 And here we go. Hello and welcome, my name is Shannon Kemp and I'm the Chief Digital Manager of Dataiversity. We'd like to thank you for joining today's Dataiversity webinar, applying artificial intelligence in all the right places in the data value chain, sponsored today by Information Builders. Just a couple of points to get us started due to the large number of people that attend these sessions, you will be muted during the webinar. If you'd like to chat with us or with each other, we certainly encourage you to do so. Just click the chat icon in the bottom of your screen for that feature. For questions, we will be collecting them by the Q&A section in the bottom right hand corner of your screen. Or if you'd like to tweet, we encourage you to share highlights of questions via Twitter using hashtag dataversity. As always, we will send a follow-up email within two business days containing links to the slides, the recording of the session and additional information requested throughout the webinar. Now let me introduce to you the speakers of the webinar, Vincent Dini and Aditya Srinram. Vincent is the Senior Director of Worldwide Strategic Services at Information Builders. He leads the Global Strategic Services Team that supports analytic data integration and data governance. He brings almost 20 years of experience in the data integration and data management space, including data quality and master data management. Aditya is the Senior AI Strategist of Worldwide Strategic Services at Information Builders. He is a PhD candidate in the Faculty of Engineering at the University of Waterloo, where he is a member of KIMIA Lab, Laboratory of Knowledge and Inference and in Medical Image Analysis. He brings an extensive understanding of how artificial intelligence moves research and industries forward. Since 2011, his research activities encompass content-based retrieval of medical images using machine learning, deep learning and computer vision approaches. He has developed learning schemes and descriptors for medical imaging and published his works in top-tier journals and conferences. And with that, I will turn over to Vincent to get us started. Hello and welcome. Thank you very much. So, and thank you everybody for joining us today on this webinar. So, notice the title and notice the highlighting one of the certain words there, applying artificial intelligence in all of the right places. At Information Builders, we end up talking to a lot of customers, a lot of prospects, a lot of businesses that have many different types of use cases. And everyone's trying to find out new and more innovative ways to solve them. Artificial Intelligence Machine Learning is certainly one of those potential tools to be able to solve and expedite business challenges, be able to overcome them and get better or lie out of the data that you have as the information assets that you have. But something we also notice is that very often when talking to our clients, we identify that they may be looking at utilizing artificial intelligence not in the right place where another solution is going to be just as simple or just as not simple, just as easy to implement in terms of getting business value but will be much quicker and lower ROI. So, in today's session, we're going to be doing a few things and you can click to the next agenda tab. We're going to be looking over first an introduction to what artificial intelligence is. And this is something that if you're joining this, you probably have a well-rounded idea of what artificial intelligence is, but still for this conversation, it's a good idea to make sure everybody's on the same page and is coming from the same set of terms. Then what we're going to do is we're going to talk about some of the artificial intelligence journey that we've seen at our clients where they've been successful and where we've seen clients implementing these types of processes. With the journey, of course, comes all of the potholes along the journey. So, after we talk about what a standard journey looks like, what an ideal journey looks like, we'll review some of the common problems that we see over and over again as we help organizations to make better use of their information assets. And then finally, we're going to finish this out with a case study and then, of course, some questions and answering at the end of the topic. So, with that, I'm now going to hand it over to Aditya to go through the introduction to artificial intelligence. Take it away, Aditya. Perfect. Thank you so much, Vince. And thank you for the introduction. Thank you, Shannon, for the introduction. Welcome, everyone, and good afternoon to this session on artificial intelligence, applying it to the right places. This session is going to be essentially, to put it at a very high level, it's going to be a primer before getting started with your AI journey. So, as Vince had pointed out earlier, the agenda is pretty straightforward. We want you to enable you to start thinking about AI in a more strategic mindset and start taking proactive measures on what are the good practices, what are the best practices in order to implement AI in your organization, what are the few things to look forward to, what are the few faux pas that organizations commonly have, and outlining all of that so that you have a good foundational understanding of AI before thinking of ways to applying that within your organization, within the organization, or within another vendor's organization. So, an introduction on AI. So, let's get right into it, right? I'd like to start this section with a topic that I'd like to talk about for hours. It's to debunk common myths that you may have heard of within your organization, or perhaps you may have read up on the internet. So, to give you all better visibility on what these myths and realities are, let's get a foundational understanding of this, right? So, given all the traction around AI, of course, is raising a lot of questions. A common one I get is a belief that the main beneficiaries of artificial intelligence is technical organization, technology organization. And this is simply not true. In fact, what I have seen a surge of both use cases and AI adoption across different industry verticals, this includes financial services, which is encompassed of both credit unions as well as banking, but healthcare, agriculture, government, public sector, transportation, logistics, just to name a few. So, certainly artificial intelligence can branch across various verticals. It's not purely for a tech organization. Another misconception is, in fact, this is more of a question, which is, is the impact of AI leading to workforce reduction? The truth is that in some cases, the jobs are likely gonna change as some of them are gonna be more automated, right? It's gonna be more automated and it's gonna create, it's gonna lead to creating new jobs. But the intent behind AI is not to replace but to complement a workforce, right? So, AI is a very suggestive approach. You gotta