Webinar: how yo use AI as your innovation co pilot

Awesome. Great. All right, let's get it started. So if you could tell me where you are dialing in from. And one funny question I have. If you could describe your current relationship with AI in just one word, what would it be? Okay, we have people from London, Bangkok, UK and Rocio saying love us. Okay, we have people saying super curious love. That's awesome. Exciting, explorer, fun. Keep it coming, keep it coming. AI as my co founder. I like that. Usman, a business partner integrator on and off relationship. Yilmaz. Thank you. That's exactly what I'm sort of feeling. I was thinking about this question and I was feeling my relationship with AI has become turbulent. Especially after you prepare for a webinar with ChatGPT just changing all its features, rather updating all their features. I had to really redo my webinar. I would say my relationship is a bit turbulent, but I'm really, really loving this very, very exciting time to be in a challenging relationship. That's it. Okay, let's dive in. So today's topic is how to use AI as an innovation copilot. So what we will do today, you might have seen multiple webinars from Board of Innovation. Today we will sort of do a deep dive into the innovation process. So I'll bring in some examples and processes that we have been using in some of the projects that we have been working on in the past couple of months. And some of the mentions here could be specifically for certain industry, but that doesn't mean that it is not applicable for everyone. All right? And I hear that some people saying that my voice is breaking a bit. If it continues, please let me know or if my voice is fine, please give me a thumbs up. Superb. So before I get in, hello everyone, I'm Gaia. I'm talking to you all from our Singapore office, Board of Innovation. We are a global innovation firm. We imagine tomorrow's products, services and businesses and we create them today with one of the world's most ambitious businesses. We work with businesses in food, health care and industrial sectors. I really, really appreciate all of you coming in here. If you have any sort of questions, you can please drop it in the chat or in the Q and A section provided to you. And at the end of the webinar, I will try my level best to answer most of the questions I can with the best of my knowledge. All right. When we started writing this slide, we wrote AI is AI will change how we innovate in food and beverages. But we had to strike that will because AI is already changing how we innovate in FNB. So the examples you see there right? Be it McCormick and IBM's one using AI to create unique flavor blends from recipe and preference data, or be it Heineken's Beck autonomous beer where every decision from recipe to marketing is being made by AI or food pairing AI predicting successful products by connecting the right data and insights. Or one of my favorite validation tool taste wise helping CPGs to validate new product ideas by leveraging their data set on food consumption. Or when we come into the experience completely like Coca Cola using Gen AI in its freestyle soda fountains to personalize drink suggestions based on user preferences. We know that AI has infiltrate the world of food innovation very, very strongly. And what's interesting is when Gen AI tools were launched, it had the fastest uptake ever in the history. So you know ChatGPT currently has about 180.5 million users. When it was launched in November 2022, it only took 5 days, literally 5 days for it to hit the 1 million user mark, while in comparison it took 3.5 years. For Netflix or even Instagram it took about 5 months. And even in their infancy, the gen AI tools are starting to make an impact on the real economy. So with Alphabet, the Google's mother company, losing about 100 billion in market value after a demo of their AI model, producing a factual error with all this background, we started looking into the research. What are these unique capabilities that AI bring forward? All the research is pointing us towards saying that AI has unique capabilities for innovation, ideation and modeling human behavior. A particular study that was conducted this year by Cornell University shows that the average quality of ideas generated by GPT4 is higher than those of humans, especially when you rate it on purchase intent. Very, very interesting. In line with that, with the usage of synthetic Persona or human proxies, Brigham Young University in 2022 also concluded that the language models can be effective proxies and they can model customer Personas and panels. This research information, this was where it all started. Proposal Board of Innovation to use Genai in our innovation process and understand how we can use AI as a copilot in the innovation journey. Some of you might or most of you would have attended our previous webinars and in our creative AI webinar we introduced this process. So throughout the innovation process we saw that many of these tools can be used. What we did was in the previous couple of months we have been working on sizable number of projects, playing with these tools and trying to help our clients to make life what life need next with the use of these tools. As we started working with these tools, we started seeing the traditional front end innovation process is shifting into a new form. So to kind of refresh all your memory, the traditional front end process would look like a double diamond, right? So when compared to an AI powered innovation process, the traditional innovation process usually requires extensive resources. For example, even at the first diamond, where you are sort of identifying the trends, your brainstorming, the problem, you're defining the opportunities, all of it usually take months and then you're