 Hello everyone. I extend a warm welcome to each one of you, and would like to express my heartfelt gratitude for your presence at today's webinar centered around the theme of artificial intelligence, AI, and machine learning ML for the fintech industry. I'm Vinod Jain, your host and presenter for this enlightening session. In the next few minutes, I will take you through the roadmap of today's presentation, highlighting the key area of focus. Our objective for today is to foster an interactive and engaging environment where we can collectively drive into real-world business challenges and explore how AI and ML can provide solutions. Though this we aim to build a comprehensive understanding of how AI can effectively address the challenges faced by the fintech industry. To provide you with a glimpse of what in store, here is a brief overview of today's agenda. So the agenda consists of introduction. We will kick things off with a short introduction of myself, allowing you to get to know your presenter better. Setting the context, I will spend some time understanding the sudden surge in popularity of AI and ML. Why has it become the buzzword in recent time? Let's find it out. The third will be AI, ML and fintech. Next, I will delve into the core concept of artificial intelligence, machine learning and their intersection with the fintech industry. This will serve as the foundation of our discussion. I will explore a few AI and ML model and construct shedding light on their practical application within the fintech space. I will try to spend time on key drivers and market trends. It's imperative to grasp the driving force and prevailing trends behind the pressing need of AI and ML in fintech sector. I will take a closer look together with you on these aspects. Practical application, then I will roll up our sleeves. We together will do this one and get our hands on together. I will take you through a journey of a business use case, which is fraud protection and embark on the journey of building AI, ML model right here, right now. With that said, let's jump into my self-introduction and start our day. I'm Vinod Jain, Global Product Management with 17 plus years of business and finance industry experience within that six years at the JP Morgan Chase and more than 10 years at the top innovation companies like PayPal, Oracle, DEA Systems, PaymentTec and VNYMallon. There, I have played enterprise-wide roles focused on new product development, strategic transformation, digitalization, fintech startup and setting up client-first experience as primary goal of product and delivering that. Moving further, I would like to spend some time to set up some context which we talked earlier. Some of the most influential voice in the tech industry met with the federal lawmaker on September 13 to discuss regulating the fast-moving artificial intelligence industry. This quote which I picked it out is from that part. There's another quote which I found it out was that with AI, we can't be like ostriches sticking our head in the sand. Another incident is that artificial intelligence has been uttered many number of times which you might have experienced yourself. In this case, 827 times on the 76 sample calls total out of 221. And that is 3.5 times more per call has been mentioned. AI has been mentioned 175 times during Microsoft Q4 earning call. So you can understand immediate focus and need towards this topic and widely talk about the emerging topic or we should say like we are in the midst of AI and ML market. So what is achieved? Global AI market has been predicted approximately $9.5 billion in 2021 and it has been expected to grow 16% year after year till 2030 which is a phenomenal big potential of the growth are there. When we build the only and we want to know what are the key drivers and the biggest driver behind this paradigm shift which some of the expert industry experts is calling it as a paradigm shift in computational industry or in technology is because of there is a endless flow of the data in print that companies and it is impossible to manually track all of those processes and report correctly significantly using the human or manually. That's the one of the driver. Another one is that AI could potentially disrupt the business. A lot of people are talking about these things. Life ranging from the increasing commercial productivity. Some of the folks are talking that fact in the job and national security and intellectual property. So those understanding perception is also driving a big need to understand and digest the AI and machine learning aspect. And third but not the least is print tech industries are aggressively converging in last three years towards either becoming a stable traditional banking aspect and vice versa, the traditional banks are trying to become the startup and print tech aspect. And yet there is a long journey to be achieved in quick time to market aspects and product of features. So why now all of a sudden we started thinking about this topic or this area of technology. If you look at the cumulative spend in the financial crimes and compliance is approximately $214 billion in 2021. And it is growing. It's not slowing down. It's growing exponentially year after year, which is a big amount of money going in financial fighting with the financial crimes or costing that financial. The global fraud exceeded approximately $1 trillion and it is identified that $1 is lost in fraud typically cost four times to the financial services firm who is handling that fraud later on. So you can multiply and do the math how big is the spectrum and scope of it. Same time post COVID and due to digitalization the payment processing and the things done transactions are increasingly expanding on the online and digital payment has gone to 9.5 trillions or approximately more than that and are growing 15% year after year. So now you can think at higher level why all of a sudden this