conversations with personas representing the segments.
So, the way we gel and overlap is around this notion of longstanding heap consultancy expertise. And I really wanted to stress, it's not just tech, but we're developing smartly tailored solutions that connect with the needs of insight professionals.
Paul Metz
Yes, I completely agree, Andy, in that our very custom approach and that our desire to really bring forward human attitudes, emotions, behavior as we step into AI really dovetails well with the principles and the work that you've been doing.
There are a lot of reasons we're excited about using AI personas and segmentation.
Chief among them, for me, is that AI can bring segments to life conversationally. We can quite literally talk to the segments. We think this is a great way to foster stakeholder engagement and it creates an easy, fun and memorable way to learn about a brand or company's target segments.
Andy, you've done a number of these. What feedback have you heard about the value and benefits that AI personas provide?
Andy Dexter
Well, I think you're spot on actually, Paul, in terms of using these approaches to socialize segments and bring them to life. But beyond that, actually it means you can do real research with them.
A few examples, you could use them for iterating, testing ideas, concepts, comms line extensions, pricing strategies and more. We're finding that on the wider canvas, our clients are using the simulation platform across a really wide variety of use cases to get pretty much instant feedback on pretty much anything because they don't get bored. They don't get tired. You can ask questions that you didn't ask in the segmentation studies, you can ask questions that didn't fit, but more than that, the simulations will actually come up with new ideas.
So, clients are also using them almost like a brainstorming thought partner in workshops and these personas configured correctly will even draft ideas for effective products and comms, even social media posts, things that would appeal to them and engage them personally.
Paul Metz
Those are really great points, Andy.
Kathleen, you spent most of your career as a client and worked on a lot of segmentation projects. What most excites you about the possibility of AI personas and segmentation?
Kathleen Blum
Yeah, thanks Paul. I think what I'm most interested in are some of the things that you mentioned.
You mentioned the aspect of it, makes it more fun. Andy mentioned that it isn't boring. I think the important thing with segmentations is while the best segmentation can be understood by researchers and insights personnel, being able to translate that to your marketing department and your sales department becomes a little bit more difficult. You want them to digest it, make it their own so they really know how to action it.
And to me, this helps with that actionability because now they can actually interact with the personas, ask them questions and start to get really excited about potential ways they can use it and just even hypothesize with the personas about strategies.
Andy Dexter
I do agree with that, Kathleen. It's almost like having your target audience in the room with you a lot of the time.
Kathleen Blum
Exactly.
Paul Metz
Yeah, those are great points. I think that the ability to immediately extend and apply segmentation to assess marketing tactics is really exciting.
Well, let's move on and talk about one of our segmentation projects that now has AI personas.
C+R has an established and experienced shopper insights practice. Early last year we completed our own shopper specific segmentation study that we leveraged to be smarter when we consult with our CPG and grocery retail clients and we can also integrate this segmentation lens into any custom shopper research, whether quantitative or qualitative.
We felt that this study would serve as an excellent use case to develop AI personas because of the numerous questions that the segmentation can address. It was really important to us that the AI personas be anchored in high quality segment specific data.
As we collaborated with Andy and Signoi, we decided to field additional research all qualitative among each of the shopper segments. We conducted online discussions for depth and fielded a large sample survey comprised entirely of AI probed open-ends to collect shopper language at scale. Signoi took our original segmentation data along with the rich qualitative data and used it in their AI bot's platform to create AI personas for each of our six segments.
Let's take a quick look at our shopper segmentation before we introduce the AI personas.
Kathleen Blum
These are our six shopper segments, so let me take you quickly through them. I'm going to start off in the upper left-hand corner with our Deal-Savvy Enthusiast.
This really is the person who really loves the art of the deal. Finding those bargains really makes them feel very accomplished in their shopping and they tend to shop more often than the other segments because they are always looking for those deals. They also will shop a wide variety of channels, traditional as well as online because it's all about the bargain.
A Convenience Seeker is also an online shopper and the reason they're very heavily online is because they are all about fast and efficient for them.
Then we have our Weekend Browsers down in the lower left-hand corner. For them, shopping is a fun chance to get away from the stresses of their work week and enjoy seeing what's new, what's available out there.
As we move forward, we'll go over to the top on the other side to our No-Nonsense Shopper. They have some similarities to our convenience seeker. They want a hassle-free shopping experience, but that experience is not online for the No-Nonsense Shopper. They're used to what they're used to and that's traditional grocers and mass retailers where they can find what they want easily and quickly.
Then we have our Penny Pinchers. As the name implies, it's all about how they can get the most for their dollar.
