Webinar: How to run an AI-powered Innovation Sprint

Awesome. Let's go. So, welcome everyone. I'm Vincent, I'm partner at Port of Innovation and the founder of our offices here in the US And I'm absolutely excited to present today's webinar on how to run an AI powered innovation Sprint to you all. And throughout the presentation we're going to go through first of all a perspective on AI for corporate innovation, then specifically the AI powered innovation Sprint. Then we're going to dive a little deeper into the process and tool stack for each of the different steps of that Sprint and we're going to end off with running how to run your own AI powered Sprint. And finally, I'll share a few links to a few tools and resources and of course make the time to do a Q and A and to capture any questions you might have. For that, feel free to post any questions in the chat and we're going to curate some of those questions and I'm happy to answer some of them at the end of the presentation. So don't hold back, feel free to post your questions throughout the presentation and if, yeah, we'll, we'll definitely get back to most of them or some of them at the end of the. Of the webinar. Good. So, so with that I'll start with the perspective on AI for corporate innovation. I think there's been so much talk about AI in the corporate world and how it will impact creative work and innovation more specifically. And the question is, is it actually having the impact and will it have the impact that many predict? And we did some research on that to look for studies that can provide us a perspective of the future impact of AI. And this study for example by MIT and Nvidia actually predicts that the impact of AI is going to be five times bigger than the impact of when the Internet was launched. So it's definitely predicted to be a very impactful technology that will change the way on how we do business. And with that it's good to look at today and see how AI is already impacting some of the creative work. So for example, we see inspiring collaborations between big brands like Ikea and Patagonia. This one was featured in Fast Company. Other startups launching purely based on ChatGPT, very large funding rounds being done based on startups that have only that haven't existed several weeks ago, but as well very important platforms like GitHub publishing studies on for example the use of their copilot saying that more than 46% of the code developed on the platform is already AI generated. So that's obviously an immense number if you know that most development is happening using that same platform. So most of it is already auto generated. And finally as well you can see examples of video games being used or developed in AI and finally also new web applications being for example fully developed based on voice command. The question is obviously is this going to, is it reliable? Can we actually use it in innovation processes and can we rely on the results? So we typically take a dual approach. On one hand, experiment with it, try it out and see whether the quality of the work is equally as good as a human powered AI human powered innovation. And then second, look at some research and if I would start with that first of all in the areas of idea generation we've looked at several studies and several studies show that the average quality of ideas generated by GPT4 is higher than those of humans when you would rate them on purchase intents. And especially knowing that for younger generations like Gen Z there's a very big say do gap and AI can actually take that buying intent into account. And so potentially the idea generation is potentially more accurate. And actually studies show that that is actually the case, especially when you feed the algorithm with ideas that have been proven to be successful in the past, it will take that pattern and reuse that for any idea generation afterwards. Second is when you're using it for evaluation. There's several studies, it's early days but definitely first studies are coming in on large language models on whether or not they can be used for qualitative and quantitative research. And this study by the Brigham Young University tells us that language models can be seen as adequate effective proxies for specific human subpopulations in research. So what it says is that the qualitative answers that a large language model would provide to you, and this is ChatGPT3, the answers that it would come up with or the feedback that it would give as if it's a human are nearly as accurate as actual human populations. So it's super interesting both for qualitative and quantitative research, but as well in the later stages of innovation when you're selecting ideas or prioritizing them. And so that's really on the research front. But what we've seen is by applying it with the on dozens of projects over the past months where we compare, we really a B test human tracks where all the ideas are created and tested using a human only approach and then you have the fully AI powered approach and we really see that the quality is better or at least at the same level, but taking into account that timelines can be massively reduced. So it's very intriguing and, and first signs definitely seem to be very, very positive. What's important to note though is that there's multiple AI maturity levels and that differs from human only innovation all the way to fully autonomous. And we can see those different scales. And it's important for you as an organization and as an innovation team or as an executive in an organization, to understand where you want to be, where you want to be this year, next year and the years to come. And typically organizations start from human only or AI assisted. And then the question is where, how far do you want to go? Do you believe in a fully autonomous AI engine where AI can be, where innovation can be run fully autonomous? We're actually hosting a webinar that on the 25th of September, so feel free to tune in on that one. It's