 Thank you. Hi, guys. Thanks for coming tonight. And tonight, we just skip the blockchain. Wrong direction. OK, we want to talk about AI strategy and techniques for leaders. This is actually for leaders and product managers. So I wonder how many of you are product managers, Steve? Nice. And how many are going to become a product manager? It's also nice. How many of you are entrepreneurs? Oh, we have quite a bit of entrepreneurs. That's nice. Now, I worked at Yahoo before as a product manager, senior product manager. And then we formed with my partner, Moot AI. And Moot AI is an AI consulting company. And what we do, we go to mid-sized businesses and analyze the business requirements and analyze the business, then actually provide them solutions. So unlike regular technology, technology solution companies, we actually approach from the product side first. So look at the product. So that will be more likely to your liking. Now, today I want to talk about a couple things. And we are not going to talk about a couple other things. So we are not going to talk about why AI. I assume already you already know that why companies have to move to AI or why they are moving AI today. And we are also not going to talk about algorithms. You can find lots of other presentations, also lots of information, which algorithm to use, which framework to use, or which technology to use, so to say, to run your AI solution. But what we are going to talk about, and these kind of presentations are very rare nowadays, how can you strategize AI in your business? And that applies especially for product managers today, because product managers own the products and they can understand how to apply technology to these products. And when we think about strategy and business, we are looking at five different aspects. So the first one is people, the people aspect of the business, then the practice aspect of business, then the solution aspect of a business, and environment, and partnership aspect of the business. Now, I said we are not going to talk about why AI. I just have one slide. Today, a business is either integrating AI technology or they just die. It's a no-brainer situation nowadays. It's a fourth industrial revolution. So if the company, the product manager, the CEO, CIO, if they don't have a strategy in mind to implement AI in their different processes, the company is going to die in the future, because there is huge competition going on. If you look at these numbers here, for example, healthcare companies who have AI strategy have about 15% to 20% competitive advantage over all other companies. And these numbers are, these margins are really interesting because most of the industries, their margins are not that high. There is no like 80% margin. So we're talking about here a couple percent of margin. It makes a big difference, especially in areas like tourism or package good retail. Now, there is one more reason I want to really underline. These are just two slides about why AI. And the first one was, okay, you have to do it because there's competition. Second one is coming from Jeffrey Moore from the crossing the chasm. Product manager's most proud to know. How many people know Jeffrey Moore? Right, almost everybody knows Jeffrey Moore. So crossing the chasm, very, very famous book. And he is a really good point here. He says in the U.S. especially, we will never have the chance or we will never have the luxury to have the lowest cost of labor. That's not gonna happen. But what we have is we must continue to exploit advantages further up the value chain. That means new technologies like AI, new technologies like IoT, or any other like even blockchain, those technologies, we have to really push the envelope every time. And that is our responsibility as product managers. Now, when we look at businesses, and we did that with my partner in the last six months, we went to like 50, 60 mid-sized businesses. We talked to C-level executives, we talked to VPs. And the problem we have seen is that the companies look from the business perspective and that's what we named business domain. We talk about revenue, we talk about cost, we talk about marketing, and we talk about partnership. But when you look at it from the technology perspective, like AI technologies, this is the language, image recognition, recommendation engine and which algorithm to use, what prediction, what is the bias in these algorithms? And what we've seen is that there is a big gap. Companies don't know how to jump this gap. And honestly, technical people also cannot figure out how to jump this gap here to the other side. And there's a big opportunity we see that, especially for product managers and service companies, to close the gap, to connect the business domain to the technology domain. And that is also what's happening in bigger companies today, like leaders like Google, Amazon, Yahoo, these companies can understand the business domain as well as has the technology side and there are lots of product managers who close the gap. Now, what we shouldn't be doing is this. So you have a business here, your current profit, and hopefully this is your target profit, it's much higher. And picking a technology and sticking with the technology is not the solution. The solution is not picking Keras and saying, okay, I'm going to bring AI to my company. Or I'm going to pick TensorFlow and that will solve the solution. In fact, today, if you go to some companies, like if you consult with Amazon, of course they're suggesting their own technology base. They're giving you that one tool which you think maybe it will solve all your solutions, but this is not the case. And this is what we name, of course, mouse loss law, hammering the problem. So you have a technology and you try to hammer it. We also see it in smaller companies, like we had a client and they had one data scientist who could use only one language and the company couldn't scale because that language didn't allow scalability in the platform. And they couldn't, oh really, sorry. There was a speaker, there was a mic actually