decide in a way that's very suggestive. It's to complement a persona hence attaching a subject matter expert at the center of equation. It's very important. Going back to my background in the medical domain, the intention there is to provide suggestions on how to best go down the path of diagnosis, for instance, right? It's not to replace a subject matter expert. It's to complement them and provide them the appropriate knowledge and background that they may have perhaps overlooked, right? The third is AI is a magic box, right? That can make sense of any and all of our messy data. I'll say that this is one of the things that you hope for the best but the truth is AI can't solve everything. The ideology behind AI is that it has to learn something and generalize its behavior in the format of a data force. If you provide it corrupt data then it's not gonna learn anything, right? So to provide it good, concise, harmonized, cleansed data is very good. It's considered a prerequisite to going down the path of an AI initiative. And finally, and this is a heavy one for many organizations, which is would I need a scientist or AI experts to create a tangible predictive or AI application? Frankly, this was true a few years ago but in a very short period of time, things have changed. There are a lot more strategies you can keep in mind. A is there is a out of the box model experience where it's a pre-trained model that you can just adopt given a targeted use case and you can just run your data points against this pre-built AI model and just get your predictive results right off the bat. The second piece of that is there are self-service tools that exists out there that enable data analysts to create and deploy AI models without even knowing or understanding code, right? So all of these different facets are available and it really maps into what the organization's needs are and what kind of personas exist within the organization. The reality is that AI is everywhere. It's across industry verticals and multiple lines of businesses. It can be embedded technology as well. And from a user perspective, it often is. These are smart embedded softwares that is becoming part of our lives in the context that AI is working in the background and enhancing our day-to-day lives without us even realizing it, right? So think about optimized search engines. Think about home automation, personal assistance like Siri, Alexa. These are embedded technologies that are out there and that are fueling our lives on a day-to-day basis. And finally, AI is great for all types of problems. If we focus on the healthcare vertical, we can really derive really great examples and use cases, right? From optimizing bed count to image diagnosis to performing remote surgery with high precision. We're talking about a really wide gamut of possibilities when using AI. So before we apply a definition around AI, just sort of set the stage of the importance and impact of AI. So fundamentally, I believe that artificial intelligence, it's going to be in every piece of software that we write in almost exactly the same way as databases have historically. Now databases are great at storing things and counting things and running reports. AI on the other hand, it's very good at recognizing patterns, extracting features. It's very good at understanding languages, for example. It's very good at making predictions. And because these different use cases, these different positioning for AI is becoming so valuable, especially from a operational perspective, you're going to see a flavor of AI inside many of the applications we see commonly just like databases has historically, right? And this brings me to the quote that fits in well with the positioning, which is coined by Andrew Ng, who is one of the founders of deep learning. And he said that artificial intelligence is the new electricity. And this immediately gets into the core of why AI is so important. And I like to show this quote to sort of set the stage on what is to be expected in this session, right? Why is AI important? And why is it, you know, why is the hype? So why is the hype escalating and how can you apply that into different organizations? Is it really tangible to apply it across different verticals, right? That's what we want to answer today. Now, AI historically has been perceived as an act of fiction, but over the years, there are certainly a surge of overall adoption of AI across different verticals. And this graph speaks volumes to support that, right? It outlines a flavor of the addressable market of AI adoption in US dollars. And so we can take a look at the exponential, the exponential growth moving forward. And the reason for that is not because we have, you know, quicker access to data scientists or even because we have, you know, a better hold of mapping the algorithms to a use case across verticals. But the catalyst, the reason for the exponential growth, the reason for this projection is that businesses can now understand the operational side of AI, which is to enable AI in a way that can help drive business insights into actionable outcomes. And this fits in well with my next slide, which is why artificial intelligence now? So why all this hyperbole? Why is AI gaining such traction? The reason is surprisingly digestible. The first is due to practicality. This is, so when I say practicality, we now have tangible use cases across verticals that embed AI into business processes to provide daily insights, whether it is from a government sector, from public sector, whether it's healthcare, whether it's retail, logistics, transportation. All these verticals are adding a flavor of AI to differentiate themselves. The second is because we have, now we have faster and cheaper computing resources. With everything going on the cloud, it's much easier to configure your environment to create and deploy AI models by attaching it to higher configurations, like higher RAMs, higher spec, high specifications of GPUs and TPUs. All of these different resources are now available and it's much easier to grab on to it and start training your model or even deploying it for operational purposes. The third is the evolution of data. Every organization I come across is collecting enormous amounts of data and setting strategies in place to monetize their investment in data. But what does more data mean? More data doesn't just mean, doesn't just let us see more of the same type of data. More data allows us to see new types of data. It allows us to see better types of data. It allows us to see different patterns of data, right? But I have to emphasize that although organizations may have more data, it's more important to have good quality data, right? For any AI implementation, one has to make sure that the data is clean prior to going down the path of strategizing an AI initiative. Finally, the fourth is the evolution of algorithms. And this is certainly more on the technical side, but at a high level, there has been certainly groundbreaking advancements in