Video: Transform Data into Strategic Advantage with AI | Summary: Strategic data handling reduces the "price of not knowing" and accelerates AI advancements.
Video: Elevate Your Career with AI and Intelligent Automation | Summary: AI's rapid evolution demands daily learning; it enhances jobs, not replaces them, with IA's support.
Video: The Evolution of AI: Navigating the Data Dilemma | Summary: text discusses the evolution of data analysis, highlighting AI and its dependence on data quality.
Video: The Digitization of Anything: AI's Numerical Transformation | Summary: Everything can be numerically expressed, enabling AI to digitize and analyze all data types.
Video: Optimizing Data Governance: AI, Compliance, and Cost Efficiency | Summary: Prioritize data sovereignty, optimize AI, and prevent leaks for robust regulatory compliance and efficiency.
Video: The Essential Role of Information Architecture in AI Success | Summary: Seasoned data expert shares insights on information architecture as the foundation for AI.
Video: Cybersecurity as a Value Creator in AI Strategy | Summary: AI strategy requires auditable data lineage, addressing security as a value creator, not a cost.
Video: Legal Responsibility and the Unpredictability of AI Models | Summary: AI decisions are unpredictable; ensuring alignment with business goals and understanding data sources is crucial.
Video: Embrace New Collar Jobs: Navigating AI's Impact on the Workforce | Summary: Exploring new collar jobs and AI's transformative role in the evolving workforce landscape.
Video: Reimagining Business Processes with AI at the Core | Summary: Redefine business processes with AI at their core to maximize benefits and scalability.
Video: Bridging the Gap: From Data Collection to Understanding | Summary: Data collection excels, but understanding lags; being data rich often means information poor.
Video: The Price of Not Knowing: Navigating Data in Las Vegas | Summary: Exploring Las Vegas, I ponder data collection versus understanding, emphasizing we're data-rich but insight-poor.
Video: Accelerating AI with Unified and Deduplicated Data Strategies | Summary: Leveraging deduped, unified data boosts strategic advantage and accelerates AI development.
Video: Compressing the Price of Not Knowing: Why You Can’t Have AI Without an IA (Information Architecture) | Duration: 3616s | Summary: Compressing the Price of Not Knowing: Why You Can’t Have AI Without an IA (Information Architecture) | Chapters: Introducing Information Architecture (27.98s), Augmenting Human Potential (83.28s), AI and Data Revolution (167.76s), Intelligence of Everything (269.445s), Data Understanding Challenges (363.28497s), Data Saves Lives (435.49s), Data Understanding Gap (517.17s), Evolution of AI (581.555s), AI Model Implications (688.45s), AI in Healthcare (769.4s), Data-Driven AI Differentiation (842.025s), AI Business Implications (1051.075s), AI Model Explainability (1129.01s), AI Security Challenges (1208.05s), AI Plus Business (1294.62s), AI-Reimagined Workflow Example (1401.57s), Democratizing Data Insights (1502.6549s), Data as Product (1631.05s), Enterprise Unstructured Data (1725.4451s), AI Data Optimization (1938.01s), AI Boosting Productivity (2137.6548s), AI's Transformative Potential (2271.225s), Business Value Framework (2384.3948s), AI in Customer Service (2502.645s), AI Capturing Expertise (2597.68s), AI Agents in Action (2726.595s), AI Agent Demonstration (2853.1501s), AI's Evolving Impact (2945.695s), Democratizing Data Strategy (3005.7349s), Conclusion and Book (3071.96s)
Transcript for "Compressing the Price of Not Knowing: Why You Can’t Have AI Without an IA (Information Architecture)":
Hi there, my name is Paul Zakopoulos and I have over thirty years of experience in data. I've been around for a long time, written 22 books on data and AI. From databases to data warehouses, I wrote Hadoop for Dummies when the big data craze came. And then into data science around Spark, AI and now generative AI. And there's one thing that data lets you do. It lets you compress the price of not knowing. And so there's so much craze and excitement around generative AI and agentic AI today that some people forget this tip I'm gonna give you. You can't have AI without an IA. What is an IA? An information architecture. And in this session, I'm gonna take you through what you need to get going and have the best AI journey possible. Let's start off with this Steve Jobs quote. I'll read it to you. What a computer is to me is the most remarkable tool that we have ever come up with. It's the equivalent of a bicycle for our minds. What did he mean by that? Steve Jobs was famous for being obsessed with beauty. You see it in Apple products today in font, eloquency, and more. But he was also obsessed with the efficiency of mechanics. And when he studied mechanics, he figured out or read up on what was the most efficient creature to cover a mile. You're probably thinking a leopard, a cheetah? It's actually a condor. Humans? About halfway down the list. But when they got on a bicycle, they went to the top of the list. And that was the point of that quote. But there's some interesting things here. By the time we're two and a half, we can reason about the space among us and around us, the equivalent of an adult chimpanzee. And of course, humans, I'd like to think, we keep getting more intelligent. And so what became the opportunity here? What sets us apart? Our ability to create things. That's what generative AI is. It's an ability to accelerate what we're capable of. We invent. And this is the opportunity that we're in now. And it's pretty key because if a computer is a bicycle for our minds, then AI and data are going to be the bicycle for your business. But AI is not a promise of prosperity. And so today, AI is transforming our lives at such a pace, what seemed like science fiction just years ago is reality today. Has anyone checked out these new AirPods? You put them in your ear, you click on each side, they translate for you. Imagine the friction that that removes from travel, from business. But I'll tell you one thing, nail salons going silent everywhere because anyone can understand what