somewhere here, right? Brian, do you have the mic? No, this is for recording. Oh, there we go. Thank you. Test, all right, here we go. Better? Wonderful. Now, picking a technology and trying to solve all your codification is actually not the solution. So what do we need to do? We have to change the strategy. And Henry Minsky, the famous management scientist, mentioned about this, doesn't end the strategic change. Very simple, there's nothing smart about it, but the idea is that it's strategy is an organization's response to environmental change. And AI is an environmental change, not just from competition perspective, also from other perspectives. Now, how it works is that very, very simple graphic here, there's an intended strategy, of course, there are some strategies dropped and then emergent strategy comes in and in this case, AI is the emergent strategy and you join it. This is the definition. He talks more about it, but basically we need a strategic change. And what is the strategic change? We name it, Cognification. I don't know where this originates from, but it's interesting word which summarizes lots of things into one, Cognification, meaning that your company can utilize AI technologies to gain competitive advantage, even become a market leader. So we're gonna dive a little deeper into Cognification, but not deep because there are many, many broad concepts around it and we're going to talk a little about it. And when we think about Cognification in businesses, we really think about these five aspects. We mentioned people, practices, solutions, environment and relationship. Now, let's dive a little deeper into it. Let's start with the people aspect. And the people aspect is really important because Cognification is about everybody in the business. It's not just a data science team or it's not just a solution team. We are talking about finance, sales, IT, QA. Everybody has to really have put the AI head on their head and then come up with solutions and solve their problems using AI. And the most important part is on the shoulders of the PM. And we name it AI PM, Artificial Intelligence Product Manager. And then AI Artificial Intelligence Product Manager differs from a regular product manager by these two specific traits. One is a specific AI solution understanding. The other is AI product lifecycle knowledge. One is about execution, the other is about strategy. But in order to do this, you have to be a good product manager, have the core product management skills and that is actually what product school does, teaches you the product management skills. Then in those two specific domain expertise, that's something you need in order to actually bring those solutions to market in that company. And this is usually something either you learn or you study but there's not many schools about industry specific domains. Now, if you fulfill all these traits, we name it Artificial Intelligence Product Manager today. Now, why do we need AI product managers today? Because technology is different. AI technology is not like a regular software. I mean, you don't write AI code like you would write in the past, regular software code with rules. In this case, we are talking about teaching the computer to do something instead of telling the computer what to do. It's a totally different mindset especially from engineering side and that requires actually new understanding for the product manager to be able to use new methodologies to drive AI solutions. Another thing is teams, as I said, teams are changing, there are different teams. We're gonna go a little more into deeper into that. Then businesses, the processes are different. AI processes, we're talking about decision support system, we're talking about decision making system. If you bring a decision making system in a process which started working, you most probably gonna cut that process somewhere and disconnect from human. That's why businesses, business processes are changing. That's why we need artificial intelligence product managers. And also effects, there's lots of things going on about AI and security and privacy. Nowadays, you most probably heard about it. That's another reason. And we know that today there's a huge shortcoming. There's a shortcoming on product managers because 10 years ago, product manager was the guy who would actually decide what color the Kellogg's box should be, right? That's what the product manager. But today, product manager leads to products like big products, technology products, like search engines, driving cars, self-driving cars. I mean, if there is no product manager, you cannot develop these kind of solutions today. And that's why there is a need, but there is also lack. We need more product managers who understand these technologies, at least from the business perspective. And there's no school for it. I don't remember. Is there AI product manager school somewhere? So that's why it will come most probably, but it will take time. Then industry experience, it is really varying. One project doesn't apply to another industry all the time. You're working on some self-driving car solution. You most probably cannot work on other NLP related technologies. So technologies kind of change, but the core product management skills won't change. And there are other things like, there is no standard right now. It is the frameworks are coming together. It is really on the rise right now, so to say. Now, if you look at a team, like how does a production team look like in an AI organization? And that's something we have seen in larger organizations. Also, smaller organizations trying, I mean, mid-sized businesses try to get there. And so we are talking about a CAIO nowadays. And Andrew Andrew also points us out. Every company in the future will have a CAIO. It's not the CTO, it's a new AI officers, chief AI officer. And that person leads, that role leads the AI initiative. Then AI product manager and AI product owner. These are really interesting, two new interesting roles which applies to you all. Which we need to be able to drive AI products. And