research and developments of advanced algorithms that has immensely helped AI become more powerful and efficient. It's not just algorithms that are yielding higher and higher accuracy across a certain vertical. It's also algorithms that have been optimized to run much quicker, that minimizes runtime, that minimizes training time, and that enables scalability and repeatability, right? So all of these different strategies are coming together in the AI cycle that's complimenting the evolution of algorithms. Now, to keep true to our core message around AI and data governance, I will find a few common use cases that one can implement across industry verticals. Of course, when you're talking about data governance, irrespective of the industry vertical you are, as long as you collect data, there's a data governance play with it, right? So how does data governance overlap with artificial intelligence? In short, data governance helps an organization to better manage their data, whether it's availability, usability, integration, or even security off the data, right? By using the right technologies, data governance can drive business values and support an AI transformation. Technically speaking, it includes, but it's not limited to the use case that I've outlined, which is anomaly detection, right? To detect what the outliers are in your data set. Now, some organizations use these outliers to capture seasonality, use these outliers to essentially drive some sort of insights. Others use outliers to essentially scrap that data, right? They think that the outliers are not valid and it's considered noise to the data set, but detecting these outliers can certainly, AI can certainly help propel you to get to that end state. The second is metadata classification, which is an automated way of bringing about structure in data. So think about multiple sources of data. You want to classify what a particular data value means, is it an address, is it a name, is it a phone number, is it a, is it a SIP code, is it a SIN number, right? To classify that and bringing about structure to unstructured data points, that's a great application of AI as well. Data integrity is another big one. The ability to match and merge multiple sources of data using artificial intelligence, this includes any movement of data that has any kind of change in shape or form, right? We also have augmented data profiling, that is to make data, to make sure the data is accurate across systems and it's usable. This would mean to profile every record and feature, including how many values are unique or missing with an attached statistical significance to it. This will allow to get visibility into the distribution of data for each of the features that you hold, right? To get an understanding of where to apply transformation or how to best normalize your data. Finally, I have something to share that's on the screen, which is data completeness, right? How do you complete your data? How do you, do you include the data? What's the best strategy for that, right? AI can, AI is a very elastic approach. It will enable you to provide suggestive reasoning on how to fill in gaps in your data as well, right? So it really depends on how you want to position AI in your organization, but certainly apart from business application, data governance is a huge lead forward for artificial intelligence. How do you best use your data in a more trusted manner, right? That's when AI really comes into play to provide just suggestive reasonings. You know, Aditya, it's sort of interesting how when you think about artificial intelligence with data governance, it's sort of the artificial intelligence and data governance having a yin and yang sort of symbiotic approach, right? Or a relationship because as you just described, AI is going to help to support a data governance initiative by doing those things, getting the data to be in a better state, which then of course will drive or enable better outcomes with artificial intelligence and machine learning, utilizing that data further downstream. And then of course that information helps to foot to support the data governance initiative. So once you start this process of AI in both places, sort of you build momentum all along the way, don't you? Absolutely. And I want to add another point to that, Vincent. And I'm glad you kind of bought that up. The traditional data governance approach is a very static approach, right? You know, you have defined set of rules, right? And it's not elastic. Whereas the differentiation when you have AI coupled with data governance is that you have an elastic approach. And the advantage to elastic approach is that first you get to learn your data, which means that you can adapt as your data matures. And this becomes scalable and repeatable, right? As opposed to a traditional data governance where it's a static approach. If you have newer types of data, then you have to essentially fill in your data dictionary to abide by those new rules, right? This is more of an elastic approach. That's a great point to bring up, Vincent. So now that we sort of set the stage on outlining the capabilities of artificial intelligence, let's get a definition out there, right? So what is AI? Well, simply put, it's an approach that analyzes current and historic data to make predictions about the future, right? And it does so by understanding and uncovering patterns in data that is otherwise difficult for us to comprehend. Think about the terabytes of data. No one is gonna go there and intervene the data manually, right? You need some smart solution in order to do that process for you, right? And if I look at this from a business perspective, the value-added component of AI is to really step away from making decisions based on personal experiences or quote unquote, gut feel and transform that into a data-driven decision based on the current data and your historical trend, right? Whether it is a capture seasonality or whether it's for capturing X number of years that justifies the use case that you're driving. Now let's understand the difference between the two facets of AI that is machine learning and deep learning. These are by far the most common techniques around AI. And this is a very common question I get out there, right? What is breaking down the nomenclature of AI, right? What is data science? What's the difference between data science and artificial intelligence? Where does machine learning come in? What is deep learning? And why is that, you know, why is the trajectory of deep learning going higher? What's the advantage of that? I'm gonna cover all that in this presentation as well. So, so AI was invented, you know, back in the 1900s, just a bit of history lesson over here, back in the 1900s by theoretical mathematicians, philosophers and even psychologists, right? In this diagram, you'll see that deep learning is a modern subset of machine learning which itself is a branch of artificial intelligence, all of which in its entirety is under the discipline of