 Hello everybody, Ola Comestas. I think I'm saying it right, but I'm managing in Spain for so long and I just love it. Thank you all for putting up such a great show. Before I start, I do want to acknowledge the organizers, especially Lydia, who's been changing at least 50 emails with each of the speakers here and Manuel to make this go flawlessly and all the team behind. Let's give it up for the team today. Thank you so much. So today I'm going to talk on three big pillars. The first one is investment trends in AI. The second one is the practical use cases that you see, because AI is such a big buzzword now and everybody's talking about it. What is practically possible in companies? And third, if you do like the first two, how can you go tomorrow and absorb AI in your organization? I judged it a great job describing this, so I'm going to skip through this, but I lead the AI teams at Target, which is a retailer competitor to Walmart in the US. So we will be talking, like I mentioned, the three themes and we would predict the future growth areas based on the investment themes. It could be autonomous vehicles. It could be your visual recognition software or it could be healthcare. We will look at that from my perspective and then we will think about the broad use cases and applications of AI. $1.5 billion. Does anybody have a clue of why I put this number here? Anyone wants to guess? I'm not giving away a gift or neither is Big Data Spain giving this away. It's what the European Commission has promised to invest in AI technology and boost it up by 70% to match US and Asia. They are lagging behind now. There's 400 AI companies in Europe, but they need to do a lot more to be on the center stage. By 2020, they expect private and public investments in AI to total about $20 billion to be competitive and prevent a brain drain from Europe. Before we go any further, I know you all are experts and you've learned to different definitions of AI, but for our purposes for this talk, I'm defining AI as the ability of the machine to do what the human does, but there's two important key things here marked in red, which is the learning and the inference piece. So when you want the machine to learn and infer something, that's AI. It has three broad pieces. It has the data preparation piece. It has the analysis piece and the decision and modeling piece. That's what I broadly call the learning plus inference or AI for our discussion. The final piece is the action because once you infer something, you can act on it. You can act by describing, by prescribing, predicting, or reacting, just like us humans would do, right? So if you, for example, you have a diabetic patient's health record in front of you, you look at it, you describe that, that's description. You predict something saying it's gonna get worse or not worse, that's your predictive action. You prescribe a diet plan based on that person that's prescriptive and reactive is based on what the person does as a health goal and what you do in return of that. So let's first look at the investment themes and follow the money. When I started presenting in this topic on investments in 2016, US was leading investments and Silicon Valley, the place where I'm based was about 50% of those US investments. But within a span of a year, China with its support from government investment themes is now dominating the global AI fund with 48 to 50% of the global AI investments. We'll see where this money is flowing, but to start off, you can see how within a span of a year that China is now dominating this investment trend. From my perspective of looking at various different startups, this is where the money is flowing. There's been about 15.5 billion approximately in 2012 to 2017. And the broad areas of where it's flowing is auto tech and manufacturing is the first one. We all see driverless cars. In the auto technology, security, safety, the intelligence in the car, connected car intelligence, fleet management, services like your Uber lift, parking services, all those are part of that big bucket. Manufacturing, intelligent manufacturing is another big part of this bucket. Amazon already has over 100,000 robots in its manufacturing warehouses. Computer assisted robots are quite common. In Arkansas state, Adidas has actually signed the contract or MOU with a company which does robotic sewing for its t-shirts and it's reduced the cost of employment drastically for Adidas. This is also, you can see for manufacturing 4.0, you can also see farmers and farming at a large scale adopt AI so that it's personalized for each plant that you're not wasting resources and you're giving each plant the tender care because it's done through AI. The facial recognition and cybersecurity, this is where China's major investment comes through. This is where they are investing in deep learning hardware and software. There's a project called Zou Liang or Sharp Ice which is a pilot project where they're creating a social credit score for every Chinese citizen. Like Face++ is a company which has about 1.3 billion faces which have already been recognized of Chinese citizens so you're monitoring the video surveillance in public places. Are they stopping at stop signs? Are they paying their loans on time? Are they defaulting? And based on that, you're