And then finally, our Sophisticated Traditionalist who tends to be a little older, has the disposable income and at this point in their life, they deserve quality.
Now we can move from these segments and start to bring them to life into our AI personas. Let's see what that looks like.
As you can see, our personas start to bring things to life. There's a lot more texture in what we get with the AI shopper personas.
Now we have our Deal Savvy Enthusiast, Samantha, she's 32 and she's a marketing coordinator.
You have the Weekend Browser who after a week of teaching art to high school students, she needs a break and she's going to enjoy that shopping trip on the weekend.
Frank, our No-Nonsense shopper really is about convenience getting in and out. He likes what he likes. He's used to what he's used to.
Our penny pincher, Martha.
Finally, Charles, our sophisticated traditionalist who likes quality, no surprise, is a semi-retired corporate lawyer.
Paul Metz
Thank you, Kathleen. So, as we work with Signoi, they were armed with all of the data that we collected. They use their expertise in AI to develop comprehensive and nuanced personas in their AI bots platform.
What you see on screen are just snapshots that hint at the richness of each persona. Each persona has a bio of text that goes on for pages and pages and makes them a lifelike representation of real consumers. The underlying models for each persona are anchored in deep segment specific research and language from actual shoppers. This is what helps ensure that each person, when engaged in conversation, reflects what is essentially an imitation of life.
Andy, what steps do you take to craft personas that have an aura of authenticity?
Andy Dexter
Well, let me pick up on the imitation of life point here.
The key thing is that these personas are not shallow. Our AI process expands from all of the segments info to generate persona profiles that include not only things that were present in the research, but also for example, their cultural values, their personality traits, their backstory, their lifestyles, their hopes, their fears, their dreams and their guilty pleasures. All this helps give them an identity and a personality and a tone of voice.
The key point here I think, is that the AI bots don't merely parrot back what they've been fed. They're not just dry constructs based on dry data. They are kind of imbued with the wider spirits of a given segment and they genuinely have their own opinions. We give them room to breathe, to be creative in their responses, albeit without ever breaking character.
One thing I would add is we never want to push that too far. They need to retain their core segment identity to be useful.
Paul Metz
Thanks Andy. So far, we've been really happy with all of the collaboration and the contributions that your team has added to all the rich data that we've provided.
Okay, well let's do a live demo.
Here is our AI bot’s platform. It resembles the interface of ChatGPT and similar LLMs. The majority of the real estate on the screen is devoted to a chat interface. There's a handy dropdown menu that provides a short descriptive snapshot of each persona. There's also a menu that allows us to select who we'll talk with.
Included in the AI bot's persona platform is a research expert. Selecting the research expert allows us to query the body of research behind the personas. We can simply ask questions and even retrieve data from the segmentation.
We'll show you a few examples.
We see here the expert researcher is awaiting our questions and so we'll enter one into the chat interface. So, we're just going to simply ask, tell me about the segments with a brief profile of each one.
Here we see it's giving us a nice synopsis in bulleted format of each of the six shopper segments that Kathleen outlined for us. And it does a little bit of formatting.
Let's back up and take a look at what it did.
As you can see, it just gives us this snapshot in brief of the Deal Savvy Enthusiast. It gives us the percent of shoppers and a few factoids about each one that stays in character and is tied to the underlying data. We know that Deal Savvy Enthusiasts are younger and tech savvy.
Our example persona is named Samantha Chen, one of the Convenience Seekers. We have Alex who values convenience and prefers online shopping. The weekend browsers where we have Emily and so on and so forth.
Now let's go a little bit deeper. We're going to ask it to give us some data to retrieve some data from the segmentation that forms the basis of data underneath the segments.
So what we're going to do is ask for a number of facts, the size, the grocery share of spend, share of trips and grocery shopping frequency for each of the segments. We've also asked it to assemble that data in a table.
Here we can see that the platform is assembling the data. We'll give it a second and it will format it for us as well.
Here's the result, and you can see it gave us exactly what we were asking for. It arrayed the segments down the side and all of the data points across it retrieves it from the underlying data set.
This is one of the nice things about the AI bot’s platform. It can serve two different purposes. One, it can just give us an interface to interact with the segmentation data in a real easy Q&A type of fashion. You can see it's very quick and it's also accurate.
The other use is that it allows us to actually talk with the personas as if they were real people that are part of each segment. We're going to demo that next.
So, to do that, first we kind of need to get out of the expert researcher mode. We need to call all of the personas to order. So, we'll do that with a short prompt.
Okay, so all of the personas are ready to chat and the system lists them out for us.