data science. So the question I commonly get asked is, is to outline the difference between data science and artificial intelligence. And the answer to that is quite simple. Data science is the science behind data. And it has multiple facets. A part of that is artificial intelligence. So in reality, data science is an overlap between traditional software, research and artificial intelligence. So what we know what AI today is nothing new. It's a fabrication of what we invented back in the 1950s. So at a macro level, the general idea of AI is that instead of instructing a computer on what to do, we're gonna simply throw data, we're gonna simply throw data at the problem and tell the computer to figure it out by itself, okay? So artificial intelligence, machine learning, and deep learning follow the same core principle. The only thing that sets them apart is their architecture. And the way, you know, and the way they're computed and the way they design, I'm gonna give you a glimpse of that as well. Now, regardless of which approach you take, at the end of the day, it's all about one thing. From a business perspective, it's all about one thing, right? That is to unlock that information and really take advantage of it and really monetize your investment in data, right? You collect the data so far, you wanna monetize and invest and you wanna gain insights. And that's really the purpose of artificial intelligence. So what is the difference between machine learning and deep learning? Before we dive into that, let me talk about the commonalities between AI and between machine learning and deep learning. They both follow the same principle, which can be broken down into four steps, right? Data pre-processing, feature extraction, model learning and model evaluation. The difference lies in the way they process it. So what are the two main differences between these learning techniques? The first is feature engineering, which is the act of extracting features from the data, right? Machine learning requires manual intervention of extraction. That means that AI developer needs to spend their time on the data and derive important features. Deep learning, on the other hand, has automatic feature extraction and that's why it's getting more traction because the ideology behind deep learning is, hey, give me the data, I'll figure out the rest of it, right? So the intent behind deep learning is, give me the data in its rawest form and I'll automatically extract the features for you. I want to, but I want to be very careful over here. When I say raw format of data, it has to still be good quality data, right? We can't throw corrupt data because it's not going to extract any feature and leaves it's not going to be valuable features. So what does it mean to extract data? What are features, right? Well, if you look at an image to the right, a feature are important data points. In this case, given the image of a cat, the first feature is a location of the space. The second feature is a distance between the eyes. The third feature is the shape of the ear, et cetera. Those are important features and those are features that deep learning algorithms can extract automatically as opposed to a manual intervention for machine learning approaches. The second piece that's a difference between these two learning techniques is a scalability. That is, if I create a model right now, what happens when I get new data, right? How can I attend new data points to my model? Well, here's another downside to machine learning. They need to be rebuilt once you get new data points, right? So if you train a model right now and you're using a machine learning approach, when you grab new data points, say, end of FY20, those new delta of data points, you need to use that with the initial data points you trained against to relearn the entire machine learning algorithm. Whereas the deep learning approach, that can be done automatically, right? All deep learning requires is the delta of new data points and you can just take those delta points and adjust it back into the deep learning model and it will automatically adjust itself as opposed to relearn the entire use case from scratch. At the end of the day, the intuition behind either of these approaches is the same, right? The core remains the same, which is to extract features and see where it fits in the data. The process and the design behind it is different and that's why deep learning is, A, it requires a lot more data as opposed to machine learning and, B, organizations that are collected a lot more data are leaning towards deep learning because it's a more automated. So onto the data science process, and I noticed that, kind of noticed that the data science process, I mentioned data science as opposed to artificial intelligence from the title as this was an end-to-end approach. So how does the data science, life cycle look like? It starts with identifying a use case, right? Identify the use case. Thereafter comes the organization of data, the harmonizing of data, the integration of multiple data sources that can be leveraged or attached to the aforementioned or the underlining use case. Once the data is collected, thereafter comes the data quality piece which is perhaps the most critical step to ensure the AI use case is tangible. Once the health of data is deemed good, then comes the AI development phase, which includes a pre-processing sampling of data, training of data, validating and evaluating the capability of that model, right? Once the model is evaluated, the final piece is to integrate the AI model with a business intelligence environment and create a quasi-real-time or real-time visualization against the predictive results. Now, this of course alludes to prescriptive analytics. To put it very simple, this is the entire data science process. It starts off the data integrity, right? Data quality, harmonizing your data, then comes the AI piece where you have a use case attached to it, you know which data points are important and you know what outcomes you want to drive and then you take the outcome of the AI model and you embed that into a business pipeline, a BI pipeline that will allow you to have more actionable insights based on your predictive results. And this is a recommendation that I would strongly advise for each and every AI use case that you have identified. To the respective of use case, this is a process that this is the best practices to follow. So how does AI enhance data governance? The first step is to organize the data. So if I get a little more granular on how the overlap of AI and data governance comes into play, it starts off with organizing your data, right? So data organization starts with aggregation of data, bringing the data together from internal external third party sources and also correlating the master data with transactional data, right? Then comes the layer of intelligence wherein you can do an automated match and merge, you know, use case, refinding or even reconciling your data to find and