anyone is saying at a moment's notice. So the thing with data is it can all be expressed numerically. And if we can express something numerically, AI can go to work at it. Well, as it turns out, everything can be expressed numerically and that is the opportunity for your data and AI. Think about me saying the word hello, or maybe someone more famous than me. Hooray, Jerocco Leo, hello. Hello, Jerry. Hello, Newman. Those are WAV files. Well, at 16 megahertz, there's 16,000 numbers a second representing uncle Leo saying hello. That's pretty significant because those are numbers that I can put into an AI. And as it turns out, absolutely everything can be represented for AI. And we call this the digitization of anything. So you saw how I digitize the word hello but we can digitize anything. Look at BNSF Railway. They have 230,000 rail cars that move cargo and people across 33,000 miles of track. On those trains are over 2,000 sensors and 1,500,000 wheels. What's the biggest contributor to a train derailment that's almost impossible to spot? A cracked wheel. So they've gone through all kinds of techniques, manual inspections, computer aided visual inspections to try to figure out preventive maintenance on a cracked wheel or indications of tampering, etcetera, etcetera. They finally settled on the sound of the wheel. The vibrations in a crack start to change the dynamics. Think of this as shazamming a cracked wheel versus shazamming a song. But the bottom line is we can represent anything in data, which means we can put AI to work on that data. So we're in a really unique time. You see, have the internet of things, which connects everything. So now we're in the world of the internet of everything. And now we're gonna go into the world of the intelligence of everything. And that's why AI is so important. So we can put it on the edge where the data is because everything is connected. Honestly, I can't even vacuum my house without a WiFi connection. Think data on the edge intelligence of everything. That brings us to quite the conundrum, doesn't it? You see, when I first started writing books on data warehousing, I'd always tell people to find a needle in a haystack. But in this world, you're kind of looking for a needle in a stack of needles. That's why we need help. We don't need AI to replace humans. We need it to assist us. So I'm walking through the city of Las Vegas the other week and I realize I've become this kind of cartographer of the city and we're in this perpetual state of documentation. I'm getting offers, I use OpenTable, I'm looking around for directions, I'm constantly interacting with the city around me. And it got me thinking how everyone's so good at data collection, but I don't know how good we are at data understanding. And if I was to take data collection and understanding and look at the difference between them, there's some things you start thinking about. Data collection, that originates in real time, but understanding these days is executed in stale time. We're data rich, information poor, and when you subtract data understanding from the data that you collect, you have the price of not knowing. So let me give you an example of the price of not knowing. I assume many people know who that is, as Stevie Wonder. Not everybody knows he's blind, but most do. But most don't know why Stevie Wonder's blind. He's blind because of something called ROP, the retinopathy of prematurity. So when you're a neonatal baby, that Stevie Wonder was, you go into the NICU, the neonatal ICU in a hospital, and they oxygenate you. But too much oxygenation can burn the back of the eye and that's why he's blind. And back in the days when he was born, they would start to read data or collect data, maybe top of hour, and that's the kind of oxygen saturation they thought they were giving. And if you look in today's measures where we can measure data at the millisecond level, a thousand times a second, perhaps the oxygenation was at that level. And that's why Stevie Wonder was blind. So the price of not knowing was his eyesight. What's the price of not knowing for lives saved, opportunities lost, upsell, cross sell promotions? And today, we actually use this type of technology measuring physiological markers of a neonatal baby thousand times a second to predict the onset of diseases like pneumocycosis, which is all to say that more babies go home because we're putting data to work. So let's look at the price of not knowing in a different context. You're this sailor, you're stuck at sea, you're thirsty, there's not a drop to drink because all the water is saltwater. It's the same with your data. So let's look at that on this particular grid. We have our collected data, it's available data and we have our understood data and they have very different slopes. And the difference between those two, as you heard about, is what I call the price of not knowing. And your goal is to kind of go and shrink this price of not knowing. And you do that with various efforts and that's not smooth sailing. It goes up and down. You have starts and restarts. I always say fail fast, fail safe, and fail forward. But now you have to ask yourself a question. Is data a strategic advantage? What if you deduped your data? What if you could connect in a single namespace to all your unstructured data? You would start with a steeper understood data curve and that would accelerate your AI journey. Okay, so we have all this data and we're trying to steepen the understanding curve and we put technology to work, assuming you have the data and we'll talk about that later. And so over time, technologies have changed that have allowed us to understand and analyze this data better and better. So if I think back in time, we kind of start with expert systems and there we really had handcrafted symbolic representations of our data, which is to say that you had to write the rules yourself. Describe a cat, cat has triangle ears, triangle nose, almond eyes. You can imagine how brittle and how effective those systems really were. And then we kind of moved along in time into machine learning. And that brought us to an environment that was less handcrafted. In that