creating a social credit score and that's why Chinese companies are investing heavily in anything which is deep learning based or hardware related to visual recognition. Expert automation and augmentation, there's a lot of talk about employees losing their jobs but this is more about augmenting their jobs. Take for example, talent management. You want to recruit the best talents but you have to scan millions of resumes as an individual and it's gonna be difficult whereas if you use AI software, you're able to get a small group of those and even for the speaker series out of the whole world you can target who you need for something like Big Data Spain with using these assisted softwares. Or for legal assistant, another use case is you have to scan so many similar cases for criminal cases for preparation. If you use NLP search, it tells you what are all the similar cases based on what you are addressing or you are suggesting. So it reduces time and it makes it much more easier for people to augment their jobs and do a much better job. Healthcare diagnosis and applied AI, this is kind of the oldest discussion piece here. This involves disease diagnosis, this involves predictive analytics, remote monitoring of your patients, pharmaceutical analysis, drug development and clinical trial management. There's a lot of regulatory issues here because if you diagnose my skin has a lesion and it's something who is responsible? Is the AI machine responsible or is the doctor responsible? So that's legal issues are there and those are being solved as we speak. Finally, you have so many IoT products these days like your driverless cars or you have these Nest security cameras, Ring security cameras, which need a prompt action right then and there. You can't wait for the data to flow to the cloud and come back slowly. So you need computation when somebody suspicious is walking around your house and you have a webcam looking at it, you want the action or alert right then and there. So you need AI on the edge and that's a big investment area as well. So like I mentioned, there are about 5.4 billion which I tracked from 2012 to 2017 with a GDP impact of 15 trillion by 2030. Now, if you look at the corporate side, the Googles and the Amazons of the world, they are also investing heavily in AI but their investments is more from the perspective of a lot of acquisitions. So there were 120 exits in AI just in the last part of the year of 2017 and 115 of those startup exits were acquisitions. On an average rate, the big five companies, the Amazon, Google, Facebook, Microsoft, Apple were acquiring one company a month in this space because they wanted to remain competitive. They want to become the do-it-yourself platform leader when compared to like a Sage maker in Amazon or you know, TensorFlow and Google. So that's something which they're obviously fighting over. Voice is a new text. When I redo this slide again in the next couple of months, this will be so part of the process that I don't even need to put it here because everybody is gonna think and talk without typing, they're gonna use text and voice-based searches, voice interaction with the Alexas and Googles. So that's the new normal. So you don't even have to call it out. Business process automation, energy cost cutting or like form filling, those are other things which these big companies are investing in. Google saved a lot on energy costs when it acquired the DeepMind project. So that's something they're constantly looking at cost savings and business process automation. Finally, there's talent wars. Data scientists are really hard to find and there's about 180,000 jobs if you go to Indeed today and look for AI, ML, data science. There's like 180,000 open jobs. And it's very difficult to retain these folks. So that's why it's exciting to see as you'll see today that there are platforms which are making data science so easy that anybody who's a citizen's data science or a business analyst can also use and become experts. Now that we've looked at the broad investment trends, let's think about where it's being used. So if you look across the industry impact and this was a McKinsey study, you can see that retail has the highest impact with about 600 billion followed by transport and logistics as well as travel and food. If you take a retail, the common use cases are customer marketing, customer segmentation, sales and promotions, supply chain, demand forecasting, recommendations engine, as well as customer service. In transportation and logistics, route optimization, inventory management, those are common themes. Travel and food, I mean, we've all traveled in airlines, there's a lot they can do with optimizing routes and optimizing our operations. So that's something they are looking at as well as food, creative food and personalization. There's like a whole industry, impossible foods is all plant-based meat and it's all done genetically and scientifically and they are selling about 500,000 pounds of meat just recently, they've signed a contract and then there are companies which even do this where they can monitor each egg which is created using AI so that it's used for baking versus fried eggs and so on. So the purpose is so different that they're specialized in using AI for that. Healthcare, we