Harnessing Natural Online Conversation and AI to Take Concept Development and Testing Next-Level  

Editor's note: This article is an automated speech-to-text transcription, edited lightly for clarity.      

On January 30, 2025, Sophie Wright, managing director at discover.ai shared a new method of concept development that the organization has been working on. Discover.ai was one of eight companies to sponsor and present during the Quirk’s Virtual Session – AI and Innovation event.  

This method is very new for the company but has been used on a few projects. In fact, Wright used a case study with Nestlé to demonstrate the process. She walked through each of the steps and explained how the AI streamlines the process for efficiency.

Session transcript

Joe Rydholm 

Hi everybody and welcome to our session, “Harnessing Natural Online Conversation and AI to Take Concept Development and Testing Next-Level.” 

I’m Quirk’s editor Joe Rydholm and before we get started let’s quickly go over the ways you can participate in today’s discussion. You can use the chat tab to interact with other attendees during the session and you can use the Q&A tab to submit questions for the presenters during the session and we will answer as many questions as we have time for during the Q&A portion.  

Our session is presented by Discover.ai.  Enjoy the presentation!

Sophie Wright 

Hi everyone. Thanks for coming today. I'm really excited to talk to you about this new offer. We're looking at how we can approach concept development and concept of testing in new and pertinent ways.  

I should say, first of all, we've got half an hour for this webinar. I'm going to aim to stick pretty precisely to that timing. If you have questions or suggestions or challenges as I talk, please put those into the chat. My colleague, Suzie, is there in the chat all the way through the webinar and we'll pick up your questions and send you answers or suggestions on how we can communicate about those.  

First of all, let me introduce myself. I'm Sophie. I'm the managing director at Discover.ai. I've been here since we started, so just over seven years ago. Prior to that my background's always been in consumer insight in brand strategy and marketing, sometimes client-side, sometimes agency-side. 

I should also just quickly introduce Discover.ai, who we are and what we do. I'm aware there's people on the call who may not know us, who may not have worked with us.  

So, we're a team of strategists, very focused on delivering insights and ideas for our clients on the team. We've got a group of people, people with backgrounds in semiotics, cultural anthropology, consumer insight across all sorts of different methods and in brand strategy. We work with our clients to deliver insights and to deliver ideas.  

Now the way we work is that we always start our journey to get to insight. We always start by pulling content from online conversations and we gather that content to create a dataset in a very curated way.  

That's the beginning of any of our projects. We kind of build the data set to fit the brief and fit the project in a sensitive way, but we're always working as a start point with what people are saying naturally in their conversation, spontaneously in their conversation. That's kind of the start that we kick off from.  

The other thing I should say is that we’re working with a proprietary platform that we've built ourselves since we began. This platform also has AI capability built in. I'm going to talk a little bit later as we get into some of the details today about how that works. But it is really important to say that the AI is just a part of what we do. It's just a small part, it's a support to what we do.  

Clients often ask us, ‘how much is people, how much is the machine? Is this all just about ChatGPT coming up with the insights?’ We always say it's 100% about us. It's 100% about the people on our team and what they bring to the analysis and the meaningful bit of the analysis is done by us and not by the machine. 

But the way the AI works is to help us get across a real breadth and depth of content and hone in on the rich, interesting, exciting, resonant bits quickly. The AI helps us go further and go faster basically. I'll explain a little bit more of the details of that shortly. 

Today we're here to talk about concepts and ideas. We're talking about a new approach to concept development and concept testing.  

Now where's all of this coming from?  

What we hear from our clients, we hear it often and we know from our own experience, is that oftentimes concepts can get in the way of our ideas. There's a kind of irony here. We all use concepts to capture ideas and to share and communicate those ideas to other people in order to get their feedback. Often that is with respondents in research. It can also be with internal stakeholders, but the truth is it's really difficult to capture the essence of an idea, to capture the essence of a good idea and get it into a concept and get it into words. 

Our clients tell us this often. So often we put a lot of labor into creating concepts, but those concepts then instead of helping us to share ideas, can obstruct the sharing of ideas.  

People can get hung up on the words or hung up on the graphics or the images that we choose can be distracting and it can put them off ideas. And this is how great ideas or good ideas can kind of get killed off by the concepts. 

What we're looking at, what we're asking really, is what if there was a different way? Is there a different way to capturing the natural essence and the natural energy of ideas as they live out there in people's conversation? Like whether we read it or not, whether we hear it or not, whether we go to that conversation, not out there in culture, out there in conversation.  

Are there the kind of shoots of these ideas? What if we could read consumers' minds? And what if we could read the psyche of culture at large to understand the kind of natural heartbeat and the natural essence of ideas as they're alive out there?  

That's really our starting point. All the time in the work that we're doing, we are using natural conversation as I've said.  