We'll start by talking with all of them and we'll start with a simple question and we're going to ask them to explain what they like and dislike about shopping for groceries.
Samantha's first, not surprisingly, she loves the thrill of finding great deals and discovering new products. Sometimes it limits the fun of exploring everything she wants to explore.
Alex is more convenient oriented, prefers online shopping because it's quick and fits into their busy schedule. He dislikes when he has to go to physical stores and face crowds.
Emily, as a Weekend Browser, really enjoys the shopping experience such as the colors and the smells of the food and stumbling upon something new. That whole element of discovery. The downside is she probably spends more time than planned in the stores.
Frank, who's my personal favorite persona, the No-Nonsense Shopper, likes the straightforward nature of getting what he needs and moving on with his day.
Marta is budget constrained, as our Penny Pincher, likes finding deals as well and stretching her budget. She dislikes when prices rise unexpectedly.
And Charles, who's the opposite of our penny pincher segment, enjoys really looking for the high-quality ingredients.
We can see that they stay in character with the segments that Kathleen introduced to us just a few minutes ago.
I'd like to ask Kathleen now what do you think of these responses? Are there any personas that caught your attention that you'd like to hear more from?
Kathleen Blum
Yeah, Paul, I do love how this comes across and I'm getting more texture and they're coming alive for me. But I would like to talk a little bit more to Charles.
I'm wondering if you can create a dialogue between two personas? Because Charles and Frank have varying opinions on shopping for grocery at mass, and I would love to see if you can create a dialogue between the two. Is that possible?
Paul Metz
Yeah. Actually Kathleen, that's one thing we can do with this platform. Instead of the dialogue like us researchers to personas, we can have personas chat with each other.
So, just give me a second to quickly get a prompt going. Alright, so we're going to ask Charles and Frank to have a friendly debate. Let's see how that goes.
So, Charles started the conversation and they're doing what we asked, which is kind of taking turns and putting their opinions forward. Yeah, so Charles started, he's kind of speaking into Frank's life a little bit, which is interesting you imagine he finds shopping at places like Walmart, practical, but they don't meet Charles standards for quality. He prefers stores where he could find high quality and carefully selected products, even if it's spending more.
Let's see, Frank says; well, I see Walmart as a good solution for getting everything I need in one trip.
So, Frank is after the convenience and the reasonable price, sufficient, fits his routine, doesn't see the point of spending extra when you can find decent quality there.
Then poses back to Charles; what about you? Aren't you just paying more for the same items?
Charles: I see your point, but my emphasis is on quality and curating a shopping experience. I appreciate stores where I can trust the provenance of my goods, not just the price tag, but enjoying the process. Don't you ever feel like the environment at a place like Walmart is a bit impersonal?
Getting very specific.
Frank: Honestly, I prefer the no-nonsense, straightforward experience.
Okay, let's jump down back to Charles.
Charles: In those cases, I plan my trips carefully or use a trusted smaller store nearby.
So, he's referring to quick pick up trips.
Charles: I value personal service and do my shopping during times that aren't so busy. I suppose it comes down to what you value more, efficiency versus experience.
Frank: True enough for me, efficiency wins the day, keeps my life simple. I'm fine with that, but I can see if you have different priorities. It's all about what fits best to your lifestyle.
Charles: Right, exactly, Frank.
Okay, so good. They wrapped it up in a friendly style, they each made their points.
What did you think about that, Kathleen?
Kathleen Blum
I like it. It actually is very true to form from what I would expect to hear from Charles and Frank. I'm also really impressed by how they respond. It's a healthy and truthful discussion. It feels very authentic, but it's still polite.
Can we go a little bit further, Paul?
Paul Metz
Sure. Yes, absolutely.
Kathleen Blum
So curious with Charles in particular, does he go to mass for other things outside of grocery because he seems like that's not really fitting his needs from a grocery standpoint, but does mass fit other needs that he has outside of grocery?
Paul Metz
Got it.
We can ask Charles specifically if he buys other items at mass and which ones he thinks are okay to buy there.
Kathleen Blum
I really love the healthy discussion you get from the two of 'em.
Paul Metz
Yeah, that's one of the fun things about the platform and in experimental trials with it, we've had situations where we've had one persona moderate a discussion between two other personas.
Here's what Charles said.
Charles: When it comes to non-grocery items, I'm more flexible with shopping at mass retailers.
So, you were onto something Kathleen.
Charles: For products like household basics, paper goods or electronics where quality is more standardized, I find it acceptable to purchase from places like Walmart or Target. The difference in quality between stores isn't as pronounced. However, with groceries, especially fresh produce, meats and specialty items, there's a clear distinction.