a more future perspective on how we believe AI is, is potentially creating an environment where innovation can be done fully autonomous. But for this, taking a step back to today, or tuning in on the AI powered sprint, which literally situates it in the AI powered level. So in the middle level, where it's definitely a human powered approach, with that, you can see that when you define where you want to play as your organization, you can identify multiple ways on how to deal with AI, starting from providing a training or developing a simple playbook that people can use several prompts, that they have guidelines on what tools to use and whatnot. But also what we find very important is that you have a project level, AI sprints or demonstration projects where you show the value beyond only inspiring, actually applying it as well. And second, if you want to apply that scale, you could really roll out an autonomous innovation engine where multiple projects would actually be AI powered. And then finally, when that's all happening, people are inspired and know how to use AI. And you've done several pilots. It's important as well to provide that more strategic framework and so more an autonomous innovation strategy, for example, or really starting to embark in a transformation of your organization because that is in the end required to really make AI the best use of the technology. Again, we're zooming in on demonstration projects and AI sprints. So using AI to come up with new ideas, to test them, and to come with strategic recommendations at the end. So let's start with, before I jump into the deeper presentation, I'll take the time to quickly explain Board of Innovation, our company, so you know why we do this and, and who we are. So we're a global innovation firm that imagines tomorrow's products, services and businesses and creates them today. And we do that with a team across more than 20 countries making that impact on a global scale. But having main hubs here, for example, in New York and in in Amsterdam and Antwerp and Singapore, we join forces with the world's most ambitious businesses. So a diverse set of food and beverage companies, health care organizations, but also B2B and industrial players all around the world. With that, I'll tune in on the AI powered innovation Sprint specifically. And I think this is the easiest summary for us to explain what an AI powered Sprint is all about. In our opinion, a Sprint without AI feels like running a race without wearing shoes. So the idea of running a human only design sprint, for example, feels completely outdated because you have such an amazing tool stack at your disposal that it feels super weird to compete against an approach that is just by far better. And I'll explain it a bit because we've been doing a lot of design sprints, but with recently we really stopped doing it. The reason why is that we believe that the AI tools can be used in so many different steps to really make it a much more effective approach. And I'll highlight a few things. First of all, obviously getting everybody together in a room for five days in person or virtual, it's a lot of commitment. You're relying a lot of the brain power and brain capacity of people in the room, whereas potentially technologies can do a lot of that, reducing the time investment from the individuals in the room and therefore making it much more effective. Then second, sometimes a lack of concept quality, once first ideas are coming up, and typically that second day of ideation, the idea quality isn't that good. So actually the prioritization that's happening might not be as good as it should be because you don't have the information needed to potentially assess the quality. Then if you select, there's often selection bias. So although there is in many cases there's a lot of research and information available, it's not always taken into account. And actually the people in the room might decide versus the data making the right decisions. And then there's too much overthinking. Once you're narrowing down, the process needs to proceed, so selection needs to happen. And then people start to overthink, iterate again, add another feature, adapt it to a consumer segment. So a lot of overthinking, which again sometimes takes away the main focus, which is going towards just a qualitative outcome. And then finally low fidelity concepts are created that might not have the quantitative backing. So qualitative backing is often encouraged to get people in the room to get a handful or a dozen of consumers to provide feedback on your solution. In a digital context, you could do some digital testing. But again, using AI and using the new technologies that are available, like synthetic testing, you can actually do quite high quantity validation in again the AI Sprint format. So with that, I'll highlight a few benefits of the AI powered innovation Sprint. First of all, as I said, it removes that bias because all available data is considered in the selection of ideas and synthetic testing of concepts. It absolutely improves the quality of concept writing and visualization. As you have certain technologies at your disposal that really allow you to push the boundaries of what was possible in the past. You can move a lot faster because of the efficiency gains at every step of the innovation process, resulting in a much faster go to market. So where in typical food and beverage, for example, where we would take six months, four to six months for concept design and validation and testing, you can actually do that in a much, much shorter amount of time, allowing you to go faster to market. And in the end, if you're in a fast mover context, fast mover consumer context, that is obviously a competitive advantage, you can look much, much broader to look for options that you might have not considered in the past. And then the final two would be to iterate sooner because you can come up with those ideas and concepts much faster. Within the same amount