committed team to be part of it. Then to adjust to the resource constraints, we tend to also narrow down the TA or for the other reasons. Opportunity spaces identified here often becomes more as a means to end, more like ticking the box. And we also start to narrow ideas but very, very quickly because that's something that is able by humans, right? So we, we are not able to kind of work with so much of data sets. So we tend to narrow ideas too quickly and also because of the limited resources we have and we start into this process of designing, validating and experimenting. And this double diamond process also, and you know this, it also has a bias towards like conventional concepts. Of course, when we are trying to keep resources at bay and also sometimes consumers say something, they behave differently. When we actually go in that direction, there is a bias towards conventional steps and the prevalence of say do gap. This is how the traditional front end processes and how AI is right now surfacing inefficiencies in this process. What is this new process? How is it going to look like AI powered front end innovation process is going to be different and it is different. It gets us further along the validation journey in very, very less time. Instead of a double diamond, now it becomes a stingray. With the help of AI, you can ideate and validate the concepts with feasibility or sustainability or viability in your mind right in the beginning of the process itself. This creates an opportunity to infuse backend innovation to the front end innovation. It's a good blend of human LED and AI led. So if you look at the colors there, you'd see human LED is the darker blue and AI LED is the lighter blue. So first you're sort of doing a human LED process where you're sort of training the model, you're generating the hypothesis with all the information you might have in house. And then you can also use the GPT engines to kind of do some additional research. Then you go into exponential concept development. So here you can play with a higher number of concepts than ever before. So your first step of iteration could be not based on real human feedback, but could be based on synthetic feedback. And even this, you can make it progressive. For example, you can use your ChatGPT or your internal LLM to craft synthetic Persona first. Then you go for specialized platforms which give you synthetic Personas. And then you start bringing in human validation again, progressing, narrowing down the concept, stretching the concept time and again. As you progress through this journey, it becomes more and more human led. What it also allows you is you can redistribute your resources efficiently. Your bulk of your validation, which includes a lot of your human resource and other capabilities, can actually come at the end of the process. This AI LED stuff could be done with very, very limited resource and only for those ideas which are promising enough. That's how the whole model shifts. Perfect. With that. One thing I wanted to remind all of you, this is also considering some of you might be experts who might already know this, but if there are freshers who are really looking into using Geni for innovation, I just wanted to introduce this concept as well. So when you use Geni to effectively copilot innovation, the quality of the prompt is super, super critical. So otherwise what happens is garbage in and garbage out. So what we have seen, even internally, this is something that we need to kind of really spend some time, need some training, need to really figure out. But I'm just going to go into some of the basics for you and also sort of give you some tips on how to do it. And there are a lot of resources out there for you to read and develop this or we are super happy to have a call on understanding this. Better prop to engineering is the new buzzword many of you might have heard. But what is it? So this is a picture. A colleague of mine put it in the slide and I really liked it because it's interesting. When you think about the entire gen AI, you need to think that you're actually interacting with a robot. You have to imagine that you're teaching a robot to do task. Prompt engineering is giving that robot special instructions that help it understand what you want it to do. Prompt engineering becomes possible only when you start doing prompt design. As users, that's what we work with, the prompt design. What is prompt design? Now you talk to the robot, you need to say things in a way it understands. It's actually kind of figuring out the best way to ask or tell the robot something to get the kind of answer or the action you want. So for example, if you want to know the Weather. A simple, very, very simple example could be you'd say tell me the weather. But then if you ask the Robert, tell me the weather, it's not often that you get if you are imagining that you want to hear, want to hear about the weather today, it's it might not often give you that input. So you need to know tell me today's weather. So that's where the prompt tuning starts come. So sometimes any of the gen AI tools that you're using, you really need to kind of tweak it. And that's what prompt tuning is. And this tweak in this tweaking or in this tuning is where throughout the innovation journey. So you're kind of creating a lot of internal capabilities for yourself, right? So for example, you want a job to be done and the chatgpt to identify job to be done. So in the tuning you sort of define what is the kind of functional jobs you're looking for, emotional jobs you're looking for, or social jobs that you want to bring in. For example. Sick. So bear this in mind and I'm going to next slide. I'm just going to give you a couple of tweaks before we go it and I wanted to remind all of you that as users