we have AI architect. You can find today AI architect. It's not that hard, but it is much harder to find AI product managers. Data scientists, many schools actually teach data science. So it is easier to find data scientists and data engineers who deal with data, who actually clean data or bring data and it's actually much easier to find. And here are a couple of strategies. If you are a product manager and you have a plan to integrate or use some AI technology and you don't have the talent, there are ways to get them. There's a hiring strategy, training strategy, consulting. And this is very interesting, Cognify. There are tools and technologies today, you can replace some of the data science job, not the whole data science job, but part of it. And in fact, I was talking to somebody in one of the meetings and they gave a job to a data scientist. It took him like one month to do it. Then they found online, there is a tool named DataRobot, you may have heard of it. And he could do it in one week using DataRobot. And it's gonna get more and more simpler. So once the tools and technology will improve, we will see more and more cognition here. And there are ways to acquire these. For example, you can send your current engineers, your scientists to deeplearning.ai course in Coursera, it's pretty good, Andrew Eng is giving this. It's like a three, four months course and they can get up to speed with certain things like machine learning algorithms, AI algorithms, that's a pretty good course. Then on the consulting side, you can actually consult with these companies. Now NVIDIA, Amazon, Google, and Microsoft, they started creating these special teams who support their platforms. And if you're a big company, of course they work with top 500 or top 1000 companies, you can get consulting and they will take care of your AI needs. They will actually create your architectures. And I also put our self year mood to AI in different places. We provide the support for mid-sized, small to mid-sized businesses unlike these companies provides for the bigger sizes. And cognition, of course, DataRobot, cognitive scale, crowdflow, they changed the name now import.io, these kind of companies will be more and more in the future. And we will see actually they solve these AI problems on their side. And these tools are really important for you, for product managers to be able to pick the right tool. If you need something fast done, maybe with 90% accuracy, okay, not 95% accuracy, you may try it out on DataRobot or cognitive scale and solve the problem. If you have to hire somebody, it's gonna take you three months compared to that, okay? Sure. Could you define the cognition once again? I love track of it. Of course, cognition is implementing AI technologies, using AI technologies in a company, in your whole company, in your whole business, such that you gain competitive advantage or become a market leader in that area. And we're going to talk about the process, like what I mean by business converting, where to use AI, right? Now, let's talk about practices, right, Tom Point? And practices, when we talk, when we say practices, practices are business practices. And when you look at the company, when you look at the business, you can look at the product manager or as an entrepreneur or a service provider or consultant. From the AI perspective, you're looking for opportunities. So you ask yourself, how can I cognify this business? That means how can I bring AI in to gain advantage to actually improve something? And there are really four different ways you can bring advantage to a company. And one is automation. Automation means something is working with human power right now. And you want to bring AI to convert it to some technology. And it doesn't mean there's a software solution. Today, most of the problems don't have a software solution because software, rule-based systems don't apply to it. But you can find AI solutions to those problems. That is automation. You find the working process, it works great, but maybe there's high human error or maybe there is high cost and you can bring in automation. Another one is optimization of the objectives. So that means you have a working system already, an AI system, which is suboptimal and you're bringing some new AI technology which is much better and you're increasing the performance somehow. Another one is expansion opportunities. So it happens actually with bigger companies. Yahoo, we were expanding internationally. So we went and deployed to 30 different markets. That requires a different perspective. You need new technology, new hardware, and you have to implement it, expand your technology. We had to rewrite our recommendation engine. And that actually is an opportunity there for AI. And the last one is innovation opportunities. That is if you're a research team, if you're looking like five years out, like 10 years out and you want to have competitive advantage, want to be market leader like Google does, they're working on projects like 10 years out, then there is a big opportunity, you can use AI technologies. Of course, all of this comes with a cost and there's some complexity associated with it. And the complexity increases towards right. So an automation project opportunity is fairly, comparably, in general, easier than an optimization or expansion opportunity. And of course, innovation is the most uncertain one. It is more complex, the complex comes from there. And this complexity can be anywhere in the organization. It can be organizational, it can be technological. It doesn't mean the complexity is coming from the algorithm. I mean, the complexity sometimes comes from the organization. For example, you want to automate a human process and people maybe are not okay with that. Now you have to deal with it and explain everybody that what you're bringing to the table or it could be regulations. We started deploying to Italy recommendation engine and then we figured out Italy doesn't allow you to crawl websites and put them on the website. So there are interesting regulations around