of time, you can do much more cycles of iterations, allowing you again to get to a better quality outcome. But it does definitely force us to think much harder because we are not the creators anymore. As humans, we're actually the editors. We, we have to spend most of our time in thinking critically and strategically and also creatively challenging the output. So we're definitely, it's not a sit back and relax type mindset. It's really an all in editing mindset. So what are the three pillars of an AI powered innovation sprint? It starts with generation. So that can be in the early phases, generation of insights, generation of ideas, or all the way to the generation or actually production of a product. Then you validate, you want to make sure that consumers like it and that you'll be able or your customers would like it, and that you're able to then make money out of it. And finally you curate the ones that are strategically the best decisions for your company to move forward with. And those are connected through concept development, insight generation, the curation to then make that final decision. And it's an iterative loop. So although the different steps, the three steps of such a sprint go through, generate, validate and curate, it's an iterative cycle where every time you go through these loops over and over again, what you see is we came up with that simple yet quite powerful process where before you start, we typically take about a week to train, to train the models, to train the LLMs, to do data review and to make sure that you're all set for doing this in an effective manner. So it's the preparation, let's say then in about one week, but you can actually condense it to one to three days. You go through generation, validation and curation. So you create exponential insights and ideas. You synthetically test and design, you synthetically design concepts and test them. And then finally you do the high fidelity concept design to then come with final recommendations at the end of that track. And afterwards you can iterate on the outcome and potentially already work on the plan to go to market later on. So let's do a deep dive into the tool stack on each of those steps. What in the next chapter I'll go through is on one hand, the activities within each of the steps, several tools that we use, but also a highlight on what a typical team is that you would need for this because there's some edits to that as well. You can't come in with the same team you would take into the room in the past. The higher level overview looks like this where we have again the same four steps with the go to market at the end, where at first on the training side, we look at gathering all existing competitive user and trend research to generate much more insights. So what we do is we bring together all that information. Sometimes it requires editing, meaning some information needs to put into Excel sheets, needs to be put into, taken from visuals or images into raw text data, for example. Then you also want to train and build your own LLM. So that could be ChatGPT, that could be Claude, that could be synthetic users. Any model that you feel is relevant throughout should be trained in advance and tested in advance so that you can really move fast when you're doing this. And then finally obviously scoping and setup of all the tools, so making sure that everybody has access to, to often your internal GPT or any other tool that you would require. Then when we're in the generation phase, you want to, before you start, you actually do all the data review. So you literally a few days before you start, once the models are trained, you can run all the information through the LLM and You can get the first insights synthesized and you can start exploring the full problem and define some first opportunity areas. So typically you have different types of research, competitive research, and your own consumer research that has happened. And you have some strategic priority areas. You can literally bring all those together and ask the model to come up with opportunity areas. That obviously needs some human curation. But with that you can really have a pretty effective start where every data set is taken into account. You can then just build on that and start making jobs to be done for each of those opportunity areas and straight away go all the way to idea generation. So based on those opportunity areas, you can ask them all to come up with hundreds of ideas based on each of the specific opportunity areas, market segments, etc. Then you validate them. So typically we like to start with synthetic validation, so we build synthetic Personas or synthetic users and we ask the model to evaluate the different ideas based on the segment that we've identified. And based on that, you can refine the ideas from that short list and iterate again to start working on those first visual prototypes. Again, that can be hundreds already on that second day. Then finally on the third part, you want to work towards high fidelity concepts. So you do. Once you've selected a subset, let's say 10, 15 different concepts, you can do the concept writing by adding specifics like descriptions, claims, benefits, reasons to believe, scientific backing, competitive analysis, all that can be fueled automatically into the concepts. You can cross check the competitive positioning and then invite, for example, consumers to end off with real consumer testing. And this can also be curated by AI, where AI would create the interview guides, it could even run qualitative interviews and come up with side questions when consumers give special or quirky answers. So even that can be AI powered. And then finally, once you've finalized this, you can come with strong recommendations and strong business cases to then come with recommendations of how to go to market and take it further forwarder. There's different, different tools at each step of the process. And I'll go through several tools at each step of