we design and tune prompts. I mean that's how the engineering gets limited. The other overall engineering could be in the hands of techies. So the tips to design better prompts, first and foremost, be super, super specific. The more precise your prompt, the more accurate and relevant the response will be. So instead of an example of asking them tell me about chips, you should be asking what are the common ingredient combinations? Especially when it comes to an innovation usage. And the context is super key. So sometimes articulating that you are an innovation consultant from a particular organization or you're an innovator in a particular organization, maybe not name the organization, give the details of details of the not the organization, but like your e intent for it. So give the context a bit more and limit length when needed. If you want a concise answer, specify it in your prompt in one sentence. Explain this in product. Explain this product concept. So be be very, very specific, right? You can actually limit the length when needed. Use examples. So for example, for a specific format or type of answer, provide an example in your prompt. For example, you want a summary of the latest culinary food and snack and trends in the style of a tweet. Say that. So this is a language model. It really picks it up really, really fast. The next point is super, super important because you need to guide the structure and provide step by step instruction. If you're looking for a list, mention that. If you're looking for a debate and form debate format, mention that. If you're looking for strength and we can weakness, mention that or any specific structure. And later when I'm diving into the process, I'll talk, I'll be talking about why specific frameworks, specific structure is super, super key to kind of get the results that you really want. Always avoid ambiguity. Don't ask how tall is he? An example would be rather ask how tall is Michael Jordan? Set the tone. So if you want a specific tone of answer like be it, be it more formal, casual, be it more creative, like set the tone rightly. And for instance you can say explain extrusion to me like I'm a 4 year old or a 5 year old in this case. Also ask for sources. And this is something right? When you're working with Bing AI, you're working with Bard or even with ChatGPT recently, ask for sources, ask for our sources and steps. I mean one thing that you will later realize as well is that it requires human intervention and judgment to verify the input or the output from ChatGPT is something relevant there. It is good that you see the steps and that allows so always it's a good practice to do that. When it comes to prompt tuning, prompt design, always, always iterate and experiment and this is where your innovator skills start coming in. So if you don't get a decide response initially, most of the times it doesn't happen that way. It's not that straightforward. Try rephrasing your questions and us and try to ask it in a different way. Okay, so now let's get into the journey. In our observation, what we see is that using AI as a copilot is providing superior performance in terms of quality, speed and cost compared to a human. Only across the journey, in most of the projects that we have worked on, if you ask me, would you use Genai as a copilot? Hands down, yes. Because the amount of data that we are able to use, the amount of quality that it brings and the speed and the cost, it is amazing. Now you saw the process earlier. So for today's discussion I've looking into a process which is a little bit more simplified and where I can show you clearly the usage of human and usage of human plus AI in the innovation process. First, where you see human and AI combined, obviously it's in the strategy area. Here it is trends, competitive analysis, consumer problems and existing Insights, custom coming in. So here if you look at it from a step point of view, so trends competent analysis and consumer problems is something that AI can really, really sort of help us. But existing insights is something that you can sort of augment. But when you look into trends, competitive analysis, consumer problems also there is a way that you need to intervene as a human being and that's something that we will cover. And when it comes to opportunity areas you would see it's more skewed towards genai gen AI. It's for 40 of kind of triangulating all these trends, competent analysis etc and then as you go into concept generation, stretching and refinement here practically it's a blend of human and AI. So is it with desirability, feasibility and viability validation. But again from there when you go into prioritization of the concepts, you can sort of start using AI a lot more with less human intervention. And then from there you can derive top concepts. Bear in mind what I'm going to share now, it's only based on the couple of the tools here in focus and it has been done over the couple of months. AI is fast changing, there are new things that's coming in. So some of these things could become irrelevant in matter of weeks or months. But the tools today we are focusing on is ChatGPT, Bing AI, USCAN, Altitude AI, Adobe Firefly, and we have been also using Bard. So these are the tools with which we have played to sort of bring the analysis for today. So first step, trends analysis. So as you look into a trend analysis, right, the first step of trend analysis often starts with identifying trends. So here we often sort of use ChatGPT and Bard specifically to sort of identify the trends and obviously charge, you know, the, it's not like up to date. So that's where you kind of take these trends and you put it in Bing basically to validate. And in many of the projects that we have been working at one go, we have been not just looking at one