environment, computers started to find patterns. We did a linear regression to try to understand how much a house would sell for based on the number of bedrooms, proximity to a school, proximity to a police department, shopping, etcetera, etcetera, etcetera. And then we go into another era and this is deep learning. This is where computers get more and more data than ever. Do you see us kind of come into the modern time and to get more power than ever with GPUs. And now they can create very strong rules of their own, which are extremely powerful for task specific features and their representations. Now, it's a matter of taking a picture of a cat and the AI is like, this is a cat. And that brings us to the world of today, which are foundation models, which is another word for large language models, which is all deep talk for generative AI. This is a fundamental change because we have encoded knowledge within these models like never before because of new innovations, one of them called the transformer, but there's others, but I'll put it to you like this. Whether you use Google Gemini, whether you use OpenAI ChatGPT, Anthropic, these models are simply indexes of the internet. They've taken all of the internet and put them to an AI in the new feature data representation. And that creates enormous opportunities, but enormous pitfalls as well. After all, ask yourself this, is everything on the internet true, safe? Is it free of racism? Not really. So you've got all the data in one place, a new way to analyze it and that is going to create new opportunities and new things you gotta think about. So here's the deal. Everyone's enamored with these large language models because of what they can do. People will show me how they wrote an entire email using a prompt. And while I think that's cool, technologies gets better every year. I've been using GPT since version one. Of course, ChatGPT is at 5.2 right now And every year it gets better and better. Just like that TV of yours gets thinner and thinner. That's technology. What democratized AI, what this moment in time is, is the prompt. It's the ability for anyone to interact with AI using natural language. That is the magic of today. And that allows employees at all companies to no longer walk by problems that could be solved or made better with technology. What an enormous set of opportunities now to compress the price of not knowing. Think of medical professionals and clinicians. If I think about healthcare today, maybe I'm gonna come in and 10% of that data are clinical factors, 30% are genomic, 60 are exogenous factors. These are the things that we can capture in numbers, just like Uncle Leo's voice, maybe it's coming from my watch, maybe it's all kinds of things tracking my diet. We can create vectorization representations of our lives. And now imagine what our health outcomes could be. And when you look at the ever increasing percentage of GDP cost that healthcare is and the aging population, this is one area that needs critical investment and AI's gonna help if we have the right data. So now I want you to think about your business. You might be in the business of saving lives, but maybe you're in the business of travel, insurance, manufacturing. How do you differentiate yourself from competitors? Well, distribution economies of scale, there's one way to do it. But you know what? Not everybody is UPS and Walmart. They don't have the supply chains that these companies have or the ability to move goods. So another way you can differentiate yourself is with network effects. Do you connect the many to the many but not everybody is Twitter or X, Meta, Facebook? But what is the opportunity that everybody has? Your data. Your enterprise data. And the ability to analyze historical data, for example, in manufacturing, and monitor equipment failures, to find leaks, faults, predict failures before they occur, minimize downtime, extend the lifespan of your electrical components. Think about this. This is enormous opportunity if you put your data to work. And I'm gonna put my data to work in the language of Pac Man. Okay. Here's my AI built Pac Man world. It's not winnable. It doesn't seem right. Like I'm kinda stuck there. I can't even get to these pellets. This is not gonna be good for me. What's the point in this? This is taking a regular large language model to try to understand your business because it's built on what? Everything on the Internet and not your business. So let's go see what happens when I feed a model information about my business, which is the business of Pac Man. As you can see, my world is a lot different. It looks more natural. Maybe there's an era I don't think I'm supposed to be able to go into their home. I'm still not very good at it, and I could use AI to help me get better at it, but that's the difference when you put your data to work with AI. If you just take a regular large language model off the shelf, you're gonna have a very different set of results than if you teach your model or provide your model data about your business, in this case, a winning Pac Man business. Now, not everybody is in the business of Pac Man. I understand that. But this is real business. Remember, to an AI, data is just data. Doesn't matter whether they're Pac Man ghosts and bullets or cosmetics. Over a century of formulations, science that they can put to work of consumer preferences, of taste, of colors by segment and cohorts, they can move all that to work. So realize we've gone through these epochs of data analysis and everything's the same. I have my data, I put it into something like the machine learning we talked about and I get some insights. But there's one rule that has never changed in the thirty one years that I've been involved in data. Data in is data out. We moved along to AI. Same bad data in just gave me more colorful data on the way out. And then along came generative AI, and that gave everybody the ability to work on data and get all kinds of interesting shapes of that output. And we're gonna talk about this later, but along comes AgenTik AI. This is the danger zone because you're going to create a veritable storm if you don't have the right data. And this is why