that process. First of all, in that first part of where you would generate the the insight, it's important to scan the market and to market to scan for example, for entry risks at a glimpse. So what you can do here, you can use clothe, you can use social listening and other tools like Bing to source information that is out there. That can be proprietary data and you can fuel that into your internal GPT environment in Azure, for example. But you could also take external data that you source from the web. And what's powerful here is that you can really do that in a way that it structures that information and already takes the next step in summarizing it and highlighting where you as a company have the biggest risks or where you feel where the impact might be the biggest for your company. I've mentioned it earlier as well. There's very powerful tools like synthetic Users that allow us to do to create virtual individuals. Where in the past you had segmentation studies, Personas or demand landscapes or anything of that sorts. That's really powerful. But it would be even more powerful that you could ask those consumers in real time throughout the entire process. Any question on ideation, but also on the validation side. And that's what Synthetic users is all about. And then you have other tools like if you want to use your own large language model, you can really build it for yourself. So for example, Cloud2 can handle up to 100,000 tokens. If you have the latest version of ChatGPT4, you can go for 32K, which is again allowing you to have multiple sources of information, dozens of documents, dozens of research reports to be uploaded into the LLM and keep them safe for you and within your proprietary environment. I've been using a lot of consumer products examples, but it's actually really important to highlight that you can definitely also do this in a B2B context. And often it's even a very good starting point because there's so much information like technical research, but also annual reports or finance sheets that are publicly available. And you can really analyze all that information in a heartbeat using models like Claude or using models like like ChatGPT4. And again, if you want to input information, there's very interesting plugins like ask your PDF or any other tool that allow you to even read PDF files. What we're looking for in that information is obviously discovering patterns. And that is the key thing that any generative AI model is actually very good at is uncovering those patterns. And that's what we're looking for, both in the competitive set, but on social listening as well, or even on different consumer research studies, overlapping them and identifying patterns where maybe in the past you might have not uncovered them. There's interesting tools like Tastewise that have a really interesting angle on doing social listening and identifying trends that you might that is happening. For example, this is if you would take the example of coffee, if you would search on that term in taste wise, it said that Vietnamese coffee is something very intriguing. Well, first of all, as a social listening company, as A company, if you would be in that space, it would be interesting to know that. But second, what's interesting as well is that it automatically generates visualizations or concepts of what that could look like. So based on the social media and social listening, it comes up with recommendations and other tools like you scan for. Social listening have also incorporated technologies like the Microsoft or ChatGPT Copilot, which really allow you to go further in giving recommendations based on social listening and not just giving the output of the amounts of people that are talking about coffee on TikTok, for example. Finally, it's really important to generate those ideas using all that information. So Once you're in one thread, one chat, for example, within ChatGPT, you want to use all those insights and then come up with ideas. And again, ChatGPT4, even 3.5 is actually quite powerful in doing that. Here's an example of partnership ideas that we came up with for, in this case ikea, to come up with multiple partnerships of companies that they could work with. And first, ideas that ChatGPT would recommend to work with. And that's the power of this is as we all know, the quality or creativity of ideas is actually defined by the amounts of ideas. In the beginning, you have to go far and explore as much ideas as possible, so very wide to then go to a final selection of ideas, because that allows you to uncover ideas that you might have not thought about in the first place. And that's again where the power of generative AI plays is when you have an automated infinite amount of ideas that can be generated and you prompt it to look for the extremes, you prompt it to look for things that might not be the main idea. So in the beginning, what you'll notice is that you prompt and that it will come up with more generic answers. But the goal is to keep pushing the algorithm to give you more and second to also ask for extremes and more interesting ideas that you might have not thought about in the first place. So you end up with a large amount of ideas that you want to select. In the past you would do that manually, but with AI you can use tools again, like synthetic users, where you would select the ideas based on all the data that you fed it at the start of the track. Then when you're going into that idea selection in that second part, in the next part you actually want to double down on the top rated ideas. And as I've shared earlier, the studies really show that when you feed the model with, let's say you did concept testing in the past before you launched a product and you've, you take all those ideas that worked in the past and you feed the algorithm with those ideas, it will actually take the parameters into account and will reuse it for any idea that you can come up with. There's for example, tools like Storms AI that is super