the world also in the US, especially nowadays with privacy and security. You have to, this can become really, really complicated and as a product manager, you have to take care of it. Nobody else will take care. Engineer won't look at it. Scientist won't look at it. CEO won't look at it. Product manager has to take care of all of these issues here, okay? And if you look a little more deeper where the complexity originates from, I'm just going to point out a couple here. For example, optimization. Optimization is really a technology bound, comparably technology bound process. If you find the optimization, you really have to work on the last couple of percent sometimes and that will actually take lots of time. And expansion, expansion opportunity complexity usually comes from regulation and organization, especially if it is geographical. And of course innovations, you don't know. It can be any of them depends on the innovation. But usually when we talk about innovation, we're looking at a really long term. And here are a couple of methodologies. These methodologies were created a long time ago and we were talking about Azure a really long time ago or design thinking, GV design sprint, just maybe a year, I don't know when they start, but I just saw it this year. Now there will be maybe newer methodologies. I don't know, related to AI because the process is a little different. But what you can do today, you can use a combination of these and create your own methodology. For example, for automation, you may mix up agile and GV design sprint. Especially GV design sprint gives you good view. That's a new method they're using. But what we came up with is this. We name it agile AI product lifecycle. And it's a flexible framework where we then remove things. But the idea is that first, you have to collect the information requirement and design sprint does it in five days. How many of you did hear from GV design sprint? Google Venture design sprint? That's a really interesting framework Google start using with their ventures. And it's fairly new, maybe six months, I don't know. I saw it like a couple of months ago and I started working on it. And in five days, you come up with a prototype. It's so fast. And they say there is no project on earth. You cannot do it in five weeks. They even tried it on some jet engine project. It even worked in manufacturing. Now that gives you really good head start. Then requirement analysis, sometimes it's a short, sometimes it's a couple of days, sometimes it's two weeks. But the main idea is here that AI projects have a second cycle here which is optimization, which is experimentation. You try to find the right solution. AI solutions are about search. You're searching for a solution. So it's not like writing software where you have the rules. You just write the rules, then you test, test, test. I mean, maybe you introduce new features here. You already have the code. You just try to optimize it. You're searching for the best algorithm. Sometimes you're running 20 algorithms. Sometimes you're changing the data multiple times. So this process is interesting. This doesn't exist in other processes. Then of course, once you're done with it, there's an engineering cycle. And this whole process is a cycle on its own. And you can add or move to this. And we have seen actually a good advantage using this, especially the rapid experimentation and AI solution research is an interesting area. Here, for example, with AI solution research, you sometimes go look at GCP solutions, Amazon solutions, or different vendors outside, not just internally. And that was practices. Now let's look at solutions, like from the AI perspective. How do we approach a codification from an AI solution perspective? And first, we need to understand where your business stands in the AI timeline. When you look at a business, when you look at a business, or a business process, so to say, you have to understand where they are so you can come up with a solution. And the AI timeline looks like this. We had handcrafted knowledge. This is software today. We use it. It's as regular, if then else, rule-based systems. You wanna write a thermostat or some controller. You say, is it 5 p.m., turn on the AC? That's so simple, right? And then we have statistical learning that's next up. And this is what we named AI today. This is machine learning. And machine learning system, it has lots of input variables, lots of state, and what it outputs is that, for example, set AC278. It used that information, the input data, to figure out a correlation to the output data. And it tells you what it is. So everybody knows this. Contextual adaptation, that's the next wave. And contextual adaptation is self-generalizing systems. We don't have this much today. We have gone like generative adversarial networks. And it is going towards that. But a system would look like this. You have the context of the whole world. I mean, it can look at the environment. It can input every data. You're not pushing the data. It only consumes, takes the data. It's data hunger systems. And then it has a memory, of course. Today we have similar solutions like LSDM, a long short-term memory systems on recurrent neural networks. But it's still not this size, like not in the world context. And what it does, it sets AC, for example, to 75. And at the same time, it decides to order a script. And that's going to be possible in the future, in the next five years, I believe. And next up is, of course, AGI. And it's an interesting name, Artificial General Intelligence. 