build relationships across these data fields. This includes, you know, relationships between people profile, product profile, store profile, location profile, right? And of course, an embedded analytics approach like this will allow for future, you know, further analysis to score your data to understand the data quality and to rank your profile. So really the AI piece over here is a layer on top of the organization of your data and the solidation of your data to bring about intelligence to it, right? How do you relate different data fields together? Is it valid? Are these data profiles valid? What's the anomaly here, right? All of these different use cases can be derived from a data governance perspective, right? From a data governance AI perspective. So after the intelligence piece comes the insights piece wherein the intent is to drive actionable outcomes from an AI cycle. This includes gaining insights and recommendations that is available to business users in the form of data-driven application whether it is, you know, through a consumer 360 profile or a supplier profile like a supplier 360. The idea here is to make all this data available to a data-driven application such that you have a sense of confidence that trusted data is going downstream to other operational analytics like CRN systems, right? And this will enable trusted data to all the business users to consume that data as for their business goals, as for their requirements in the context of their data data, right? So that's a bit of background about AI and how we can couple with the traditional data science approach and how we can couple with the data governance and AI scheme, right? But now I want to spend the majority of my time to outline the journey required to attain a successful AI initiative. As mentioned before, you know, this journey is broken down into two facets. First, from a business perspective to increase overall adoption, what's the right business strategy to ensure AI can fit into my organization? That's the first half. The second half is more of a technical journey. So what is important from a technical perspective that serves as a prerequisite to having a successful AI implementation? So the first is to define a strategic goal, right? And that strategic goal should map to your organization's initiatives, right? A common use case, sorry, a common question one should ask themselves when strategizing AI, at least in the beginning stages is our questions like, you know, what are specific problems that I want to solve or what opportunities do I want to take, right? The purpose here is to really define a use case that can have a direct impact to your organization. And I would recommend identifying between three to five different use cases, AI use cases. And once these use cases are attained, it's critical to prioritize these use cases into a transformational roadmap that covers both a long-term vision as well as a short-term win, right? So understanding what the strategic goals are and what the organization's vision is, couple that together and that would be the first step in identifying a tangible AI use case. The second is understanding the data. So once the use case is prioritized, then comes the time to tap into your data sources and define which data fields and attributes maps into the aforementioned use case. This exercise is to essentially ensure that the available data is, you know, it's addressable to the business objective. It's very common in this case to have a technical persona as a partner to drive the data preparedness phase. Hence, a thorough collection and exploration of data is required, right? So to get a good understanding of what data resizing your organization, what have you collected historically, all of that is going to be extremely important and it should map to the aforementioned use case. So some companies over here, you know, have a pressing need for getting an AI initiative going and they don't want to spend much of the time or the data governance space, right? They don't want to spend too much of the time to cleanse their data. They're targeting an AI initiative. So for such organizations, I say seek expert solutions for data management, right? Because you need to have data management or data governance strategy to enable the data to be of good consulting structure before going down the path of AI. The third is elect a use case owner. So it's important to elect a use case owner to be responsible for driving the AI use case from a business perspective. This use case owner should define which KPIs are, you know, are most important to the business in support of that use case. These KPIs should measure and should be a guidance to your organization to kind of give them, you know, a promised land of how that project is gonna yield a successful deliverable, right? These KPIs are gonna serve as your backbone to ensure a successful project completion. It's also beneficial to take a second look at these KPIs, you know, after an appropriate duration of time and you're going down the path of maturing your project, take a look at the KPIs, go back, reflect on it to ensure that, you know, these KPIs are valid and if so, that's good. If not, then you've gotta address the discontinuity over there, right? It's very important to be aligned with the initial KPIs that are outlined. Now, I've been on opportunities where businesses are unable to identify the right KPIs to measure success. And if that's true, and if your organization is struggling with that, and I would say that the project is too complex, I would recommend taking it back to the growing board, understanding what the core objectives are, and then addressing mapping the KPIs into the AI project and defining that, you know, preliminary before even going down the path of the initiative. Understanding success metrics. So it's important to understand success metrics and possible outcomes of AI project once the strategy and the use case is defined and derived. From a business context, a use case owner that you've elected should define what the success metrics should look like, right, for the AI project. And how will, and it's very important to outline how to measure that success, right? It's essential to be specific and to identify which business metrics and KPIs to define to track the AI project as a mid-source. It's also important to ensure that the success metrics is compliant with the technical team. You don't want a side that approach where the business leadership team is defined the KPIs that siloed from the technical team, the engineering team, the product team. You want to have a more cohesive look at the AI project, right, so whatever you define as a KPIs and success metrics needs to ensure compliance with the technical team. The measure of success will serve as a foundation to the technical team to design and build the AI model. They're going to take those KPIs and they're going to build on that, right? The fifth is technical and legal, sorry, ethical and legal issues. This is