Our method completely relies on first of all, gathering content from what people are saying in their natural conversation, whether we listen to it or not, whether we go read it or not, what's naturally being said. This approach here is quite different and it's about saying, can we go and find the ideas our clients have? Can we go and find your concepts out there in people's natural conversation? And can we find the kind of language of the currency today that people are using to express those ideas?  

And can we understand from finding those ideas, finding their presence in natural conversation, can we understand how big they are? Can we understand how much energy there is in those ideas?  

That's the sort of premise that we're working with here. That's our ambition for this new offer.  

Now there's two different aspects that we're looking at and I'm going to just go through each one in term.  

The first is really about looking at natural language, looking at the natural currency to talk about ideas and trying to capture that and harness it for expression. This is all about metaphors, turn of phrase and people's natural selection of language and selection of phraseology, but particularly metaphors. I'll come onto the other approach after.  

Let's talk about metaphor now. First of all, what is a metaphor?  

Metaphors are really, really interesting. We all use metaphors all the time without thinking about it, it's completely instinctive. It's kind of a default. It's totally unconscious and a metaphor, what's a metaphor?  

A metaphor is really a way of expressing the significance of something and kind of getting across how important something is.  

As an example, if I say this morning I'm dying for a coffee, I'm not literally dying, whether I have the coffee or not is not going to dictate whether I stay with you here today or not. But what it does get across is that that coffee feels really urgent to me.  

It also gets across that coffee's a fundamental part of my morning routine. It's something I do daily at a key moment at the beginning of the day to prepare me for the day and to give me the get up and go and to keep me going.  

To talk to you about this metaphor approach and what we do here and how we bring it to the world of ideas and particularly concepts, I'm actually going to talk you through an example of work and work that we've done recently with Nestle.  

So we do a lot of work with Nestlé, we do a lot of work in coffee, which is good really because my coffee is really important to me. I wasn't making that up. And what happened here was Nestle asked us to help them with idea development and to help them with expressing those ideas in concepts.  

What they were looking for particularly was how to capture some of the richer emotion that people have about their coffee, about their coffee routines and their coffee habits. They were looking to understand and capture the role that coffee plays emotionally in people's lives.  

But importantly, they wanted to find a way to communicate that emotion. It is quite hard when you try to communicate emotion, particularly to respondents using concepts.  

So, on the one hand you want to touch on the significance of things, and that means we're digging into emotion and that's our role as insight people. But on the other hand, if you get the expression of that wrong, you go too far, you exaggerate, you're too touchy feely, it's a bit naff, it's a bit corny, it's a bit OTT or it's kind of marketing bullshit. 

There's a difficult balance in getting the words right in concepts. And Nestlé very specifically talked to us about that, and they asked us very specifically to look at language and to look at metaphor.  

What do we do?  

We go out there and look at what people are saying today about coffee. We bring into our platform a really good sample of the content in conversation about coffee things people are naturally saying.  

In this particular instance, we're looking for what people who drink coffee are saying about drinking coffee, kind of consumer voice stuff. We look at what's being said in the chat on a whole range of social media platforms and we're looking very much at consumer forums. 

That's a really big source of content for this particular project. We have a platform that gathers that content and organizes it so a strategist on the team can hone in on the interesting insights.  

Now in this particular case, we're looking very explicitly at language. What language are people using to talk about their coffee moment and what does that tell us? What does that reveal about the significance of coffee, but also what does it give us as currency to use to talk back to respondents about those coffee moments and about the significance of those coffee moments.  

Now I promised that we would get on to a bit more detail about the AI bit in the platform of how we work. This bit is that bit. So, this is the busy chart, it's the more technical chart, looks a bit scary, but we're just going to focus on a couple of places on this chart.  

First of all, you'll see there's four big blocks of black across the top of the chart. These are the key steps in our journey to get to insight, working with our platform and starting with this sort of sample of content from online conversation. 

The first thing over on the left, we pull in the content that's relevant and that's going to be rich and helpful here. We have a very curated approach, and we pull in a really big sample from that conversational content into our platform. And there's a bit of organizing, filtering and editing that happens around that at the beginning of the process.  

Now the other thing that happens if we go onto the second step, and this is the AI bit coming to play.  

So, what happens is the language model that sits inside our platform that we've built over time and that has learned from our work and all our projects, it organizes all that content. You can imagine there's a massive amount of content in our dataset. It reads everything super quickly. It would take us weeks and months and years to read through all the content. It reads through it super quickly and it organizes it in a very meaningful way so that we can then come in and navigate through that content very quickly and focus on all the really rich and interesting bits.  

So, it will organize it into human needs. We have a model of human needs which it understands, and it will organize all of the data in the dataset according to those needs. 

We can open the door onto each one of those needs and unpack the content that sits behind them. It will also help highlight the language in the data set that's most resonant, that feels really significant, that feels rich in insight, and it will push that to the top of the project so that we come and look at that first.  