market, we have been looking into multiple markets at one go. The speed at which you're able to collect and identify trends of a, of a particular category across market is like unprecedented. But at this stage, when you identify the trends, again this is where the human intervention come in. You would see the slider is more towards AI. But bear in mind it is super important that we define the framework. The framework could be as simple as something like a pestle or you're saying that the innovation fusion diffusion model, again the examples here, I'm referring back to could be more on consumer products. So when it comes to like industrials, et cetera, there can be tweaks that's made and then comes the analysis of trends. And as you go into the analysis of trend, here is where we see a combination of ChatGPT and Bing AI being used. So ChatGPT usually sets the direction Bing AI because it's more updated and it's a smarter search engine, we use it for specific signals. However, here we can also kind of do a cross reference of these trends across multiple markets using ChatGPT. What ChatGPT does really, really sort of brilliantly is that they are able to sort of analyze and determine key uncertainties for each cross market trend. So obviously these data we have to feed in, but this enables like creating and developing future scenarios. So with extensive training on trend and signal data, creative AI models are able to provide foresight into future scenarios. And the narrative is interesting and I'm going to show you a hypothetical example. Obviously this hypothetical example is rudimentary, but it shows the writing skills of ChatGPT when it comes to future scenario by keeping some of the trends in mind before we go in. In terms of trends analysis, research and analysis, the speed and quality of creating future scenarios based on multiple data points and it being adjusted to multiple market is super, super impressive. Bring always stays as a smarter search engine. Whereas ChatGPT excels in creative foresight. However, the cons like some of the limitations here definitely there is limitations on creativity and as I said earlier, there is lack of like up to date data and sometimes like some hallucinations as we call it. It's like yeah, the data is completely sort of disparate and come so you need to kind of pay your due diligence. So our recommendation would be like regular due diligence is important use like leverage a blend of tools like Charge, Bing AI or the new tools like Waldo, et cetera to kind of offset the individual limitations for efficacy and comprehensiveness. So as I promised an example future scenario in the context of pet care market. So some of the trends that we used here range from gourmet dog snacks, customized nutrition, eco conscious choices, interactive pet communities, tech, enhanced shopping. So many of those trends, right? And then there were a couple of assessment criteria that was given and then this is one of the example future scenario that ChatGPT came in. And so it's sort of, so it's sort of really interesting and I would say this is like a starting point, definitely not the final point. In our project work we definitely tweak it and because of confidentiality reasons, I'm not sharing it here. Now as we go into the competitor analysis and one of the really big benefit that when it comes to competition analysis, like imagine a food product that you're trying and you're able to sort of look into multiple markets and you're able to look at like 100 brands or hundred products at one go. So when we were looking into a competitive analysis, that's the amount of data that we were able to bring in like 100 competitors at one go. And also you, you are able to sort of analyze the data and once you define the framework again. So it could be like a 6W framework or, or product attributes such as benefits, claims, packaging, any of those attributes. Once you define that tangibility allows you to analyze how each of this product or brand would play in each space. So the graph which I'm showing on the right is a hypothetical example. But then if you look at it, product attributes are there. So against those product attributes you see how other brands are performing and that those white bubbles are kind of a hypothetical opportunity area. So this like with human intervention often it becomes super, super difficult, especially with a lot of data to come in. I mean it is possible but it's a lot more tedious. However, with the intervention of Gen AI, it becomes a lot more faster. You can do it in a matter of like days. However, as I said there are like in identifying the competitors, in analyzing all of that. Genai is amazing. But when it comes to cross market analysis, every time we sort of work like we were doing it, even though it is like English speaking countries, sometimes it requires a lot of human intervention. So you need to sort of again like do your due diligence and you kind of do human analysis and then you sort of validate it with ChatGPT. So the pros of this is definitely ability to analyze and learn from multiple competitors and adjacent brand at scale. However, the cons obviously at this time when we were working on this project, we worked without an API key and if you work without an API key, it is extremely inefficient. It results in a lot of time spent in copying, pasting, clearing the data, filtering the data, etc. Our recommendation would be that you need to start think about new tools. As I said earlier, you can think about cloud, etc. In a combination. Obviously you would use ChatGPT and to kind of result in exponential efficiencies. I'm pretty sure you might be thinking okay, how do you sort of firewall chatgpt So I in my presentation I'm not talking about internal LLM engines or GPD engines. So all of this