huge, it's absolutely huge. It's very underrated. The implications are absolutely huge for ethical and legal issues around AI. From a legal perspective, ensure consent and data privacy, especially GDPR, one of the central questions one has to answer when implementing AI is the evolution of law, right? Whether to imposition of new laws and regulations or through modifications of existing rules to comply with technology advancement, right? Overall, the last few years, to the best of my memory, the last few years, AI is changing the data privacy, right? And a lot more legislation. So it's important to stay on top of that. In terms of privacy, the volume, exposure and residency of the data is perhaps top considerations, right? AI systems use vast amounts of data. Therefore, the higher the volume of data, the more questions are going to be raised and more concerns that are going to be raised. So keep a lookout for that. Common questions one faces when regulating AI pipeline is who owns the data, right? Who owns the data that is shared between the AI developer and the user and the data be transferred? Is the data on premise or on the cloud? If so, what is the appropriate legal direction to take if you're going to migrate it on the cloud? Should the shared data be anonymized to protect privacy concerns, right? Frankly, depending on the use case, the ethical and the legal compliance could be a bottleneck. So it's always recommended to run a parallel stream with a legal team as the data is being flushed out. Technology and infrastructure. As AI moves from experimentation to productization and adoption in your organization, it will certainly require technology and infrastructure changes, right? So depending on the model of choice, the volume of data, your organization may require to scale up the computing resources or even scale down depending on the model and the application. And you have to consider the infrastructure cost and you have to consider the resourcing cost, where the computational resources cost. So as AI becomes more complex and you start growing and maturing your AI model, it's going to be resource demanding, right? What you have right now may be a subset of what it could be a year from now. So organizations need to take a proactive measure to find cost-effective environments, right? But once you're set up, then the coupling of AI with infrastructure could serve as a competitive advantage. Skill and capacity. So it's important to consider any skills gap that may prevent you from achieving your AI goal and have a strategy in place to kind of close that gap, right? A common option for AI development is to attach a technical persona, like a data scientist or machine learning engineer to create and optimize a model during against the data. But the truth is, you don't need these personas in-house, right? So if you're not looking for, if you're not looking to hire data scientists and you're not looking to fund contractors, then it's fine. What we're starting to see a lot more in the field is that there's a move towards democratization of AI, a surge of self-service tools, enabling data analysts to create and deploy AI models, right? So the option is, it's widely popular these days. Another option is pre-booked models, like I mentioned earlier. There are vendors out there that provide you pre-booked models for a targeted use case that you can just piggyback from, run your data forms and get the predictive results, right? And these serve to be very robust ways of getting AI initiative out there. Finally, change management. It's really important to manage the change carefully. And particularly if you're automating or streamlining process, this may have an impact on the work that your employees do. So hence, it's imperative to have a plan on how the organization manages to change, right? While promoting a positive culture of AI in the business. And let me quickly pause here. Anything to add here before I move on on the organizational process? I think you nailed it. Nope, so nothing, thank you. Perfect, so in terms of technology process, I'm gonna skim right through this. There are three pieces over here, right? Data preparation, the algorithms of choice, the model of choice, infrastructure of choice and finding the visualization piece, right? Which eludes to prescriptive analytics, right? That's a summarizer, right? Why is data preparation so important? Well, the net of it is that a recent study that was done by a cognizant, Cognizant Lidica. He suggested, I read this article a few days ago. He suggested that data scientists spend a lot of their time labeling, cleansing, and augmenting the data, right? In fact, about 80% of the data scientist time is spent on preparing the data. While this is a very good sign, considering that good data goes into building the analytical model, the reality is that data scientists should ideally be spending more of their time interacting with the data, right? Which includes training and evaluating that model. So data governance is very important. Data preparation is very important. If you don't have a good data preparation strategy, if you don't have the appropriate personas or the technology, then, and you regard as you go down the path of an AI implementation, that the concept of garbage in, garbage out qualifies here, right? Which is you provide garbage data to an AI algorithm. It's going to give you garbage results. So you want to stay away from that. You don't want to go down the path of maturing in an AI lifecycle, knowing the data is not trusted, right? You want to ensure good data before going down that path. Most of the, you know, yeah, go for it. Ben, anything to add to that? Yeah, I was just going to say, you know, again, this gets back to, and you've been touching on this over and over again, the importance of a data governance program overall. But, you know, the cleansing is one thing. I just wanted to highlight the fact that a lot of times, you know, when we're thinking about cleansing, cleansing is a larger thing just is, do I have the right data, right? Is it correct? Is it complete? Is it the, is it accurate, all of those aspects? But then also, is it easily identifiable and relatable to the rest of your information assets? And I noticed that one of the other statistics if you have there, just so much time is spent cleansing the data as augmenting the data, right? And so, and those two exercises are very much interrelated. So this is another place where having a very clean, a very organized data governance program will help the data scientists because they may want to augment that data using information that other, if they didn't have a program they would not know about, or it would not be time-efficient to actually try to join it too. Agreed, agreed, absolutely. Absolutely, and both said on that. So on the technology side of things, as Vincent mentioned, data quality, augmentation of data, extremely important as a prerequisite to AI, right? So what happens once you have good quality trusted