10 years ago, this was AI. Now, AI started being used as ML. That's why now we need a new term. And now we have a new term named Artificial General Intelligence. And Artificial General Intelligence means some artificial intelligence which can do a human-like task, but not just do human-like tasks, but also acts free. So it doesn't care about human. It just does what it does because it wants to do it, right, like human. There is no reason. Any other system is a servant to human. It just serves to human. This one doesn't have to serve to human. I don't know how far, but it looks like we are going towards that still. I think there is lots of solutions here. We should be able to create solutions here in the next five years. And then contextual adaptation systems will come along. Now, you determine where you are in the process as a product manager. Let's say you have a system like Nest Thermostat, right? Nest is right now here. If you have Honeywell or Honeywell, what's the name of it? You're most probably a little tourist there. I don't know if they have a new technology. And then you can say, OK, here's my solution. Here's the market. And we are going to run towards this. And now, depending to your solution type, there are really three different solution types from AI perspective. One is task processing. It just does mundane tasks and repetitively. And decision support system. Something supports human decision. Your lane alert in the car when you drive is actually a decision support system. It tells you, stay on the lane. Don't cross the lane. Well, this is a making system on the other end. It decides for you. You don't have to decide. It's your self-driving car. It just drives between the lanes. It doesn't even alert you anymore about the lane. So given these different solutions on the AI are in a different point today than this graphic going to change in the future. But today, if you have a task processing, if you have a task to be processed, you can use any of these systems like adaptive handcraft. It's going to do better than human. I'm repetitive constantly calculating something. And you cannot give it a human to calculate every day the same thing again and again. But I can run a job there on some server. It's going to calculate every day the same report. That's something easy to do today by the same systems. But decision support, it's not at the human level right now. What we see is that if you need a decision support system, you better jump to some adaptive or statistical system to be close. If you have rule-based system like handcrafted software, that's not going to help you to do decision support that much. And if you need a decision-making system, it's even more or, you need an adaptive system for sure. And there are not many decision support systems today. And this is going to improve. There will be systems which make decisions for us. That's what we see. And you've most probably heard of these, like supervised learning, unsupervised learning, or reinforcement learning. That's just from the technology perspective. But let's look at it from product perspective. And that's actually your perspective. If you have a prediction problem, if you look at it, and you need a solution, prediction solution, or if you have an image classification problem, or you need a solution for that, you can pretty much go today mostly with supervised learning. If you have a recommender problem, or compression or targeting problem, you can go with the unsupervised learning solution. And if you're dealing with a solution about dynamic pricing, next best offer, then you could actually go today with reinforcement learning solutions. And the line is kind of blurring nowadays. You can use different solutions at different places. But pretty much what it says is, if you model by mapping input to output, the first one is good. If you're looking for patterns in the data, then you can go for the second one. But if you want to create models, policies relative to environment, then you can go with the last one. And your data scientists, data people will actually know this already, it will give you the required information. That's why I'm not going more deeper into that. And let's see what time, seven or six, good. Environment, when you cognify your business, what are the environmental aspects? And not the environment in terms of nature, but environment of your business. What are those aspects in co-ignification? And the first one is, everybody heard of this most, Robert, the hidden or the black box AI problem. And many companies hit this problem. Here you have the Google example, Google News. Actually, if you look at the world, world embeddings of Google News, Bolut Pesce told it the research, and they figured out that most of the Google News have incredible gender bias. And it thinks actually that men are engineers and women are receptions. But if you use an NLP system, which is trained on world embeddings from Google News, and it's very, very common to train NLP algorithms with world embeddings from online sources, you will end up with a system which has bias. Think of a system or a recruiting system where it looks at CVs and decides, tries to match it to different positions in the company, and you create an AI system. If you don't know the bias, if you are not aware of the bias and nobody still thought about it, then you will end up with a system which has this bias and that will actually create a problem for your company. And the responsibility is on the product measure to think that, okay, we have this model. Did anybody look into this? Like, is this even serving every aspect of my business, not just that high performance or that high accuracy, but also different aspects? Like, is there a bias? And so this is one thing you have to look into it. And there are a couple more from environmental strategy. So enforcing transparency, that's really important. And today you can enforce transparency using tools like LIME. And this tool allows you to look into the model and see why the model made a decision and what was the input, what was the output. You can use tools like this. And this is becoming more and more popular today. Then risk management on AI technologies, that's really important. That's your responsibility. Having a risk management strategy is really important. And adapting safety mindset. Elon Musk supports the open AI. It's a safe artificial general intelligence strategy. And there are many places like Oxford as another group and they're also supporting safe AGI. And that's gonna be more and more important going forward. And you saw on the AI