powerful where you do that. And when you click through on an idea, it actually automatically comes up with the next version or next iteration of any idea. At the end of that track, you want to start visualizing. And again, this is something that is done extremely powerfully using tools like, like Midjourney, Dall E or Adobe Firefly. They can really come up with visualizations of your ideas in a very fast way. And if you're, for example, struggling with the use of these tools, because the prompting of these different tools isn't always that easy, you can use a few tricks to come up with a prompt. For example, there's tools like Prompt Base, where you in this case, ask for a packaging design for a new snack. It'll give you different visualizations and once you click through, it'll tell you what the prompt was that it used to come up with this. And the benefit is then you can start modulating those prompts so that you get something that you're looking for. So again, quite powerful is that you don't have to start from scratch. There's tools that do this for you. Sometimes it's like a few dollars that you have to pay for it, but usually they're there for free. The other option is that you actually use the reverse prompting feature of Midjourney. It's like the describe query, where you actually throw in an image and it asks you ask Midjourney to come up with a prompt that it would use to get to that image. It's not as good as the prompt base, but it's a pretty powerful way to do this as well. There's even areas where you don't even need prompting. For example, Taste GPT allows you to analyze trends and automatically come up with those ideas. For example, in this case, if this is for a company that makes mozzarella paninis, it analyzes that there's an interesting idea to potentially combine that with garlic bread, which is trending at that point in time. And then it comes up with visualizations of what such a combination would look like. And then finally, when you want to curate those ideas, you have four different levels of doing that validation, human to human, the traditional way where a researcher would interview consumers. Then you have human to synthetic, where you Use a synthetic or virtual Persona to give feedback to your ideas. Then you have AI to human. So beyond creating an AI generated script, you can actually have the. You can have AI to actually do the interview and come up with additional qualitative questions based on the response that are the responses that the human would give or the consumer would give you. And then finally you have AI to synthetic, which is where the entire interview is run by AI, but also the other side is a synthetic user. So there's literally no human interference. And the only thing it would do is that it would come back with recommendations based on the testing that it did itself. So with that, it might be a good question to ask yourself, what's a good human plus AI team for this? Well, it starts off with the same setup as you would have in a regular project. You have a project lead, somebody that facilitates the process. And especially given that it's a new way of doing things, it's quite an important element to keep everybody in the loop of what's happening and what the steps are. Prepare everything in advance and make sure that all agendas are aligned. Because it is quite different compared to a usual sprint. Then you obviously need the scientific insights, the science of the R and D side and you need the market side with marketers. But you augment that with three different individuals. Typically a visual prompt designer, where you would use designers in the past, you would now augment that skill or that person with a prompt design because most of the design will happen through through generative AI rather than manual design. Keep in mind that most branded designs still need to happen through designers. So it's important that the designers are still skilled using Photoshop or any other tool to do that. Then you also have AI data, data analysts. So the people that actually do the large that make sure that all the information that is gathered is of the quality that you expect, but also formatted in a way that the algorithms can use and analyze it and make sure that statistic relevance is ensured. And then finally you want to loop in legal and AI ethics, ensuring that everything is compliant to the boundaries that you've defined for your company. Finally, there's an extra which is a prompt library. So this is something, if you don't have this yet, I can definitely encourage you to do so. It's a library of different prompts that you use to get to the outcomes that you're looking for. It can be prompts, but also actually a tool stack that you've built up. And when you're doing this sprint for the first Time I would recommend to document all that so that you have it at the end of the track so that you can repeat the process afterwards. So with that, I'll transition into how to run your own AI powered Sprint. And I think it's fairly simple, right? It starts with selecting the right case. So selecting the right case. I had a lot of questions about that after the latest webinar I ran and what the main question was is, should you go broad or could you use a specific case or one specific product? The answer is it can be both. But in the first instance it's nice to have a more open brief where you talk about entering into a new market or tackling a new consumer segment within a specific category, but allowing to really see the boundaries of the algorithm. That's a really good way to see how far it can bring you and come up with ideas that you might have not seen in the past. It's definitely also possible to work more incremental and focus on one specific product or solution or brand, but it just requires a slightly different setup from a team standpoint. The team that we've highlighted before is what you need. Important to note is you need less people in comparison to what you needed in the