your AI needs an eye in. AI is amazing. I'm gonna show you even more amazing demos coming up, but it's not magic. It's just numbers. And the thing that AI does is it connects data points and concludes irrespective of moral consequences. Now, let me give you an example. This is an airline and someone went to file a bereavement fair, which is a discounted fair. And they went to the chatbot and the chatbot said you have two months to apply for bereavement relief. They go to the funeral, they apply for bereavement relief and it's denied by the airlines. And the airlines rightfully points the customer to their website's term and conditions for bereavement flare, and it clearly says, you must purchase a bereavement ticket ahead of time. There are no refunds. This went to court. The plaintiff argued, the chatbot assured me I could do this. The defendant argued, these are our terms and conditions. The chatbot made it up. It hallucinated. It confabulated. The court found in favor of the plaintiff, which means you are responsible for your LLMs, which means you need to get your LLMs speaking your business. You need to create the Pac Man game that I created. What's a cat doing in my video? The world of AI loves cats. They always have. So I thought I'd put a cat in my video. Now, your AI is kinda like this cat. That airline didn't understand why the model made something up that wasn't true. Heck, they even gave it the right information, their policy controls that was clearly on the internet. And so understanding where the data came from, what did you do to that data, and why the large language model made a decision is really difficult. And you've gotta get everything in gear to do that. Otherwise, you're stuck like that airline whose large language model is like this cat who's trying to jump on this bar. And so the cat kinda thinks about it and just like the airline in a large language model, I can't see the math that's going on inside, but at some point, that cat has to make a jump, makes a response in the chatbot, and here it goes, and someone tries to catch it and it missed. The point is you need to be explainable and you need to be auditable. You need to keep track of changing data. Who did what to the data? Keep track of data lineage. Models drift. The moment you put a model into production is the moment it starts going out of date. These are all things you have to think about in your AI strategy. And this actually brings me to another point that is getting drastically overlooked and that is Audentic AI further amplifies the risks that you're already taking on with generative AI that I'm not convinced everyone understands. Look at, just like when a business found its way from solely brick and mortar to the internet, they had to deal with new attack vectors, denial of service, man in the middle, etcetera, etcetera. Well, it's the same as you become a more digital business and put generative AI work to your business. You have new attack vectors like prompt injection attacks. Now you have to protect from prompt and jailbreaking attacks, trying to trick your model into doing something you don't want it to do. There's things like data poisoning. So there's all kinds of new attack vectors. So while many organizations are still catching up with basic ones and AI ones along comes AgenTik AI, which we'll talk about. But here on this slide, you can see some of the things that you have to now start thinking about. And this is to say, and this is super important, you simply cannot look at cyber and data security as cost centers, which is how most people do. Security must. It absolutely must be seen as a value creator and your AI accelerator. So how do you get going with AI? I wrote about this in my last book and most companies operate in this manner. They have a business process and they add AI to it plus AI. This is probably 95% of AI you'll see today. In order for you to reap all the possible benefits that you can for your business with your generative AI and agentic journeys, you have to pivot. You have to think about AI plus And this is to say, look at your business processes and redefine them with AI at their core. Think about order to cache. There might be eight to 12 components in the order to cache monolithic workflow. In order for you to take the most advantage of AI, you have to break that completely apart and restart from the very beginning. And maybe 80% of those components are gonna be instrumented with AI first, maybe the others remain. That's how you get there. And this brings us back to the whole start of the presentation that you can't have AI without an IA. I'm going to say it time and time again. And a big part of having an information architecture is governance. And the problem with governance is most people see it as a least effort to comply approach. You look at a chief risk officer, how are they getting measured? Have we had a data leak, etcetera? Your job is to create regulatory dividends from your governance initiatives and repurpose this data to speed up your AI journey. That's AI plus and I like to think that AI provides the scalability for humans that provide the capability. So let's go see an example of an insurance agent and what they would do in an AI plus reimagined workflow. Now I'm an insurance agent reimagined in the AI plus world. I don't come in and log into Salesforce and log into a product recommendation. Then I don't go use my Copilot to write an email, copy and paste and put that in Outlook. Everything has been reimagined for AI plus What I just described is really Plus AI. There's AI all along the way, but we didn't reimagine the workflow and then you don't get the best benefit. So imagine I come in and I wanna write an upsell email to my customer. In the back end, one of those Bs you just saw goes off to Salesforce CRM. They go and find my clients. Obviously, can see a number of different life event changes, turning 64, turning 25. When you turn 25, you need to get your own insurance. So I decided I'm gonna work on John Collins. That goes off and calls a recommendation engine. The exact same logic on Netflix, which is telling you to watch Breaking Bad or Bridgerton, it goes in and says, we're