work, it was also done with it in conjunction with the internal engines. Only those data which is not sensitive was used from like we use ChatGPT etc. Okay, so with that, so now let's look into. I'm going into the details. Now we go into the next step which is more about identifying the problem statements. So problem identification again when you give a right framework. So here our recommendation would be you need to start detailing out a journey map. So the way you train the chatgpt should be to all the stages in journey map and look straight, start to look at it from different, different angles. So once you do that, Gen AI is able to provide consumer problems along the consumer journey from all angles. Again that's why it becomes more of human plus AI. Whereas when it comes to problem assessment, with the right assessment criteria and definitions being fed to ChatGPT, it is able to evaluate and classify the problems in a very very big. So here you should think about it, right? This is where you can also intervene as a human. For example, you can sort of like if ChatGPT gives you like 60 plus problem statements, you can also kind of have like you yourself coming up with problem assumptions. Or you could also kind of use a social listening tool to plug in with for this problem statement. So the next slide we will talk about how to sort of use social listening to come up with these problem statements, right? So and then from there we sort of look into prioritization. So we assess the problem, we identify the problem, we assisted and then into prioritization. So again here if you have a predefined criteria, ChatGPT gives you a really, really good prioritization. But again sometimes we see that the it is not something like I would say just you can just trust ChatGPT at this time. You need to kind of use your due diligence. Obviously having a huge number of problem statements that we have never worked with such a huge number of problem statement before. You're also able to create uniform problem statements at scale. These stands as the pros of this. However, some of the cons, for example, when you start initially designing this high level problem statements is what you get unless you really need to direct to search for specific items. Our recommendations would be that first thing, remember the frameworks that you use when you prompt it, you need to define what you want map all possible pain points in a consumer journey to determine exhaustive list of problem areas and Prioritize the biggest points. Now let's quickly really go into social listening. I'm not sure how many of you here have used you scan before. I'd like to see in the chat if you have used it before. So like we have been working with multiple social media listening tools and USCAN stands out. And for that very reason, they recently got a seed funding, about 3.68 million or so. It's a brilliant, brilliant tool also because of the copilot that comes in with uscan. So our experiment was basically when we started to kind of see how can we generate queries through ChatGPT and to feed in. But there we were not super successful. Because AI doesn't capture the precise language humans use in social media or other channels. Just relying on AI for generating queries can be inadequate. We have to intervene as human and then update it. But whereas when it comes to synthesizing and analyzing, and that is where I was saying, the USCAN copilot is a tool with a lot of possibilities. So it sort of analyzes for you the consumer pain points, job to be done, opportunities, etc. Based on real consumer data. And then now you plug this into back to your engine, you are able to kind of again create the problem statement at scale. So basically we, for this whole exercise from my previous problem statement and social listening altogether, we were sort of Looking at about 80 problem statements at scale at this stage. So one of the challenge here however, is that the problem statements can at times overlap and it can be at high level unless you specify directly. So just like how I said earlier, it is super, super critical that we define the framework and the objectives of the research very, very diligently. And now from there, so we got the problem statements and now we look into what could be some of the opportunity areas that a particular brand or a product can or can look into. So here specifically, opportunity area is where I personally feel that AI does it brilliantly. The AI triangulates all the information, whatever we have done so far, and it makes unexpected connections to surface opportunity areas. So the opportunities could be measured on any predefined framework. Like for example, you can sort of measure it on the impact of consumer by frequency of the consumption or novelty or anything that you fancy. So, and that can be heavily done by AI. So opportunity identification and assessment can be heavily done by AI. But again here, when it comes to prioritizing the opportunity, and this is where the human judgment has to come in, there's a lot of like tacit knowledge that comes in here. I would more rely on the human side. One of the cons for here definitely is that there's a need for specific data to detect open oceans for innovations. It's not always that straightforward. Even though they're identifying it, there is an intervention that's needed and sometimes some of the opportunities that came in we had to tweak it, adjusted with and support the data. Extensive research had to be additionally conducted here. But however, it creates a lot of benefit in the triangulation of the data and creating opportunity area write ups as well. So an example of some of the opportunity areas again for that hypothetical example on dog food or pet care. So these are some opportunity areas that quickly that it