data, right? So then comes really the AI piece where you want to map your use case to the algorithm, right? To train your model. I don't want to get into the weeds of what these things, how to get to that, but I want to give you a very high level overview of what are the learning techniques that are out there so that you're aware of it and you're prepared for having conversations like that, right? So the three common learning schemes are learning algorithms that are out there, supervised, unsupervised and the enforcement is. Now, as we speak, it's just maturing, but this is sort of the fundamentals. Now, to kind of summarize what these mean, we'll take a look at supervised learning. It's like learning with a teacher, right? Where there's labeled data and it provides you guidance on how to reach your end state. That's supervised learning. Unsupervised learning is a little bit different. It's like learning without a teacher, right? So you don't have structure in your data and the motive here is to use artificial intelligence to bring structure by recognizing patterns, right? And finally, reinforcement learning, which is relatively a newer learning scheme, but we're seeing a really high adoption for reinforcement learning. And this particular learning scheme follows the principle of punishment and reward periods, right? Which is very common in large organizations, right? So think of advertising and recommending an express product like vendors like Amazon and Facebook, right? Where they use reinforcement learning to essentially, hey, you bought this product now, I'm going to recommend these products to you to entice you to buy these, right? So how do you get those recommendations? Those are done by reinforcement learning. Understands your behavior and tries to recommend what's next best for you, right? In order to purchase. So that's a high level on the learning side. So finally, I want to touch on the visualization to kind of wrap up the AI piece, right? I look at visualization as a key driving force for artificial intelligence engagement, as you know, I like to put it as language of the eye, right? And the results are easily attainable. This is what business leaders look at, right? They don't care how the algorithm's been developed. They want to see what the outcome looks like in terms of a visualization, in terms of report. The business intelligence market in the last 25, 30 years have been a very reactive way of looking at your data, right? So you have a large set of data, you query at large scale and you provide some sort of visualization against it in a reactive manner, right? And so you're doing data visualization on data points that have happened historically. Now, when you couple AI with business intelligence, then you start to get a more proactive visualization, right? That is very powerful to your organization, you know? So to attach a use case, you know, how do I understand cash flow to ship or to hold a product? That's a great use case or how, you know, what's the right time to increase or lower the cost of a product or even predict term, right? For X number of employees or distributors, right? All of these are AI use cases that you can couple with business intelligence to show what the end state could look like as actionable insights. All these use cases are achievable when you couple business intelligence and artificial intelligence. Put a number out there, you know, to put a number out there, why is the coupling important? The rationale is quite simple. It comes down to expanding the ROI, right? When you look at market research and market demand, we found that the research, the current investment when BI is coupled with BI is significantly higher than a pure BI application, right? The BI, median ROI is roughly about 89%. And with the addition of AI with a coupling of AI, the median ROI increases by over 50%. What does this mean? It means when you combine AI and business intelligence, you create data stories that are more reflective of an organization's business process, right? Now, in terms of common pitfalls, and I just want to kind of skim right through this because with respect to time, I know we have eight minutes left, just at a high level, what are the common pitfalls for AI, right? I want to mention two or three ways here that are very, very common. The first is creating to focus on a business initiative. So it's very important to understand what the use case is and have a strategy in place to collect the data and map that into the aforementioned use case. Once you have defined it, you shouldn't deviate from that use case. You shouldn't reprioritize it, right? You should stick to that use case, identify the KPIs and drive that to an end state and have the technical and the business personas attached to it. Another common challenge for organization is not having enough data, right? Now, it's false that you need a large amount of data to do AI, right? Going back to the ML versus deep learning, if you're going down the path of deep learning approaches, yes, you need a large amount of data, but if you're going down the path of a machine learning approach, I have done machine learning approaches with very less data. In fact, I've done machine learning approaches that would plug in data sets, right? That's generally a lot lower than that of multiple sources attached together, right? So it really depends on where you are and what kind of data is collected historically. But what's more important is to have good quality data, right? More than having the volume of data, have a good quality set of data that's going to serve as a few references. Another great one is investing heavily on analytical tools with little to no return on investment, right? There are several common mistakes made when it comes to investing in AI tools. Companies often buy expensive analytics software that is way too sophisticated for their needs, right? These solutions are, you know, they come with a very high price tag, but they also come with very high adoption, right? Even the same study. It makes it very, very difficult to enable your personas to meet that technical expectation that the product requires, right? So keep it simple. And perhaps the most important, how a platform approach, this is probably the best advice you can give, never have your AI initiative as a siloed approach, right? Always have it coupled with a business intelligence platform or a data management platform. Couple it with something to drive actionable insights, right? The last thing I wanna mention over here is failing to operationalize. So predictive analytics, as I mentioned before, couple it with the business application, business intelligence application, the embedding of BI applications with AI enables you to have a more proactive view of your business, right? So how do you avoid these different pitfalls that I've mentioned in the previous slide? Frightening ROI, when tightening a predictive application, you need to consider the total cost of ownership and anticipate