timeline, in the end we have AGI and we don't wanna really end up there surprising that AGI is trying to kill us all. And so that's something important when you create models and create technologies to pay attention to that. Self-driving cars, for example. You don't want them to hit people just because it thought it is easier or cheaper or there's more gases efficient for some reason. These are all product managers' responsibility to think of these. It's not the data scientist's strategy, responsibility. And managing information security. Everybody most probably heard about GDPR and those kind of things. Now there is white-head hackers. Nowadays bigger companies hire these companies, small companies about white-head hacking. And there's also white-head AI hacking. Can your model be hacked? Today they figured out there are many papers about it. You can fake an AI model, especially image recognition, by just changing some pixels on the input image. It thinks some kid is a banana suddenly. If you can do it on the street in front of maybe from the self-driving car which is image recognition, you can actually fake it to think there is somebody or there is no person. So these are really interesting subjects you have to think of when you get to AI solutions and cognify your business. And there is standards like 27.001. It's also really important. This is information security standard, but it will be applied to AI in the future, we believe that. Now let's do the last one, relationships. And relationships is about the partnerships. And the question is when you're cognifying your business, what should you should be looking in terms of partnerships? And there are a couple of strategies around that. And the driving factors, of course, are IP know-how. If you're a fairly mid-sized company or a small company, most probably you don't have hundreds of data scientists or scientists who are writing papers every day. And that means you're going to deal with IP a lot when you want to implement some machine learning model or some AI technology. And that is a good time to partner with other companies who provide these services. And then when it's scaling, due expansion, you can partner with companies who take your technology, AI technology applied to other geolocations. And that will help the time to market so you can get faster to market. License and regulations, that's really important. Privacy concerns, security, transparency. There are companies who are helping with this. So it says you don't really have to create another team, privacy team to deal with it if you don't have a bigger company, but you can hire companies who are going to help with this. And executives, of course. Now, your product manager, you came up with these ideas, you have the solution, you talk with the engineers and you come up with these new groundbreaking solutions. Your executives need to have also the AI thinking. And this is not always possible. Like, you don't maybe have a CAIO, you may have a CTO or a CEO. And in that case, it's also possible to partner with other companies, AI consulting companies, who will provide you the solution domain expertise at the executive level, as well as the AI product lifecycle knowledge at the executive level, that will help a lot. And that's it. We are right on time. And we talked about people, practices and how to identify opportunities. And we talked about solutions, solution domain. And we talked about environmental dangers and the strategies. And we talked about partnerships and the driving factors. I hope you liked it. And this is our company. We are inception program member. And we work with Norchester University and present here as well. So that was a great presentation. We do have a couple minutes to take away questions. If we have questions, we can raise our hand and ask questions right now and dig deeper into the presentation. We've got like 10 minutes, 15 minutes there. But go for you, raise your hand. You do have a couple of drinks left over. So help yourself. At the start of the course, because I wanted to know how to confide your company with that. What's the name of the company? Vigor 8. Vigor 8. Oh, Vigor 8, yeah, CrowdFlower. CrowdFlower. Yeah, this is CrowdFlower. They don't need to figure it out, right? Well, first of all, they're a very, very big company. So that's the biggest difference. And CrowdFlower started like Amazon Mechanical Turk. They were just doing data labeling at the time. But once you label so much data and you create that pipeline, and data is everything for AI technologies, you actually have all the technology for all the enterprise, all the solutions. And they're in a really good position right now in order to provide all those solutions. Yeah, the differences, they're really big. Second of all, yeah, they're much bigger team and they have also much more clients. That's the difference. Another question? Yes. Sorry? Here. Yeah. Is there any specific case study without what would make the transformation happen in a moderately specific case? Specification to do the transformation. Specific case study. Oh, specific case studies. I see, I see. Yes, one case study. We worked on an image recognition, image classification problem. And what the client needed was is be able to recognize wine bottles on the shelves. So if you go to a grocery store, there are like lots of wine bottles. And their role was actually to be able to count those bottles because they sell it to the stores. So the stores has to put it on the shelf and sometimes they disappear from the shelf and the shelf is empty and they wanna see that. And what we did, we worked on the case, we worked on that case. We collected images from the stores and then trained machinery algorithms to detect certain brands. And they were able to go with the phone, with their cell phone and scan the shelves in real time. And it was able to give the number of bottles to them. That was one of the examples. Does that answer? Actually, my partner can give also more details about this, be more there. And other questions? There's one. Okay, the