recommending Silver Level Marketplace. Maybe I wanna go and give some ideas to it. I'm willing to give an 18% discount. Now it goes in another B, goes and actually writes the prompt, the prompt that's going to go into the backend engine. And you can see we have it here. It comes back with an email and then I would click apply and Outlook would go and send the email through an API. That is one reimagined workflow using AI plus This all comes together in what I call the wisdom of the crowd. Like, this is where you need to get with your strategy. The ability for everyone to ask new questions and get new insights. And now think about this for a moment. If everybody could ask questions, we've already said this is possible using natural language, which is the beauty of the front end of the large language model, then we could generate thousands of hypotheses to consider. And you know what? A bunch of them, they're no good. Maybe only 10% even matter. So then we'll get a 100 ideas that become validated and we could test and maybe only 10 go into production and you get this flow going round and round and round again Because the way it works today is, let me go to IT or let me go to some data analyst or a data scientist. Let me spend months getting the data set. They'll stand it up. They're not even in the business. We need to drive new questions and new insights to business end users. And you can do that now on the front end with a large language model because of the prompt, but you must deliver exceptional quality data. And that's the next part of this presentation. So here's the thing that's going on. Today, companies are purposely placing walls around curiosity. Accessing data, data in a data warehouse, which requires way more expenses than putting them into a data lake house, for example. So we guard access to data. Maybe different groups don't wanna share their data. Yes, there's privacy issues, but it's more than that. A lot of times, it's actually culture. But then there's another thing going on. They're also inadvertently applying a curiosity tax. That's because they don't have good governance. That's because they look at a least effort to comply and regulatory compliance versus thinking about the downstream dividends that come from governance such as How do you know this data existed? How do I find the data? Who did what to the data? What is the definition of client lifetime value? So we're all working with the same data, the same lineage, and we know the owner. And I'm gonna show you an example of what this looks like. We often talk about a data fabric, and this is about connecting your data with governed searchability and connectivity. If I can't connect to the data in a usable namespace, then I can't put the data to work. So when you do that, data utilization then becomes a process of knitting together data from across the business and externally. But the truth of the matter is, as the analyst, I don't care where the data resides. I just wanna go access it. And that lets me consume it, easily access it, and become self-service. Remember, the wisdom of the crowd. And then we take this to another level with what we call data as a product. And that gives you curation, governance, and lineage. You discover and access your created data. Imagine you own a piece of data and I use AI to produce data that I can share with people. These are detailed metrics on quality and a holistic view on data and how you're going to provision it and how it's being used. Now, of the crowd are those business users. They need to get at data as a product. They're just gonna shop for data. It's like you being on Amazon. Let me go get this client policies by coverage type. Now I drill into this data and I discover and reuse governed data with lineage in various sources for my AI and data driven use cases. As you're kind of getting the hint, Enterprise Gen AI is an unstructured data problem. In fact, it's an unstructured data problem at scale. And quite frankly, it's a retrieve and hope data problem at scale. The way that people are trying to use retrieval augmented generation is basically hoping that the AI gets it right. And this is exactly what you saw with that airline. Here are two things that you need to think about more and more as we wanna put our data to work and get the most out of AI. The first is 99% of the data for enterprises is not in these models. That's why the Pac Man game looks so good. And yeah, it's Pac Man. That's the language of my business. I'm in the business of Pac Man. Not really, but you get the point. Uncle Leo or maybe you're underwriting risk at an insurance company. To an AI, they're just numbers. They're just vectors. So how do we get your data safely to make a better Pac Man game for your business? Number two, most of this data is unstructured. So let me briefly just describe what some of the problems are of getting to unstructured data using the approaches most customers are using today called RAG, retrieval augmented generation. And it can't fully handle the semantics of enterprise unstructured data. Now think about this. In a traditional data warehouse or database, you have foreign and primary key relationships, referential integrity. These are all business rules. When we put them into a large language model, we basically create a vector space. That is representations of numbers of what things are close to each other. So it's like we're forgetting what we already knew. We already knew that this purchase order was connected to that client number, and now we're gonna break it apart and hope the AI finds it. Number two is inconsistent governance. When you think about it, I may have access controls of what I can access, what I'm allowed to change in the database. In this pattern, I pull the data out, I give it to the large language model, and then I have to create work on the other side to make sure I don't give you data like what is my boss's salary that you're not allowed to access. Once again, we're forgetting what we already knew. So you have to think of different ways in which you're gonna get around this and those emerging patterns for this. But let me show you the problem a little more. Here's