Then it kind of goes on to reemphasize his focus on quality and expertise. So, I think your hypothesis was validated in that discussion.
Kathleen Blum
Nice. And it really comes through true to Charles, he likes high quality, he deserves the best.
Paul Metz
Yeah, absolutely.
Alright, let's change gears a little bit because one of the other things that a persona platform can do is extend the research beyond the bounds of the underlying segmentation research.
We prepared a little concept test and that's one of the things we've been experimenting with on the platform.
So, we're using a brand of granola bars called Junk List Chewy granola bars, which was just launched nationally in Target in January of this year.
So, I have a pre-scripted prompt here. It's a little bit long because the concept description is actually in the prompt, so I'm going to drop it into the chat interface.
We explain in the prompt, and you can see it there on the screen, what it is, where it is, it's going into Target stores, non-GMO, doesn't contain artificial flavors, preservatives, it has a price and we're asking for three metrics, and the system is already generating.
We're asking for appeal, uniqueness and likelihood to buy. We can see the system is assembling the responses. Now we'll give it a second, it should format that a little more nicely for us and it does.
As you can see, we have the personas down the side and the scores across and then a quick little comment from each persona.
If we scan the third column, we could see the likelihood to buy scores. We do have differentiation across the segments where we see the higher scores with Samantha, Alex, and Emily, and lower scores for Frank and Martha and Charles, who is really the quality.
I want to look at Charles because this product seems to be positioned with no junk ingredients and so it should meet his quality. But he says;
Charles: While I value quality ingredients, this type of snack isn't usually part of my grocery routine.
I think we happen to know that Charles is a bit more into fresh quality ingredients and probably not going to go for the pre-processed granola bars. Is that right Kathleen?
Kathleen Blum
Absolutely. It fits him. He likes the quality, but this isn't exactly in his wheelhouse.
Paul Metz
Yeah, we see Martha mentioned that the price is reasonable for healthy snacks, so there's some interest and would consider them especially on sale. And
Samantha loves the fact that they're available at Target and have clean ingredients and if there's a deal she'd try them.
So, we see good differentiation, we see commentary and in their evaluation that stays true to their underlying character.
Hopefully you can see through the past few engagements with our personas that they can be used not only to chat in long form conversations but can also respond to common survey ratings and the platform conveniently organizes and formats the output.
I want to point out that as we've been doing validity testing, we found reliably consistent patterns of scaled response from the personas that align really well with the underlying data and the characteristics of the segments that they all belong to.
So we all know the Super Bowl is coming up and so we had an idea as we were thinking about what to demo in the platform and so we looked into which brands are going to be advertising during the Super Bowl this year and found out that the ice cream brand Haagen-Dazs will be airing their first ever Super Bowl commercial.
We thought it would be fun to find out what our personas think about a few ice cream brands. So, we created a prompt for that.
We're going to ask them about their opinion of Haagen-Dazs and a few other popular ice cream brands, which include Ben and Jerry's, Blue Bunny, Klondike and Edy’s. We chose this array of brands intentionally to kind of get an array from more premium ice cream brands to more cost efficient or budget brands.
We can take a look at all of these scores and see that there is some differentiation that there's, if there was a premium battle between Ben and Jerry's and Haagen-Dazs, it looks pretty neck and neck.
We see Frank and Martha with slightly lower scores as we might expect for higher price brands.
Let's find a little bit more specifically about Haagen-Dazs because that's the brand we're interested in for their new Super Bowl advertising.
Alright, so here comes the response.
They call back to the ratings to remind us what they rated. We see that Samantha gave it a four, loves the rich flavors and premium quality, but keeps an eye out for deals since it's a bit pricier. When it's on sale I feel like it's a luxury treat at a decent price.
Martha gave it a three. It's out of her price range, appreciates the quality, kind of recognizes it, but it's really a splurge item.
It fits Charles well as we might expect.
Andy, I've got a question for you. You've worked extensively with AI personas. Is there anything you've noticed or learned about how they react to brands that weren't part of the foundational research?
Andy Dexter
Yes, thank you Paul. It's a very interesting question actually.
Across all the various AI bot projects for various clients, we've noticed probably a couple of major things.
First of all, cycling back to our concept of imitations of life from earlier in the presentation, there's usually an inherent truth about their responses, assuming that they're built on high quality data because within a category they're able to essentially extrapolate from what they already know, which may not, as you say, have been in the original training data, but they would extrapolate from what they actually do know and from what the sense of the overall persona is.