generates. Right. So again the trends that we mentioned earlier, it creates opportunity areas around tailored functional traits, DIY health bites, emotional support, snack conscious duals. Is this good to go? I wouldn't say so. Again, this will require tweaking, this will require adjustment. This will require human intervention in terms of like substantiating certain things. There were stuff like culturally rich etc. Do dogs care about it? I don't know. So we need to tweak. But definitely a great, great starting point as well as you can do that within matter of weeks. That's where almost going into the final stage of this process. When it comes to concept ideation, when you come to concept ideation, the general thinking is oh, ChatGPT comes up with great concepts. At least that's what the first slide I was actually presenting. Yes, they do. However, we need to use it as a stimuli. It's up to us. We can either do ideation sessions together or you can do separate ideation sessions how you want to use it. It's up to you in your creativity. We have tried both, but sometimes like my personal preference is to conduct separate human and AI ideations. But using ChatGPT more for any of the GPT solution, more for stimulation in the human ideation session. But the number of ideas that comes in always interesting. With one AI you get about 100 ideas where with five humans you get 100 ideas. In terms of quality it is sort of comparable. But I would say certain concepts have to be stretched with human intervention. But how once you do a little bit of that idea statement stretching, then you feed it to the ChatGPT, you start getting interesting results because you start to see it creates a possibility for you to stretch the ideas based on Personas, synthetic experts and future scenarios. This is where you can use multiple tools, be it your own synthetic experts that you have created within or referring to the previous webinar we had, particularly on how we do synthetic validation. All of this comes into the step. So AI is able to stretch the ideas based on the Personas that you have fed, the synthetic experts that you bring in, or based on the future scenarios that you have created. And you can also use AI to prioritize this. Here again, I would say use your human intellect and inputs to refine and stretch these concepts further. The next point is very, very interesting. In a traditional innovation process, we often look into just bringing in desirability validation. Or when you think about the concept, you usually tend to say, okay, stay away from viability and feasibility. But OpenAI or in general gen AI is giving an opportunity for us to look into the viability of certain concepts. Basically, you're able to kind of have 80 concepts calculate what could be the ingredients cost, for example, or the Nutri score scores for it. So there are like multiple ways you can kind of look into the viability and feasibility components very, very, very early. So that is, that is another way that we can sort of prioritize. The prioritization here depends, clearly, clearly depends on your sort of need and your context. What we do often here is definitely prioritize the desirability and in the eventual sections we bring in viability and feasibility. But you can also kind of adjust it accordingly. So it again, in terms of concept ideation, it enables scale, and also in concept writing, it enables scale. You can standardize the way its concept writing is done. And also synthetic Personas. And this is something which is super interesting. So often there is a debate is synthetic Persona very close to the realities, decorate, et cetera. We are seeing very, very comparable results. And also just think about it as an early validation tool. And if it's an early validation tool and you kind of use your synthetic Persona based on your exploration interview results, then it becomes a good tool to kind of rely on as well. However, as I said earlier, sometimes the creativity could be amplified. So there could be some human behavior or complex insights that could be missing out. So you need to tweak it, amplify, etc. So the recommendation here, I would say that it should be targeted, the ideation should be targeted, it shouldn't be super broad. It is your role as an innovator to keep it specific. And often you can use it as a stimuli or a starting point for human ideation. So just I'll give you some examples. So again, now the context has shifted. Just think about a company that provides ingredients that are ingredients. But now going into a B2C kind of a setup. So here it is a customer Persona that was created fully through the insights here. None of it is from like a real human. It's all like synthetic. We had like multiple of these customers and Personas. We kind of tweaked it. We almost conducted like an AI interview with this in the Persona and then identified the function. Functional, emotional, unmet needs. And to this the solution concept that came in again just through AI ideation here and the Mario M43 is. It's in the degener. So this, this is the kind of concept that you get. Again, this can be done in a Sprint format. It can be done within like a matter of days. So finally. So once you have the concept, you have done this. Now how do you sort of interview? Do you really go in and talk to the consumers? Yes, we know this. We have been doing this for a very, very long time. So I know all of you there. No, but there's another interesting way, again a question here. How many of you have used outset AI are familiar with outset AI? Okay, drop in your thoughts, drop in your answers into the chat. So outside of AI is really, really interesting tool. So again it allows AI driven interviews. So the details about this tool has been mentioned in our previous webinar. Again, so vote of