a typical invoice. You can see I have the entity, I have some sensitive information, I have how much the invoice was. Text is stripped from the document then chunked and vectorized. So as I said earlier, the large language model can infer stuff that's labeled or is obvious to us. So it has a problem and it tries to make these connections. And of course, I mentioned earlier, there's a lack of governance and access control. So these are the things that make unstructured data a problem. It starts with just getting to the data and then we have to think about what we're going to do with that data when it enters the model. And this is gonna be critical for you because the next generation of AI agents and applications are more analytical and operational in nature. All right, here are my tips for relevant data. Number one, access any kind of data. Open source matters here, OpenTable formats, but you have to get access to your OLTP data, to your business intelligence data, to your unstructured data, everywhere. This matters. Data is a capability, not a destination. People wanna lift up data and bring it somewhere to do all the analysis. But there's so much data that's impossible. And if you remember, I talked about the intelligence of everything. That means we have to put function to the data. And number three, findable and visible. We talked about that earlier. That's data product. This is how you get wisdom of the crowd going. Let's look at tip number two. At the very start, I've talked about this. Sometimes I think culture is hard and technology is easy, quite frankly. This is a culture issue. Do not use the least effort to comply for regulatory compliance. My hint for you here or tip, use AI. When I bring the data in, AI can understand, is this data a credit card data? Then it's PII. I have to mask it with my business policy and my masking rules. Don't forget to de dupe. 20% of data right now is duplicated in an enterprise, we figure. And when you start pumping that data into agents, watch out. Next tip, prevent data leakage. I think you already know this. This is good standard governance. And I think number three is the new one that's hitting, and that is sovereign and compliant. You have to be compliant, But sovereignty of your data is mattering more and more, especially in this geopolitical world. That matters is probably the number one thing that customers are talking to me besides AI today. Finally, on the third piece, AI reliability and get cost efficient. Number one, optimize data for price performance and footprint. I want you to think about those data warehouses you have, that data lake that you've built, and I want you to think of a modern lake house. It's very different. And in that modern lake house, you'll think of a metadata layer that allows data to sit where it needs to sit. Now, this is so important because it gives you the opportunity to spend money to save money, take a lake house in, and then I take the data lake and the data warehouse, take those costs out, and I'm really spending money to make money because now I'm getting more data into the flows for wisdom of the crowd. Come up with new ideas and higher fidelity. So on the previous slide, I showed you how an AI with RAG is guessing where the vector representation is of the purchase order to the same customer order or that particular customer's order and so on and so forth, right? So think about it as lowering the fidelity of your data. If you feed that lower fidelity data into an agentic pipeline, you're wasting tokens. So you gotta spend some time researching on that kind of stuff. So we're in this lift shift, rift or cliff moment with generative AI. And I've talked about data so far in this presentation and why your AI needs an IA. But now I actually wanna show what happens when you actually have an IA, what your AI game looks like. And think about growing your business, or how about growing an economy? There's really three levers. There's population, there's capital, and productivity. And And if you look around the world today, there's basically headwinds everywhere. Only five countries have a positive replacement birth rate. In other words, populations are shrinking. There's been frictions in capital. Now we've seen some interest rate relief in North America, but still there's geopolitical, there's legislations, there's governance, there's all kinds of things that are making folks weary about investment. And then there's productivity. In all but a few areas of the world, productivity is actually on the decline. But the thing is that AI can switch the productivity trend. It can put it as a tailwind. And when you do that, you can shift left the business. Now, I used to run development teams, and when we shift left, we meant about finding a defect earlier in the production cycle. And it would be the same if you were in auto manufacturing. But now you have the opportunity to shift left your business processes that you've redefined as AI plus And let me give you an example. This is a mechanized wheelchair. Every KPI by the United Nations says, if you're confined to a wheelchair, you lead a better quality of life in a mechanized wheelchair. Here's the thing. These are 300 pound wrecking balls And over a $100,000,000 of claims are handled by property and casualty insurance companies in North America alone a year. If you're in one of these, you have a chance of getting in an accident. And if you get in an accident, there's a really good chance you're going to the hospital. What's the shift left moment? It would cost maybe $50 $50 to rig one of these with sensors. And so these are things where I say, we walk by problems we could solve every day with technology. This is a problem solvable. We could shift left personal injury and costs. And your ultimate shift left opportunity is to end what I call swivel chair work. What's swivel chair work? Just ask yourself, how often do you do it? How much time does it take? And probably, if you hate it. And there's an enormous opportunity around language with shift left. In fact, Goldman Sachs came out and said most businesses are between 25 to 44% efficiency where AI could help on these type of squibble chair