topic become critical and everyone is talking about every sentence, every news and interviews has this word embedded into sentence moving to the next slide. It's a continuation of the context and I wanted to point it out more and more companies are saying that they are using AI in their quarterly and yearly earnings and you can see from this graph that the number of companies which are trying to report and highlight the use of artificial intelligence into their business processes or any kind of their operating processes to showcase that we are not behind the curve but we are adapting, accepting and moving forward in this technology and this revolution in the market. With that said, I assume you all have heard and know some level of information as a product manager about artificial intelligence, machine learnings, financial technologies. So I thought let me spend few minutes refreshing your understanding on this topic along with clarifying some of the terminologies which will help us in the following slides when we go through building a business use case and leveraging the artificial intelligence models towards the end of this presentation. So one common thing which we always mess or mix up is the AI and ML machine learning is interchangeable so the artificial intelligence is not machine learning and not the deep learning but they are connected. So the artificial intelligence in simple English can be defined as performing a task normally required a human intelligence. It's a discipline which we define and machine learning is defined as ability to learn without explicit programming a pattern recognition. If we train a computer program to identify certain kind of patterns understanding is called machine learning because machine is learning which human interventions was required and then going further level down is called the deep learning or DL which is using the artificial neural network allowing them to process more complex patterns than the traditional machine learning. So machine learning as I defined earlier it's a subset of that. So if you see those three circles AI is on the top bigger circle which is the umbrella and the need of that the machine learning is subset of AI and the deep learning and further down and down the other things get rolled under the umbrella of artificial intelligence. Intelligence. The second box or column is more about the most widely used and known to everyone. I'm sure almost all of you have heard about generative AI or Gen AI or the chat GPTs and has used them widely used across the industry which is a subset of a deep learning and the way you can define its artificial intelligence or it's a non artificial intelligence simple mathematical model is a the thing which I shown around the right is the Y equal to fx function of X so that way X is one in simple term in very basic term if you're thinking is it's integer it's a value or it's a Boolean that means it's not an artificial intelligence but if you think X is your natural language or it's an image that is showing that it's a Gen AI because you're interpreting that particular function as a virtualization or another aspect that is done through the artificial intelligence. I hope I made something. Moving forward, I also listed it out some of the AI ML models widely used widely known just for as a refreshing purpose refresher encoder decoder model is very popular whenever we want to learn about artificial intelligence as a product manager or machine learning aspect as a quant or digital these are the widely used words and widely used models where we get started transformer model sometimes this transformer model also get pre-trained with certain kind of logics which makes them pre-trans pre-trained transfer model but model is widely used with other and part of the encoding decoding to make more meaningful outcome for different purposes different business use cases and then the massive language like you might have experienced that a lot of time larger paragraph can be read and summarized it can recheck between the historical information which has been fed to the models large language model like processing of huge document into a summary or identifying certain pattern these models as it's becoming more and more popular and powerful responsible AI comes with the responsibility and it requires understanding of the possible issues limitations and unintended consequences any model which you develop or your company producing as a product should be have these ideas and these themes to be embedded where there is no bias and these models are accountable should provide the transparency and data privacy I won't spend much time here as I mentioned earlier I wanted to show some of the encoder decoder transformer birth or attention model you can see that the vector RNNs from the encoder side each stage they get ingested used for the decoder to take a reference of those vectors similarly when you want to train in the data set you use the feed of the data set for the giving input to the neural network and error correction for the self correction purpose so this is the reflection high level reflection of those models okay as I alluded towards into the context setting that what are the business drivers or which is causing higher amount of adoption of AI ML products in the fintech industry so you can see the investment decision to enhance efficiency AI and ML in fintech is gaining more and more popularity because of the endless data flow capabilities to handle those manually is tough and these models can help it out I want to bring a two-minute story here 15 years before as part of the quant team I was trying to build up a algorithmic trade booking system for investment banks and it was such a complex because these tools who are not there I have to capture the market feed less than a million dollar trade without having data to book it or to book those trades which required a client profile or a differential