innovation. When we started doing we were using video ask. We are always being in the forefront of the technology due to this. But with AI driven interviews, it allows you to use it as exploratory or even if it for a concept validation. That's what the features allow. Basically using outsource AI, you can train your own AI interviewer by picking your own interview Persona and you can also build your own discussion guide and you can also kind of leave instructions along the way. So for example, you can sort of say, listen, if this question come in, change it this way. You if this, this is the kind of answer comes in ask why. So you can sort of engineer your. So the in holding I'm seeing a question, what is the name of the tool? Gaia Outset O U T S E T Outset Outset AI. Yeah, perfect. So that's the tool that we have been using extensively to do these interviews here. Especially your designing of these interviews should be particularly done by your team. Again, this is where you define the metrics, you define your labels, how your concept is performing, etc. With that you can actually feed outer AI or any of the tools that you want to use here in it could be video ask, etc. But here my analysis is based on that tool you can conduct interviews in multiple markets. For example, for a project that we have done over the weekend, we were able to do almost 300 interviews in multiple markets, in three markets. And out of that one market wasn't speaking English as a language, so there was a bit of translation that has to be done, etc. So also the tool allows smart tagging and also kind of efficient data collection classification. I would not stop there. It is often important you take those. You use your internal LLM engine or GPT engine and create summaries of these feedback on the concepts. Try to understand is there areas that you feel that there is a little more data. This is where we actually bring in expert interviews, real expert interviews or real validation interviews here. But by then you are actually starting your interview with like lots of data already on the concept and after one set of prioritization. So again my. I would say be aware of the classification errors that can happen sometimes. Like I spoke a lot about the qualitative interviews, but when it comes to quantitative interviews, sometimes the current state of things, the tools are not doing a great job. If you want to do a conjointly, if you do want to do it, regression, etc. We haven't seen great hope yet. Unless otherwise use the typical tools. But as I said, things are changing at Godspeed, it can be different. One recommendation as you do AI driven consumer interviews, make your objectives super, super clear and design templates for data tagging and analysis structure much in advance. So with all that, I mean you sort of looked at almost all the process and obviously this process continues there. It is more about human LED validation or you can use a combination. But some of the key learnings something that I wanted to highlight through all the slides that we have seen so far. When you use AI as a copilot, it definitely give you a large volume of ideas very, very quickly, which allows you for more concept directions and stretching than traditional human only ideation, as I might have been repeatedly saying. But human refinement is still needed to capture the nuance and cultural relevance. Also this is where you as an expert, you know how to sort of tweak things the way it has to go. And that's where your experience and innovation also comes to play. We saw that in outset AI that it is able to kind of do faster in interviews or be the other tools. We are able to kind of do interviews in an unprecedented scale. But we do see to capture subtleties. Sometimes you need to augment it by a human intervention or real interviews. Here you can be very, very choosy with whom you speak to. And as I said earlier, like you need human judgment to prioritize. So for example, it's not yet advanced enough, but this is also changing to fully replace human judgment for prioritization. It can quickly process large volumes of data to aid decision making, but the decision should be yours. It also enables initial visualization of concepts. And often we don't see like branded concepts that can be sort of done. Final branded mockups has to be done with the help of a designer. You also need structure from humans. Many of the frameworks that you have learned as innovators becomes very, very handy here. And as I said, qual and quant, qual is where we can use it a lot more. Quant still work in progress. You're able to sort of do feasibility analysis, identify early stage cost drivers very, very early in this process. However, these are again in early stage. But like anything complex on feasibility or cost drivers, like viability components, becomes really difficult. However, like as I said earlier, when it comes to concept, if you define the parameters, it's able to kind of size your concept, especially the ones with more recent data with that. We all know this is the worst AI will ever be. So what like you saw the possibilities of AI in innovation and this is how things will be now, but it is the worst. It's in the nascent stage and we really, really know that things are shifting and how is it sort of shifting? So and how do we capitalize this, how do we win and where do we get started? So winning an AI powered innovation, you should build a superior engine while working across your strategy, demonstration project and transformation initiatives to make sure it reaches success and scale. So you should develop your strategy to unlock the value of generative and autonomous AI for innovation. So what we are seeing and what we have started increasingly working with clients is to help them build an autonomous