It also understands what the project is about.  

It will organize all the content to push towards not just the richest and most interesting content, but also the content that's most pertinent to the question at hand. It will also theme all the content into themes that are relevant for the topic at hand. 

So, lots and lots of really smart analysis that happens really quickly. That's the first important step.  

That means that then when we move across to step three, we come along as people. We bring with us all of our baggage unashamedly as people. We bring our experience, we bring our points of view, we bring our passion for particular categories and particular questions, and we start to then engage with the content focusing in on the interesting bits which is helped by the AI and its organization of the content.  

In this particular instance, we're looking for language, we're looking for interesting consumer turn of phrase. We're looking very specifically for metaphor and the job from there on into the end.  

Deliverable is an entirely human job of reading, of selecting, of deciding what's the kind of nugget of insight and starting to build stories that we play back for our clients.  

Now in this particular case, we know we're working for Nestlé, we know we're working on coffee, and we know we're looking for a metaphor. And as I've said, metaphors are really interesting and we all use metaphors all the time. So, this particular example is going to show you one or two outputs really quickly.  

One area we were looking at for Nestlé was a particular occasion, the first coffee of the day, the coffee that I referenced earlier, the coffee that I was dying for.  

Indeed, that theme of dying comes up again and again and again across a whole range of different markets. We were looking at different countries in this study and the metaphor as a whole articulated it as life or death. 

The idea of coffee as literally life or death, and that mourning coffee seen by people and described to people as absolutely essential, you've got to have it. You can't live without it and it feels sort of lifesaving and life changing. 

In each instance when we came back with these metaphors to the client, we were capturing language that expressed an emotional insight that helped with idea development, but we were also very specifically feeding back to the client a kind of blueprint of how to talk about this idea in concepts, expressions, language use and vocabulary you can use.  

I'm going to just show you another one, a little bit different, which was the idea of coffee as stolen time. Coffee's a nod, nod, wink, wink, understood excuse to slip out of work, to slip out of the everyday routine, to sneak off and to get a bit of fun, to grab a bit of relaxation, to grab time for you.  

Again, there's a lot of very emotive and interesting language being used there. There's something really interesting about metaphor that's interesting in this context of talking about concepts and looking for ways to express ideas and express the emotion behind good ideas in a way that's acceptable, in a way that's understood, in a way that really resonates and is relatable for consumers. 

That's really why we're focusing on this metaphor angle. Metaphor gives you permission to talk about emotion. Metaphors are a subconscious way, a currency, we all have to express particular emotions about particular events, particular experiences, particular rituals. 

So, metaphor can reveal to us things that we don't think about because it's subconscious, it can reveal deeper emotion. That can be the start of a great idea. But really importantly in this context, metaphor also gives us currency that's emotional on the one hand, but permissive on the other hand. In that sense it's a kind of magic tool we believe in for concept writing, right? 

That was one of the two things I'm going to talk to you about. The second thing is a little bit different. The second thing is much more about evaluating ideas and is kind of a lateral take on concept testing.  

In this particular instance, what we're looking at is how we can use our approach. Remember our approach starts with pulling in content from what people are naturally saying in their conversation online, creating a data set with that content. And what we're looking at here is whether we can use that to go and find your ideas out there in conversation, in culture.  

First, we're asking ourselves three questions.  

Are your ideas present out there in the conversation?  

Are they resonant? Does the conversation and the nature of the conversation indicate that those ideas are relevant, that they're compelling?  

Also, we're working in a slightly qual-quant combo here, we always are, can we dig deeper into what's being said in order to understand why that is happening? What might be more compelling, what might be less compelling and why is that?  

That's what we're setting out to do here with this new method. You've seen the approach chart, so I'm flashing it up again and there's only really one thing I want to draw your attention to here and that's around this middle part.  

What we're doing here is we're looking at our dataset, we're finding evidence of the presence of your ideas, and then what we're doing is dividing up the whole dataset into clusters where each cluster in the dataset represents one of your concepts.  

So, we're clustering the dataset into concepts which allows us to do a series of quite significant things. I'm going to talk through those one-by-one a little bit more slowly.  

First of all, obviously we've created a dataset from online conversations to represent your category. That is the first thing we do, once we've understood exactly what you're trying to communicate with your concepts, the idea behind the concept, we're then going to the data to find those ideas in the natural conversation in there. 

We will then work with that dataset, and we will select quotes in the content that really nail the essence of your idea that represent the essence of your idea.  

We start to pull those quotes over and create buckets of quotes that sit in the data set where each bucket of quotes represents one of your concepts, one of your ideas. And in a way this gives us a blueprint for the idea.  

We can use that blueprint to then find in the rest of the dataset the content that naturally sits with that idea. We can cluster the dataset accordingly so that we have a sense of which parts of the total dataset allocate to each idea.  