talked about disease diagnosis and because of the regulatory compliance issues, it's not adopted so widely. Instead, what is happening is triage management, claims form management, all those pieces are scribing in the hospital where you have a human scribe today taking notes and we are trying to replace that with AI taking notes so that a human just predicting or correcting those notes. And in banking, the previous speaker was from the industry so we saw a lot there but risk prediction, underwriting, marketing, those are all things which you can see in banking and insurance claims fraud and pricing the insurance are common use cases as well. If you flip it the other way around, if you think about common AI techniques, anomaly detection is one of the common building blocks of AI where anybody can use it and if you look at it across the industry, fraud protection is one of the use cases that can be widely used across industries. It could be through network traffic, patterns which you're seeing where it's abnormal or it could be through payment fraud, insider trading in the finance organizations. In telco, you can even think about sell tower failure, predictive failures and then in industrial IoT or manufacturing machine settings, those are also anomaly detection. Then if you look at the other technique in route optimization, logistics and inventory management can benefit great from that. Fuel efficiency, everybody wants fuel efficiency if you manage a fleet, that's another use case for route optimization or insurance drones, they could use that as well and autonomous vehicle routing is another thing. Voice as a service, we talked about it like chatbots are becoming the norm now where when you're dialing the customer service agents, the first thing you do is having chatbots or in the retail customer journey, you're actually having a chatbot helping you predict the right pattern or predict the right product for you based on your needs. Alexa Google, we ask more questions to them than we do to our spouses or parents and so on. That's kind of becoming the trend these days. So they become part of our life. There are, the AI and voice is so nuanced that it can even predict the emotional tone of a customer. So when you have customer support and if the customer is very angry, then it directly routes them to your human life person. So it's capable of detecting even that tone. Visual recognition, this is where the majority of funding is from the China and there you have anomaly detection, autonomous vehicles are using visual detection as well. Border patrol is another use case and security is another common use case. So if you can see there's like a lot of use case, obviously this is not all comprehensive in the time slot but you can see where in your industries, places where AI can make an impact even in one small area. So based on that, if you see a need of how you can adopt AI in your industry, here's what you would do. So tomorrow if you go and your boss comes and tells you, I want to do AI to solve this or improve cost or increase revenue, where do you start? The first thing is to identify a customer problem. Many times what happens is you have these really smart PhD data scientists but they're doing technology for the sake of doing technology. There's no customer problem, there's no need. You need to find a need, you need to find a customer problem first. Start small, be it either automating something or saving the cost of something in Target, for example, recommendations. We started with recommendations. We didn't have that before five, six years ago and so how do you recommend the right products to the users, right? So start there and use the commonly known algorithms like collaborative filtering. People who bought this also bought this. People who viewed this also viewed this. Simple proven methodologies take that and implement and see what happens. Once you prove the value or you prove the ROI on that, then you can start talking to different departments in your company to see how you can help them. So AI is a platform in your company and you talk to different solution partners to see how you can bring it to their channels. And if you start proving results over and over again, then other people in their organizations are gonna hear about your group that they will start adopting it. In fact, the other extreme is they themselves create an AI organization internally and build their own solutions. So the important thing here for all this is when you start, it's like a startup. You start with one problem and you grow it but you need to have solid product management team which is following some practices and having the business and customer pain points in mind while you're developing this. So sample retail evolution is when you started shopping several years ago, it's gonna be door-to-door salesman or in-store where you're going to the brick-and-mortar store. Then you went online and mobile platforms. Then you have crowdsourced marketplaces like eBay, Etsy, Facebook, and Nextdoor and so on. Right, Amazon as well. Then now you're in the face of predictive marketplaces. Predictive marketplaces where Alexa and Google can start predicting what your next purchase is going to be because you're ordering through Alexa, for example. And then there's IoT