a return, right? Ensure maximum value is achieved. Make sure that's read it out in the beginning stages. Focus on the bottom line initiative. So create models that maps to the organization's strategic goal. Prepare the data, right? Guarantee the most accurate and trusted data as possible before going down the path of implementing AI, right? And evaluate the model without over-evaluating. That's very important. So the model must be tested to ensure that it provides better decisions. Making capabilities over a current, you know, you have to compare it against your current analytics method and see if you can get a better result. So for example, if the current AI initiative is 70% accurate and the new AI initiative is a 65% accurate for some reason, but you can see that the 65% accuracy is a lot quicker to train to run. Then it's a cost analysis that the organization has to do. Are you willing to forego 5% to increase the, you know, to minimize the computational resource required to train the model, right? Finally deploying the results. So providing that AI results to a business application for proactive strategies, right? In terms of case study, I just wanted to get a quick case study out there. You know, this is a client that I was focusing on. Clients, it's the Perry Foods. And I want to focus on the use case that I had fleshed out a few months ago. And this is one of many use cases that I've touched upon recently, right? And hopefully this will serve as an inspiration for you to start thinking about how AI can be implemented inside your organization. So, the Perry is a food distribution company serving roughly 13 states in the US with north of 9,000 locations. And they have been operational since 1963 with roughly about a little south of 1,000 employees. So we had initiative with the Perry around the realm of B2B churn prediction. So they had roughly about 20, 25 gigs of data across three different data sources. And they wanted to predict which of the distributors are likely to lapse such that they can proactively have retention strategies in place. So this was roughly a three to four week engagement where we spend about a week understanding and pooling the data sources and other week for gap analysis. So making sure it's trusted, filling in the gaps when necessary. But majority of the time was spent understanding and coming up with a data driven position and what constitutes this churn, right? So how many months of inactivity constitutes as churn, right? Is it four months, is it five months? Is it can be captured seasonality because a lot of the distributors only buy certain periods of time and they don't buy for the rest of the year. So we don't wanna label them as churn when we know historically they only have a seasonal purchase. So there's a statistical analysis you have to go there. So what we did is we created a recently plot which captures the behavior of the distributors with respect to time, right? Which captures the seasonality as well. And once we determine the threshold, I mean a timeline for churn, then it was a task of learning, packaging a model and evaluating its behavior, right? And to put a dollar value into the account we projected a return on investment of 400,000 preventing roughly four million in churn year over year, right? And so this was just the first phase where we wanted to design an algorithm that can predict which distributors are likely to churn. But moving forward, we can expand the learning model to cover more granular use cases. Think of it like an AI use case V2, which outlines what will the distributors purchase next or what else are the distributors interested in purchasing in terms of next as product? What are the percentages allocated to each of these products? So at a high level, this was a high level AI story that I wanted to share with the audience. And a quick takeaway on this piece is that I always recommend starting off simple. If you take this use case as an inspiration, we started off with a baseline use case and you wanna build upon that, right? And that's the way to kind of ensure your model, your AI model matures with your organization's maturity. I wanna make a note that we have three minutes left and we do have questions coming in. Perfect, so this fits in well because this is my last slide. And for the sake of completion, I just wanted to add these common use cases out there for different protocols. So this is for you to essentially take and kind of brainstorm on where the AI can really fit into your organization. So I don't have anything to narrate over here. This is kind of the ending slide for me. I've added additional reads if necessary. These are great reads to learn about artificial intelligence and a few interactive sessions as well that I've cited. So with that, let me quickly pause, Shannon. Let me pass it back to you, and then for mediating the questions. Perfect, I love it. And this has been such a great presentation just to answer the most commonly asked questions very quickly. We'll be sending a follow-up email by end of day Monday with links to the slides and links to the recording of this session. So just diving in here and tipping into the questions, how do you handle data spikes like the pandemic in regards to future predictions? I'm assuming the quote unquote new history might dramatically affect the predictions. Yeah, that's a great question. So that goes into the realm of the seasonality that I mentioned before, right? If you take a look at the case study, the La Perry case study, it was exactly the same situation where you have a certain period of time that you could have just explained until you've experienced it, right? And if you have gaps in the data like that, that's the beauty of AI, that you can use statistical ways to grab onto these seasonalities, right? So hopefully, COVID doesn't happen in the long run. Maybe it's just above all that's happening right now. But once you overcome this and you see light at the end of the tunnel, the period of time that we're in COVID where businesses are struggling and we have essentially skewed data points, that's going to be weighted low depending on the algorithms of choice. So the point being is the AI model is flexible enough to weigh which data points are important. And that kind of couples in well with my dialogue on the seasonality, right? That fits in really well with that definition. I love it. Well, that does unfortunately bring us right to the top of the hour. That is all the time we have for today but I'll get you guys the questions so we can get answers to the remaining questions. Just want to thank everybody. And again, remind you all to send a follow-up email by the end of Monday with links to the slides and the recording as well as their information. I think the information bill does for sponsoring and this fantastic presentation. I really appreciate it, guys. Hope you all have a great day and stay safe out there. Perfect. Thank you so much, Shannon. Thank you. Bye now. Take care, everyone. Thank you.