work tasks that are language oriented. But there's another thing that often gets overlooked. Shifting left is not just a replacement effort. It's not just replacing the way you did things. It's this opportunity to start working on problems you could never do before. Maybe they were impossible more accurately. They just didn't scale from a cost perspective. But now they can, like reviewing millions of comments on a product. And when you shift left, now you're spending money to save money. You shift right, spending money to make money. Thinking of new ways to bring services and products to your customer set. Here's a great example. Using computer vision and generative AI to do on the spot daily pricing to underwrite risk for proper gowning on a construction site. Maybe the crane operators have been violated and fined. They will pay a premium for the next month, but folks on welding are doing just fine. It allows us to anticipate risks and adjust policies proactively. That's a brand new way to think about underwriting risk on a construction site. And so the good news is we have AI because data collection was so much bigger than data understanding and technology can augment knowledge work by mimicking work capabilities. Get a watch, feel, listen, understand, annotate, observe. And if we redefine our work processes from Plus AI to AI plus we can really put this to work. And before I show you some more live use cases, I think it's important I take a couple of moments for you to get the use cases right. Like I mentioned earlier, I wrote the book, Hadoop for Dummies, and I was part of the big data craze and I could see some of the mistakes I made. And I think a lot of problems, especially the banks had, is they didn't create business value use cases. So I came up with this framework. You've heard me talk about it a little bit. Spend money to save money, spend money to make money. It's pretty simple, isn't it? You wake up every day and you have a budget and you're gonna spend money to save money or spend money to make money. And then I want you to think a grid of your accomplishments of what you're trying to do. There's value everywhere here from cost reduction, modernization, insights driven, and transformation. But as you move to the right, you get a steeper value curve. You get more for your buck. But I'm going to tell you, don't just start at the right. Everybody wants to start there. There's lessons to be learned, there are dividends to be accrued and there's value to be had on the left. Now, if I was to think of it a different way, if I spend money to save money, that's like renovating. If I spend money to make money, that's like innovating. And then I want you to come up with your use cases. Here are a whole bunch of use cases that I came up with that are around a electrician. And then you map these to the board. Sit down with your partners, internal, external, and put them into these areas. And then from there, here's the best tip I can give you. Take another level of categorization, automation, optimization and prediction. Those are the three things that AI does. And I want you to start with automation. That's the easiest place to start. Internal automation. And at all times, about how do I automate the mundane and elevate the complex. And like I say, start with some non strategic low risk internal workflows and you'll be fine. So I like to say that language, especially with generative AI is now a technology And they will become the ears and the mouths of a digital workforce, agents, and more because they can do things like summarize, read, and interpret. And I wanna show you a bunch of examples that really let you take that opportunity that Goldman Sachs talked about between 2544% of shift leftable moments with language. Let's think about you're a social media curator. You wanna see what people are saying about your business. Here's a social media application I built called Whatabuzz. Let's go check it out. I open up with a map. Maybe I wanna go and see what the reviews are because I work at the Hilton Alar in Las Vegas. Now I'm gonna go search on that. I use a Google Map API. It goes and finds hotels of that name. And when I click on this, AI is gonna go to work. That little bee is busy looking at all of the reviews that are going on about this hotel. And you can see very quickly, it just summarized about 1,700 reviews. I've been at this hotel, it's pretty good. So you go through and you can see different reviews and someone isn't very happy. Well, I wanna go respond to this person. Now I'm using AI to generate the response. I get a response. Of course, I love human in the loop. So now I can go and adjust. Now think about that for a moment. Just how much did I shift left the work for that social media manager? So remember I talked about population? In my last book, wrote about what I call the silver tsunami. Now think about this for a moment. 22% of The US population will be seniors by 2040. Think about the institutional knowledge that is leaving organizations. We need to capture that in AI because everything can be captured, intuitions, knowledge. Think about an insurance adjuster who's worked for thirty years looking at cars and accidents. They could walk by a car and with high probability and accuracy, tell you this damage is a thousand, that one is 10,000, that one's a complete write off. I wanted to capture the expertise of a recruiter advising people on a LinkedIn profile and getting that into an AI. Once again, whether it's Pac Man, Uncle Leo, or recruiting advice, we can represent this in AI. How do we build this particular AI? We interviewed dozens of recruiters, transcripted all of their calls, and fed an AI that information with a couple of other tweaks. Now let me go show you how it works. This is my friend Liz Evans. She's an accredited family therapist who works in the Durham Region social courts. Here, I entered her LinkedIn profile and I ran it through an AI. And the AI came back and starts to tell me about her profile. Likes her headline, clearly communicates her professional identity and specialization, but the absence of a summary in her LinkedIn profile is a missed opportunity. It goes on