products like Dash. You're the Dash button from Amazon where you're indicating your interest to buy a product when it's almost done. Eventually, Amazon can start recognizing it and recommend these to you rather than you pressing the button. A vending machine is also another use case where when the vending machine is almost out, you know when to refill it and to automatically reorder that from your suppliers. And now the next phase is robotics. We talk about inventory robotics already in the warehouses, but there's also humanoid robots like pepper robots which can help you make your dress decisions or just interact with you to remind you for different purchases and so on. So, given the scope, let's say you've identified a problem. Where do you start? The first question to ask in your company is the data captured. And this is this whole world of data lakes, warehousing, databases, extraction, transformation, ETL processes, right? Once you have that, then let's assume you have all the data in different places. How do you then prepare the data? Do you have access to it? Can you transform it? And then are your data scientists able to explore it? Many vendors here talked about the notebook capability. The notebook is essentially like an embedded visual tool which will give them when you're playing with data all the patterns without you going and coding something there, right? So, can they explore those things while you're looking at data from these different places? Then analysis, let's say you're exploring it and I have a problem where I need to find how many customers are going to churn this month? How many customers in this bucket are going to churn or are going to leave Target this month, right? So, what type of features am I going to select for this? Essentially, what are their past purchases? When was the last past purchases? Were they here within the last month or so? How often are they spending time on target.com or on the mobile app and so on? What is their engagement level? So, you choose what features go into the model. You can also see your solution structure to answer the question. The question is important. And then, how is it going to be parametrized? And each of these things which I've drawn here have their own tools or have their own software vendors which can help you with this process. That's why it's really complicated. And then, the decision model, like let's say you have the data, you have the features that go into the model, then you think about how are you going to select the model and how are you going to train the model? How are you going to develop the model? All those things are important. And finally, acting is here in purpose for this. It's like integrating with your code base, creating a service around it and deploying, developing and performance management of that. Outside of this whole area, you have the optimization. This is where your A-B testing tools come, like is your model better than the existing model? That's where you're doing the A-B testing. And there's even a whole new field now of continuous optimization. There are companies which are saying A-B testing is old and you're actually missing out on opportunity in the test which failed because there are people who still like your solution. So, we'll continuously optimize for people in each of those cells, right? And we look at some of those vendors as well. I define AI as a platform in the bucket which is from data preparation to actual deployment in the production for the purposes of our discussion. So, the common question I get asked is, should I build a partner? And I think the answer is it depends, but I think there is a central ground where you still control your data, where you still have AI as a service and your internal technology engineers and data engineers or regular front-end engineers and back-end engineers can interact easily with platforms which offer simplified AI solutions. So, you don't have to have a separate, highly talented AI pool because they could be focused on research, but for your day-to-day AI for business solutions, you can start thinking about AI as a service whether you're not fully outsourcing it or you're not fully doing it in-house and somewhere in the middle. And on a case-by-case basis, you can decide which direction you need to go. So, going back to our steps in the process. So, if you think about data capture, some of the names which come to mind in this are Informatica, Trifect, Talent, and they have a good bar here, so kudos to that. Then you think about data preparation. You have DataIQ, Altrix, TIPCO, and then the last two I met it and figure eight are more in the space of labeling your data because you need to label your data and it's as good as what you need to do and so that's what they do. Analysis, we talked about feature selection, parameterization. So, SigGopt is a company which is a startup just exclusively focused on telling you whether you chose the right parameters or not for your model, right? And that's their only problem solution. Then you have other standard platforms like your Teradata, SageMaker, TensorFlow, which I love you to do that as well. And decision models, here you have end-to-end platforms like H2O.ai, Databricks, DataRobot, Paytom, and then you have other algorithm platforms, I call