Now the way that we do that is our language model.  

We present the bucket of quotes, the blueprint print for the concept that we've created, to the language model. It can basically create a vector that represents the concept.  

Now a vector's quite techy language. The way I understand it is that it is a location on the language model that represents the concept. What it's then doing is finding in all the other conversations that we've got, all of the other quotes we've got in the dataset, and pulling those that are semantically close to the content of the concept blueprint. It's pulling those over to it so that we can cluster the whole dataset.  

Now once we've clustered the dataset, we obviously have pools of quotes that cluster to each concept and we can look at those and start to develop some metrics.  

We can look first at the volume of conversation. So, how much content has been pulled to concept number one as opposed to concept number two, as opposed to concept number three, which of the concepts seems to have pulled towards it? The bigger the volume of the dataset, the bigger the volume of conversation. Those concepts that have a higher volume size simply mean there's more content in the dataset. There's more content in the conversation we've pulled in that associates with that idea.  

So, that gives you a sample size sense of the size for the presence of your idea out there.  

The other metric is a little bit different and a really interesting way to evaluate your ideas.  

This is all about energy. Our language model also understands different types of language, and it assesses what we call ‘the energy in the language,’ in a conversation. Energy is really a measure of the degree of emotion and resonance in the language being used. So those pools of content pulled towards a concept that have a higher energy score indicates that the content in that opportunity space is more emotionally resonant.  

Now, once we've done that and we've done that scoring, we can start to map our ideas and plot them on quadrants where we're comparing the size of the conversation that's allocated to each concept versus the energy of the language being used in the conversation that's allocating to each concept. And this can start to give us strategic models to work with.  

It's really important to say here that we're working in this space that sits somewhere between quant and somewhere between qual as we classically see it.  

At any point when we're looking at these metrics, we can also dig into a rich bank of quotes that sits next to that concept. We can look and see how many quotes are there? How much of the conversation? Is this a big idea? Is there a lot of talk about there that's allocating to this?  

We can look at the energy, but it's really important to understand what's happening. At any point as analysts, as strategists, we can say, well, let's just go and look at those quotes. What are they saying?  

What are they about? What language is being used? We've got a big pool of quotes that have allocated to concept number one, what kind of needs are being talked about there? Is there any sense of unmet needs or frustrations? What benefits are being described or suggested? Is a particular type of product format or ingredients or inclusions or recipe getting a lot of mention? What's the nature of the language being used?  

We can be looking for our metaphors here as well to see if there's interesting terms of phrase that express emotion that can help us understand and develop these concepts.  

All of this can help us not only look at each idea in isolation, but look at what's happening across your category or across broader overarching opportunity areas and start to build a richer picture of maybe category shifts, maybe within a given space we can see some ideas seem to be performing really well on our map and some less well.  

Let's look comparatively. What's sitting behind the ideas in the conversation that are performing well versus less well? Is there some kind of shift happening where we can see there's something emerging in the opportunity space in the category that's really revealing and interesting?  

That was the second approach, which was more about a lateral way of evaluating ideas, a sort of lateral approach to concept testing.  

We actually think we're at the very early stages. We've done this a handful of times on a handful of projects. So the approach is evolving all the time. Our instinct and our sense is that this approach can be used in the journey of concept development to help us to do early testing of ideas. Then look for ways that we can prioritize stronger ideas and really improve them before we then go into more classic concept testing approaches where ideas get sort of finally selected and make it through and decisions get made about them.  

That's our sense of where this approach might naturally fall in the innovation journey. But really keen to hear your feedback as well after the call if possible.  

Right, I'm in the last few minutes. I feel like I'm a racehorse rushing through the content with my eye on the clock. I've got a few minutes left, just to summarize what we're talking about here.  

We're really excited about this, but it's in its nascent trade stages.  

We hear from our clients that concept writing is really difficult. It's really hard to capture a great idea in words and therefore written concepts can often get in the way of great ideas, can kill great ideas. That's what we hear.  

What we believe is that there's a different way and there's a way to find your idea out there in the conversation that's naturally happening around your category and to understand the natural essence and energy that's behind your idea as it lives and breathes out there naturally.  

And what we therefore offer and what we're developing is an approach where we locate your idea in conversation. We capture the language that's current, emotional, resonant currency to express the idea, and then we gauge the size of the footprint for your idea and the kind of natural strength of the heartbeat for your idea.  

That's what we've been talking about. That's what we're working on.  

That's it for the webinar. I think I'm like 30 seconds off time. I hope that's been really interesting. Thank you again for coming. I know people's time is precious and everybody's busy.  

If you have any questions that we haven't answered in the chat during this call, please drop us a line. If you don't have our e-mails, you can drop us a liner at info@discover.ai and we'll be happy to continue the conversation. Thank you.