innovation engine. So by developing and upgrading your new innovation engine, and then with that engine, obviously you can demonstrate the value of AI in innovation projects and you can learn, learn from each of this project or, and you can embed and scale the change across the people, processes and systems. So you should start with building your automous innovation engine. So and building an autonomous innovation engine. It could look like all those steps that we went, we looked in, right? There could be an insights engine which is automatically kind of doing the social listening, looking at the scientific breakthroughs, looking at your proprietary data pools all firewalled. And there could be a concept engine which is constantly generating the Concept, screening it and creating the brand models. There's a simulation engine that's doing a desirability, feasibility and viability simulation. And there's a build engine which is basically identifying the top concepts and kind of generating the design and the coding and eventually a scaling engine which kind of enables you to kind of do an autonomous go to market. So this is sort of what do we mean by engine? Is it the combination of the right tools? It could be a combination of the right tools. It could require a bit of backend work and the coding as well, 100%. But it is about going more in towards the autonomous part. So what I'm saying is what we looked into is you were sort of looking into what's today where you are curating the tools and working. But an autonomous engine is when you build something internally within your organization. So to know more about it, we invite you to come to the Autonomous Innovation virtual summit from 6 to 7 December. Exactly, Narjit. I mean, that's definitely something we have to sort of ponder and work on. As I said, I prepared the webinar a couple of days ago and yesterday is where the launch came in. So why create your own when CHAP GPT has launched away? I mean, yes, definitely. So I invite all of you to sort of come in for this Virtual Summit and I think my colleagues will help to share the link here to register for the summit. And I guess we can sort of take five minutes to kind of answer some of the questions. Okay, I'm just going to look into the question. Okay, I'm going to take this question by Brian. So we have found great ways to use AI to generate ideas at scale. Any good solutions you have found to evaluate ideas at scale? So, Brian, to your question right now. No, to evaluate ideas at scale is where your prompt tuning and we are actually using a combination of tools sometimes like the assessment criteria sits with our Excel and then we create our prioritization criteria and then we do Excel programming to calculate it or any other programming tools like that. But we still do not have any empirical results showing the accuracy of synthetic personality. There are a couple of articles out there which I don't have the links handy, but it is starting to come in, I would say. How was the competitor analysis completed? The question from Charles. So I'm struggling to kind of understand the question, but I guess obviously it started with identifying the competitors, to mapping them out, to identify the opportunity spaces. And then obviously we looked at it from multiple perspectives in terms of product, in terms of brand, in terms of features etc. And that's the thing that we needed there. Yes, I mean we also kind of, we can always augment it with experts and to kind of understand a bit about like foresighting on the competition itself. So that's something we can add. Okay, I'm going to take Rick's question. How accurate is AI in predicting trends? Instead of a hype, I still find it quite difficult to use it for trends. So Rick, again, as I said, it really, really depends on how you're prompting it. You need to have the right frameworks, you need to use a combination of tools and it's about what you find in Bard. You kind of verify with ChatGPT and then you sort of like go into Bing and validate it. You also sort of augment it with your internal research, etc. So what we have been seeing, especially in consumer products front end innovation trends, it is doing a decent job. Yes. Andre Weber, your question is very, very interesting. How is the AI approach changing the concept of getting out of the building early in the innovation process? First of all, it clearly depends on your choice. In some of the projects that we do a lot of exploratory work and understanding the jobs to be done, etc. You still do with the consumers. But if you want to do depending on the resources you have, you want to keep it completely AI, you can also you can use your synthetic Personas. Obviously here it is. It might not be getting out of the building, but I would say the getting out of the building is sort of digitized by working with the synthetic users. Because getting out of the building, what Steve Lang meant was about spending time with the users and here synthetic users are able to give you that. All right. Okay. I'm just going to look into chats if there is some idea question. Okay, I like this comments of from Naji Comment from Najeeb as humans have not been good at asking questions, we will talk more now AI is forcing us to ask better question. I wonder who training who. I would really, really like to wrap up this webinar with that. It is super, super interesting in one quote. It's like we all know, right? Spend about 95% of time defining the problem, re articulating the problem and rather than just jumping into solutions and AI will make us do so. So thank you everyone, thank you so much. It was really nice. I really wish I had a chance to see all your faces but really appreciate all of you, your comments and interactive session. Thank you. Thank you so much.