question is, when you move from one industry to another industry? As AIPM. As AIPM. Mm-hmm. Correct, you can leverage your team expertise, like how to work with the team and the methodologies and the way you work with the data science, how you get in the data, how the data needs to be processed. But the missing parts will be definitely about the specific how the technology works. For example, if you have a recommendation engine, like a recommender system where it is, if it's a content-based recommender system, for example, you would, you have different entities. You have your partners who provided the content, so you have to talk to those people, right? You have the understanding to be able to collect that data and make agreements with different businesses. Then as a product manager again, you have to collaborate with the data science team to be able to solve that solution. Then at the same time, you need to have a perspective board on the consumer side, right? How to deal with thousands of maybe millions of users, which is totally different if you have a B2B solution where you have only 10 clients and you don't have actually any other partner and you're working with the data science team only. So that is the biggest difference, that your business model, if you look at your business model, the difference actually will tell you how much you have to learn when you switch. And that's why people coming from the consumer business stay most in the consumer business. If you're in the B2B, you stay in the B2B. We talked about lots of B2B companies and we're coming from a culture where we worked with consumer, like billions of users. I'm talking about 1.5 billion users. It's totally different than having just 10 customers and this is a different problem to solve. One is like scalability. You have to talk with the data engineers, with the engineers out how to scale and other things, how to regulate that kind of user. What's the privacy? Where now you have just a couple of 10 customers and it's much more contained and you can actually enforce more rules, those kind of things. Yes. It's much bigger than the other one. I see media. Yeah, not all of them are there, of course. Yeah, it's media. Which media? News, serving or TV? News on the internet or on TV? Video or online text? All of it, okay. So yeah, when we talk about online media, we're talking about multiple things. Could be really video, audio podcast or video streaming, video podcast or also eSports is coming really hard right now. And you can use AI in different aspects. If it's text-based, you can use AI in a couple of places. First, to recommend, to find the right content, when you need it, where you are. And if it is video, it is more about analysis, understanding what's the video about. Because with text, it's much more easier to run NLP and understand what does this text about. But it is harder to analyze the video and figure out what is it about, right? It could be anything. Same thing with image analysis, understanding what does the image about. And also other things, you want to maybe ban or exclude certain content, which is maybe nudity in it or maybe it has profanity in it. So it is understanding the content, understanding the user and understanding the context. In those three areas, you can use AI. Google News, all Google, everything's Google. So there is a heavy focus on AI influence on it in terms of the recommendation system. I see. Yeah. I see. What's on your solution slide? Solution slide. All right. Next one? This one? All right. Depends to your problem. So what is your case? What are you trying to solve, for example? Or what's the business process? Customer engagement. Online engagement, right? OK. Of course, now, so you're trying to improve your customer engagement or maybe customer acquisition or customer retention, right? Customer retention. OK. Now, in order to do that, you have to pull the data from your own. There are a couple ways. First, you can pull it from your own data source and you can use the Google Firebase. It supports that now, or Google Analytics, sorry, Firebase was for the mobile. But if it's a web, you can use Google Analytics and you can download the data from there. They allow that. If you don't have many users, and let's say you have only 1,000 users or 10,000 users, and that means they're visiting, let's say, 10 times as much data to run machine algorithms, you can buy data. And you can buy user behavioral data. There are many companies who sell data. Of course, this data is anonymized. I mean, you don't know who that person is. There is just a number. But it will give you behavioral information if your website aligns with that data source. So that's also another criteria. I mean, if you're running a car repair shop website and the data from a news website won't help you anything. So you can explore those points. V2 products? V2 products, you're saying you have a closed system where it serves just the business, right? Not online, but OK. And how many customers do you have? Thousands of. Of course, yeah, with that amount of data, if they're not using a lot, then it depends what you're trying to improve there. Are you trying to improve user customer retention on this app? Or so you want to improve the engagement. If you don't have much data, from the AI perspective, there's actually not much to do if you don't have the data. But you can maybe run studies on that and experiment with that and see where it takes. But it totally depends on the solution. So if you're trying to improve user engagement, you can still bring AI technology in there, some smartness. As a PM, you can always use the intuition. If there is some solution which gives you a better user experience, you believe that. And you can try these solutions. We have seen today, for example, use TurboTax. Or there are actually smaller solutions in the accounting. And they come up with interesting AI solutions in the back end to calculate things automatically. So they don't ask everything to you. So at that point, it's more about intuition and experimentation later on.