data on top of that multiple more than 12 or two dozen of the correlations like tier one tier two tier three of the data which can be layered on top of the client profile and the market feed which is continuously coming from the market data providers for the pricing per book so in simple language if you want to book IBM trade of X amount of dollar to provide a better pricing in less than a second or 40 milliseconds I was required to bring all this massive data run on correlation on top of them and then provide the best pricing so I can win that rate against all other investment bank on the street it took me eight month and 20 engineers to build this model and to deploy it which required continuous uptation of those correlation now you can imagine those all kind of things possible in less than three to four weeks of the AI and generating those model and deploying that model for your configuration that brings back to a big business driver and use in the algorithmic trading predictive analysis and trending aspect I also alluded towards the anomaly direction or focus towards the finding the fraud and EML and finding those anomalies these models can be big business driver and that's what we'll pick it up in the later slide to build a sample model and thinking through as a product manager how this can be incorporated and can be helpful the other use cases are here like payment processing trillions of data elimination of the false positive customer services financial fighting with the financial I'm trying to rush towards the next slide because this you may be interested in knowing that okay this is all theoretical part but how I can how I should build my model what are the thought process I should be doing what should be a close loop process which should be deployed to build this kind of model so as you have seen some of the surveys has reflected 46 percent of financial organizations reported fraud which is nothing new for us corruption and other economic crimes within the last 24 months and have seen 79 percent increase in fraud and by leveraging these model or trying to identify detecting the fraudulent activities like credit card fraud, identity theft money laundering algorithms to flag this kind of transaction identify the outliers that is the scope of this model when you are designing or developing that where do I want to start and dream eventually iteratively this model so initially you build a model to find the anomaly in the transaction then you want to find the outliers you want to have that flagging of those outlier patterns and eventually self-correcting those models and it has been historically found that reducing the false positive which will be higher in the beginning because you are starting with the set or limited set of the data or structured and both unstructured data but eventually and slowly rebuilding them through the feedback it will increase the reduction of the false positive by 60 percent and increase the detection of the fraud by 50 percent so reduce the false positive and increase your detection of the fraud so I try to use the SOM called self organization map model you have a spherical data which do the pre-processing in the card processing area then you train the data using different neural networks and vectors which I defined earlier and you train the model again and again and test the model that whether your sample size set and the transactions are mapping to each other or not once you feel comforted you deploy the model and you start doing the prediction but it doesn't stop there as I give the example of my algorithmic trading every month I need to refresh those correlation but in this case the new data's are coming on the left of the data new updates and those new updates are getting feed into the historical data plus your deployed model so it's a self-correcting self-announcing error and different kind of patterns into the machine learning model so imagine millions of transactions one or two kind of outliers but they are increasing eventually slowly slowly it's hard to detect for any card processing or card fraud or financial institution now the tool is working for itself for building this is the sample of how does the product manager should be thinking or should be aligning their thought in this case I picked up example of SOM or SOMDE also it's called it's called unsupervised unsupervised machine learning system based on unsupervised learning in a data driven way unlike supervised learning so SOM can be used for clustering the data without knowing the class membership of the input data therefore it can be used to detect for the inherent of the problems logistic regression is the most basic way yet powerful machine learning algorithm you can use to predict the true or false binary values it estimates discrete values and from a set of independent variables so I combined lot of technical terminologies but if you unpack them what I'm trying to say is that these are unsupervised model neural networks which will be correcting there so if you see on the left side of this graph there is a millions of projections has the same kind of pattern which is the normal data but you find two anomalies which are marked with the plus eventually these pluses increases and you mark the circles showing that these are a normality pattern to the self-packing feedback and then they become in the production for subsequent transactions where your frauds been detected and been flagged with that said I'm coming towards the end of my presentation I hope you found this presentation useful interactive and insightful as a product manager who are trying to leverage in day-to-day activities facing the current challenges which is in terms of growing need of the artificial intelligence growing need of the automation and thinking more how I can incorporate set subset of this technology into the day-to-day product management with that said thank you and appreciate your time