That can be quite spooky actually. It can really be quite spooky. And of course you can test concepts that they've never seen before. You can test fictitious concepts, new ideas, etc. as we talked about earlier. So, that's one thing.
The second thing is that what you also get is a kind of a variation in responses that stays true to a theme.
So, if you're asking for ratings of something that is not in the original training set, the personas will respond around a kind of central value if you like. So that central value could be four if you're asking the same question 20 or 30 times. They might sometimes rise to a five or a three, but they'll kind of coalesce around that central modal north, which gives you an insight, I think, into the way that AI work actually in the sense that they're not deterministic. They actually have this kind of degree of creativity and extrapolation built into them, and that seems to be consistent across all of the models.
Paul Metz
Excellent. Yeah, that's really interesting. It's something we've noticed as we've iteratively tested this new platform.
One of the things speaking about kind of extending beyond the original foundation research, we don't necessarily need to stick in a narrow lane with our questions to these shopper personas because our underlying research also included data about their food values and motivations. We can ask broader questions about higher level grocery related topics.
So, we have a question for our AI personas. We simply ask them; how do you feel about food production standards in the U.S. versus Canada or the EU?
Here come our answers.
Samantha keeps an eye on food production standards because she's into health and sustainability. She hears that the EU has stricter standards, especially regarding GMOs and additives which she admires.
We can see that each of the personas has a slightly different take on that topic.
So, time is running a bit short, so we better wrap this up.
We hope you enjoyed the demo and that it really inspired you and stimulated ideas for how AI personas can be applied to address many types of marketing questions.
To wrap things up, we want to share a few thoughts about what we've learned along the way.
One of the most important things for us is to feel confident that AI personas respond accurately and in character that is consistent with the underlying data.
So far, every face validity test we've done has been on the mark. Additionally, when we've asked the platform to simply retrieve and report out underlying data, it does so with perfect accuracy. When we run the personas through the original segmentation typing tool, they classify correctly without fail after numerous iterations. We're currently engaging in new side-by-side research, both qualitative and quantitative, fielding identical questions to human respondents and our AI personas.
Our goal is that when we work with clients on their projects, we want them to be a 100% comfortable that their AI personas reflect real human attitudes and perspectives. Another learning we had is the interviewing mode really matters whether we engage the personas as a group or individually.
Andy, would you like to comment about this?
Andy Dexter
Sure thing, Paul.
Well, as you've seen in the demo, there's a variety of tips and tricks and prompts and constructions that you can use to get the simulations to behave in different ways. And that could be having a friendly debate for other projects. I persuaded them to have a very unfriendly debate. The choice ultimately is yours. But the crucial point here is that this construction of the inputs IE, the prompts into that chat window is absolutely critical. And sometimes you have to kind of coax them along. They can be, like I say, they have personalities. They can be a little bit recalcitrant, right? Sometimes you have to really persuade them to engage with certain things.
If they don't particularly like it, you can get 'em to do focus groups amongst themselves. In fact, we've got some models where a lots of personas are running up to 20, even 60 different personas, and you need to be able to be comfortable with interacting with the model via these fairly complex prompts.
One of the things we're going to be working on is developing a list of power prompts for people to use things that we've already done with other clients to make it easier essentially for people to interact with the models. Because let's face this kind of in unfamiliar territory, it's good if we have a series of kind of training prompts if you like, to help users along.
So, the way you talk to them is critical, actually. That's one thing we've learned.
Paul Metz
We also quickly discovered that they don't always respond identically to iterations of the same question. We actually consider this a good thing because it reflects the realities of variations in human language.
Upon close examination, we see that the nature direction and evaluation of their responses remains consistent, but the exact words used to express their sentiments do vary. And that's something we really like and appreciate and lends that authenticity to the personas.
Andy Dexter
I think this is a very important point actually.
So, it is not a deterministic system. It's always anchored in the bedrock of excellent training data, and that actually leads us down another path. So for some projects we actually end up creating multiple personas per segment. And that can be really important because depending on how well constructed the segmentation is in the first place.
It can be very tightly defined with very clear archetypes, or it can be less well-defined like a cloud, if you like a cluster of things that share a common identity but also vary in subtle and interesting ways.
That can lead us down the roots of having, as I mentioned earlier, quite large quantities of personas for certain projects. And that then plays into the ability to use the models in what I'd call quant mode in order to do a survey across quite a large amount of personas and therefore reflect the representative audience more closely.
Paul Metz
Thanks Andy. So, in closing, we are very excited by the promise of AI personas and look forward to including them in our future segmentation work.
Thank you, Andy. Thank you, Kathleen, for your time.
We welcome your questions.