them like MapR, TensorFlow, Spark, and SageMaker. And finally, if you want to deploy it to production, there are companies which do that like ParallelM, DataRobot, and H2O.ai as well, but primarily you are thinking about even if you want to, a custom DevOps if you have that capability. Finally, we talked about optimization. Optimization, whether you use AI or not, is important for a company and that's the whole A-B testing concept and seeing whether the current normal is better or you want to think about your new change. Did that make a difference or impact? So then you use optimistically, site catalyst. And so scale inference and dynamic yield are the new players I mentioned which are doing continuous optimization. So they're not A-B testing, but they're actually saying for every person who comes to your website, I can choose the right experience for them because I'm learning and I'm giving them the best option. It's kind of like a multi-arm bandit problem where I'm giving the ones which are reacting well more resources and the ones which are not reacting, I'm giving them a different experience and that's kind of how they're using it. And your standard BI tools like Tableau and DOMO are meant for your business purposes, your business executives to look at it. So if you look at AI as a platform, there are several people who are doing from data preparations to the production and these are some companies I'm mentioning here which are well-funded and they are leaders in their space trying to solve known industry vertical problems. So today if you wanna go back and do a five-step transformation in AI, the first step is your digital piece and the data in place, that's the first question to ask. If you don't have a digital platform, then don't even talk about AI, right, because it's not gonna work. So you wanna make sure that your company has the digital transformation and the appropriate data management pieces in place before attempting this journey. Second is pick a valid use case that you can choose and you can prove and have the right measurement metrics for this. If you don't have in-house talent, choose a vendor and try to prove this easily. You can probably get it done in two months or three months at the most. While you're doing this, parallel educate your workforce. You have so many tools which are even free, these days, like in Coursera, like just last week before I came here, we had a meeting with Coursera and Andrew Ng and their CEO had a panel for few executives where they were talking about how do you engage your talent for, should everybody have a PhD? And their view is you don't need it. You can use Coursera courses so that everybody who's doing engineering in the company can become well-versed in the concepts of AI for your business goals, right? And then there are deep learning courses as well which Andrew Ng has launched. Those are useful as well. If you wanna get more technical then your engineering force needs to read those papers of what's new in this field so that they can adopt it. And then play with the simplification tools. You don't have to do everything alone. You don't have to do random tools. Take these platforms which are there which have simplified the intuitive user experience for you right from data prep all the way to data deployments so that it's easy to see the entity models and the data. It's easy to see the visualizations in the data and pick one of the platform that suits you well. And yeah, like I mentioned, you wanna democratize AI which means you wanna make it simple and not have it only in your data scientists' laptop or in data scientists' computers, right? You want it across the board and you want your citizens' data scientists or business analysts or front-end or back-end engineers to be able to use due data science. Finally, I would say start with small use cases, prove the ROI, and then build out larger use cases. Thank you. For the question. Okay. So, if there is any question, this is your moment. Wow, this is really a silent group. Either it just went past the heads or they're just being too nice. Oh, there we go. The brave soul. The first row. They are bringing to you the microphone. Thank you, Arti, for this wonderful talk. One question. At Target, what are kind of the most recent or the most exciting AI initiatives that you are currently doing? So, there's like a bunch of things. I would say the standard one which we've already capitalized on is recommendations. For Amazon, for example, it's 35% of their revenue. For Target, it's a big chunk as well. So, we use AI for recommendations. That's a no-brainer. Then we use it for demand forecasting, inventory management. And there's always other areas which are by categories which I'm trying to explore now. So, for example, you have a baby category where somebody comes into the store and looks at cribs. But I can only have so much shelf space in the store. I can't put all the cribs and baby seats there. So, I have a back office problems of how do I stack the cribs and how do I stack baby products? That's something I'm looking at as well. And that's sometimes I even partner with external companies which are in the ARVR space to help me solve those kind of things. Thank you. You're welcome. All right. Thank you. Oh, that's okay. That's great.