to say, Here are some great examples that you could put in here, maybe a more impactful headline. You could even interact with the AI and it will come back with some suggestions. You're literally talking to your recruitment advisor. Imagine being able to talk to a document, an RFP, warranty, policy underwriting, or even a pathology report on my knee. That would be a really neat opportunity to meet people where they are in knowledge. So let me show you an example here. This is my medical provider and I just had an MRI on my knee and I'm gonna go see what's going on. So I type in what's wrong with my knee and back comes my MRI report. This is kind of interesting. I have posterior weight bearing surface of the lateral femoral conda. I don't even know what any of that means. That's very difficult for me to understand. I'm not a medical professional. But what if the AI could meet me where I was? So I typed this in. I'm not a medical professional. Explain my knee to me like I'm 16 years old and a huge Taylor Swift fan that has seen all of her concerts. I enter that in and now the AI is gonna have a conversation with me grounded on Taylor Swift. It talks about how my knee is like the smooth surface of a dance floor, but in this case, there's a flap. There's a hole in the dance floor. Oh, no. What's the point? We all are at different levels. Meet users where they are. What a phenomenal technology as a language use case. Alright. Now the hottest topic in generative AI right now around agents. And agents are not magic. They do the same steps a human would do, just real fast. Let me give you the differentiator. The way you use a large language model today is you probably open Google Gemini, ChatGPT, whatever your favorite model is. You start typing in. It gives you an answer. You type in. It gives you an answer. Agents are automated. They'll make their own decisions without you lifting a finger. Think about a typical order to cash processing workflow. All the things that go in there for late payments, initial billing. Imagine if an agent oversaw that, continually monitor the order. If a high value order goes unpaid by a certain point, an agent calls a credit risk model. We look at that risk. Is it going to be lost? Are we gonna have to write it down? Depending on the score, the agent escalates it to the credit control team that might send them an offer for partial sentiment or a reminder to pay. But now I wanna go show you an agent live. Go write a blog about the impact of lowered interest rates in The US and its potential impact on the housing market. That's the goal. Let's see how it works. Let's go write our blog. You can see it was the same blog that we had. Now we're gonna let our agents go to work. That lead market analyst I talked about, that's the agent who's gonna go to work. Do you see how they have a thought? I need to gather the most recent news and updates related to lowered interest rates in The United States. So it starts to work and it goes and has an answer. It passes off to the chief data strategist. It has its own thought. It starts to create data and on and on it goes running. Of course, I have a front end to that. You can see my lead market analyst is done. My chief data strategist is done. My creative content director is running on the final bit and the content officer is just waiting for that final approval. In the end, in my agentic workload, I'm going to go and have a blog written, put it on my WordPress site and post it into LinkedIn. And now you can see the chief content officer is putting on the finishing touches. Okay, everyone's done their work. Let's go see what happened. Here's my LinkedIn profile. Sure enough, here's my article. I'll go click on it. And AI put it together, gave me some imagery and on I went and wrote. Look at, I'm not telling you this is the way you should write everything, but wow, what a way to get started. All right, we're near the end, but I want to address one thing before I give you the final tips. I was walking in the streets of New York the other day and I saw this billboard. AI took my job and it got me thinking. I'm 54 and when I grew up, it was about white or blue collar jobs. And that's really the workforce of yesterday. I want you to really think about new collar jobs. Those are the workforce of the future. From the boiler room to the boardroom, AI is going to be part of the working landscape. And this landscape is changing very, very fast. I used to write how tech years are like dog years, one to seven. They're like mice years right now, one to thirty. Learn every day. And if you do that, I don't think AI's gonna take your job. I actually think it's gonna take it to the next level. And there you go folks, why your AI needs an IA. You're gonna hear lots about generative AI and AI, but you're not hearing lots about that IA. So let's sum it up. Number one, we want a hybrid platform. Any data, any place, function the data, data gravity matters. Number two, curate and catalog your data. No least effort to comply. Remember the big dividends that can pay off for your AI journey. Next, trust your data content. If you do this stuff below, you can start to trust your data content. That's how you start to get wisdom of the crowd. Next, track your data journey. How are we using these things? With trust governance, what is utilization rates? This is gonna matter more and more, especially in an agentic world. Are we feeding it low fidelity data? That's gonna waste tokens, which are the future currency software. Finally, democratize your data. Wisdom of the crowd, invite the many. They now have the opportunity to inquire their data using the prompt. So, let's figure this out. I started this talk with the price of not knowing and how to compress it. I wanna finish it with start knowing more. QR code is a copy to the latest book written by myself, Rob Thomas and Kate Soule. It's getting some pretty cool reviews out there. It encapsulates so much of what I've talked about and so much other stuff that I wanted to talk to you about. But I hope you take this opportunity to remember, boiler room to boardroom. AI is gonna be part of your lives every day. Read the book and good luck.