past. It's really three people, maybe up to a handful of folks that you need because you're going super fast and everybody's just really focused on making it happen. And a lot of the work is done with a human plus AI approach of, of the AI team. Then you want to build up that AI tool stack and prompt library so something you can create. We've got a lot of tools on the website that you can use or access completely for free, but you can also build up your own. And again, that's something that I would recommend is definitely that you do for yourself. Then you want to collect all relevant data and train the model, ensuring that you have all the question, all the everything set up and the model trained for the activities that you want to do. And then you want to run that Sprint and document those learnings. The it's really important to make sure that you have that Sprint documented because then you can again make it a repeatable process, which is in the end what you would need to make it a scalable solution. And what's important to note here is that you want to think about this Sprint as a way to do one experiment, one test of how you can use AI and see what the outcome is, but there's much more possibilities and that's when you typically do an AI strategy Sprint is when you look at a more high, at a higher, more strategic level of how you are going to change your innovation, your R and D process by enabling AI to be a part of it, meaning making it AI powered or potentially in the future fully autonomous. And again, something that we'd be happy to tell you more about later. Obviously feel free to reach out to us as well on how to run your own AI powered Sprint. Happy to assist you with that. But there's also a lot of tips and tricks on our website. If you go to boardofinnovation.com AI powered innovation sprint, you can find much more and as well a lot of free tools out there. So, so we've curated a list of top AI tools to run an AI powered innovation Sprint. We've actually just released the top 10 tools that we've used lately out there as well. So feel free to have a look and go to AI Tools for Innovation on our Board of Innovation website. And again, to end the webinar, I'd love to explain that next step of autonomous innovation. So we're, we're doing a next webinar on autonomous innovation in two weeks from now, on the 25th of September, so feel free to tune in. So thanks and happy to take any questions and I'll get them through the chat and I'll select a few of them and try to answer them and feel free to connect with me on LinkedIn if you have any additional questions. Always happy to answer them at a later stage. Ok. One of the questions that I've gotten was first of all on ethics. How do you protect proprietary data that you're uploading in ChatGPT to train the large language model? It's a very good and very relevant question. There's different routes. So first of all make sure to select the tools that are the best and most useful for your in line with your IT departments. One element that I can say is that when you use ChatGPT and the Enterprise version you can actually close the model so that the data is not shared with the that the data is not shared with the with the model itself, allowing you to keep that private. What I would recommend though is that you really create an own ChatGPT environment for your own company on Azure, for example, where you really protected from any data staying within the company and you can rest assured that all questions that you'd ask and all data you input within that model stay protected within the environment. So two options on one hand, turning off the data sharing feature in the enterprise version or in the ChatGPT Premium version. And the other option is that you really build your own ChatGPT environments within your company and turning off all those features and making sure that all data stays protected. I'll be able to take one more question and I see there's there's one more question on the diversity idea of ideas. So again, a question I've had quite a few times is when people experiment with AI, they often get pretty generic answers at first. If you analyze a lot of research, the for example opportunity spaces it would come up with with your first prompting will be quite generic. Same with ideas is you see that the ideas it comes up with at first might not be the most innovative ones. And that's again, the power of the algorithm is that it can do a lot, that it can go for quantity and therefore you need to ask it for much more ideas. You ask it for extremes, you ask it for segments, subsegments of your research so then it can come up with new ideas that it might have not come up with in in the past. So make sure to continue to prompt and to build on ideas that it has generated and tell it what it needs to do in order to come to become more creative. I'll end off with a final question on how would you describe the key behavioral success factors that can prevent the sprint to be successful? The It's a very good question. I think first of all, at this stage it's important to still see it as human plus AI. It's not a fully synthetic track. It's really done using humans as well. So often what we do is beyond using purely synthetic insights or analyzing research using generative AI, we also do a few qualitative interviews to get to insights that might not be the main pattern. So often insights are created or the aha moment comes from one or a few interviews that give you some sort of a spark. And that's very hard to trigger the algorithm to come up with that. So at first it's absolutely interesting to also do a few human insights to get a true behavioral perspective on consumers for sure. So definitely recommend to augment your synthetic research with human research. So with that, I'll end the webinar here today and thanks everyone for tuning in. And again, feel free to reach out and I'll be happy to stay in touch and answer any questions you might have at a later stage. Thanks.