Why AI Has to be the Infrastructure, Not the Strategy ft. Elon Salfati

Hello, we are Bringing Data and AI to Life, a podcast by Informatica from Salesforce. I'm Amy Horowitz. I lead solutions sales here for North America. If you've ever been lost in the chaos of data and AI, you've come to the right place. We'll be conversing with industry experts who are here to shed light on the challenges that have rocked these arenas. We're here to bring clarity to the chaos myth, busting the confusing parts and providing insights and guidance for complex data problems by delivering trusted data for analytics and AI. Good afternoon everybody. My name is Amy Horowitz and I am your host today for Bringing Data and AI to Life podcast from Informatica. Today we're joined by Ilan Safati, the founder of the Selfati Group, an AI consulting firm that partners with Fortune 500 companies and high growth startups to solve complex business challenges through AI. I am so excited to have you on today, Elon. Why don't you give us a little bit of background about yourself and, and tell us what you do and why you're here and then we'll jump right in. Thank you very much, Emmy. I'm super excited to be here today. So as you mentioned, my name is Ivan Sarfati. I've been founded Safati Group I think four years ago, kind of from this necessity of, you know, a lot of people started to came to me over the years, especially after GPT3 came out, which is quite early on as we started to do a lot of some advanced modeling and how we can leverage them to kind of help different businesses. And over the past four years or so we've been working all the way from like YC startups to Fortune 500 companies across the US, Europe, some in the Middle east and over the past four years saw a lot of the same things. So a lot of organizations having these general challenges in how they are thinking about AI implementation and where we landed over the past four years is that we were it to become what we preach. So like this AI native agency have AI help us orchestrate everything that we do and so far it's been an amazing journey. That's great. Well listen, we have a lot of listeners out there that are just starting their AI journey. We also have listeners out there that have been doing it for a long time and it sounds like within your organization you've gone through a little bit of a transition yourself. So let's talk about AI in terms of B2B. So a lot of companies think, think that ChatGPT is it that the tools like ChatGPT are exactly where you need to be as far as just a tool set rather than integrating them into a system? From your experience, can you walk us through how organizations can move beyond thinking it's just one or two tools to truly embedding AI into their end to end business process? Yeah, so it's a great point and it's really what we have done so far with many organizations. I think the biggest challenge is people kind of look at AI as like another software they install and then you magic transformation and they are an AI native organization, but really they want to. So to all of our clients we kind of recommend to look on AI as an infrastructure. Really it's not around just solving like a one use case for one person in the organization, but it's about embedding AI into the end to end processes that are happening within the organization. So really what you want to do is to shift the question from what can we use AI for which is the majority of conversations that we see right now, to what's our operation model when AI orchestrated? Right. So a lot of the conversations that are happening right now over the Internet are we can use AI ChatGPT style and like the more advanced people look on, you know, a human with an army of agents. I think we've all heard that quite a lot recently, but really I, I think it's more an army of agents with a human. Like what happens when we are starting to let AI orchestrate it and when we're starting to understand what is the handoff protocols. Right. So when does AI reach out to humans to do the things that we want humans to do? Even though some processes can be managed by AI, we don't really want them to be. So the gap always show up on like data layers first and then how we kind of validate the operations and. Great. So can you talk specifically about how a business can map their workflows and pinpoint the right intervention points for AI? Some people to your point, just use it as a tool and aren't really putting it in their full process. Specifically if you can go into what you've seen, what task should AI augment and where should humans still be in the loop? Great question. So we are approaching this in kind of a three steps mechanism. It's always starting by mapping what the internal process look like right now. And it's funny to see how very small amount of organizations actually mapped out their processes. The majority of them don't. And specifically I'm very interested when we do these exercises about where did the data come from, what triggers different steps within the process, what are the outputs that we are expecting. Step number two is kind of how we score each step. And we usually look at it on two axes. One is what is the human judgment required to do this step. And the second thing is the cost of mistakes. And here we see a lot of organizations either have a very successful AI transformation or a very poor one. And the best processes that we have seen that benefit from AI augmentation are the ones that you want to take the success rate from 10% that exist right now in the organization to 60% and it will be still like a very large value for the organization. In the majority of cases, you would end up on like 80% success rate. So the third step there would be around, which we talked about a little bit earlier about how we designed handsoft protocol. Right. So when does AI trigger a human and when a human need to trigger an AI back? And those three pillars are, I think, what makes the majority of workflows successful on the AI journey. We actually have one example. So we have been working right now in the uk. I met a client, I also had a conversation at the House of Lords, the UK Parliament, around AI. And we have a client called Keyloop. And they have had a really successful journey in the past month. The thing that was really unique with this organization is that they started from a very clear KPI, like we want to reduce churn. This is the process that we have that causes this churn. We are right now very poor at this process. So here we are going to implement AI at various types, of course, functional operations within the organization. Then we get the success metric which moves the KPI. Time is the roi, right? Really making the process go faster and seeing an immediate difference. I think that's where we see a lot of customers and prospects struggle, is they don't know where to start and they don't know what to measure and what does success look like. So it sounds like that was a really cool meeting you had. That's pretty interesting. By the way, AI is talked about everywhere. We hear it in the boardroom, we hear it at dinner, it's everywhere. So to be able to have these conversations with our customers and prospects, I think what you just described is a lot of our listeners are really asking the same questions. Where do we start? Okay, I have to talk about the data. Obviously, with my background, everyone's ready for AI except your data. So let's talk about failure. Let's talk about Initiatives that sometimes are not seen as the best choice. So AI initiatives often fail not because of the models or the tool that they're using, but basically they fail because of poor data foundations. What does AI, and I'm using air quotes for those of you that are just listening. What does AI ready data look like? And how can organizations ensure they have the right data infrastructure for AI success? The data angle is so real. I have never seen an AI deployment whatsoever that succeeds with a broken data pipeline. And I would even push further. Right. Everyone are talking right now about AI ready data, but really AI ready data means clean data. And clean data in an enterprise is a myth. Right? So the vast majorities of projects that we've seen fail are the ones that are looking on too many data sources. So they look on. This is all the data that we have and we want to implement every use case possible and we are going from there. Really what you want to be doing is you want to start from the use case that connects to a data source that no matter like how imperfect it is, it's good enough to accommodate for the workflow and then you want to build outwards from that. It won't cover the entire organization. Enterprises are too big for that. But you start from what's clean and working or relatively clean and then you scale from that. And I think this is how you get to touch this data angle more and more. Yeah, it's so interesting. I hear this all the time. You know, when we first started the AI journey years ago, data that was just okay was good enough. And I think to your point, you can't do that anymore. Not only does it risk wrong answers coming out from AI decision making, but potential company reputational damage. Imagine banking giving the wrong information or hiring the wrong candidates because of the background or not even getting to through the interview process because of bias and things built in. So I love what you just said, it's really important. But we focus on the data foundation. Now let's talk about data silos, because that's a real thing. We talk about it all the time at Informatica. We hear our customers and our listeners talk about it. How can companies address the data silos which really in my opinion undermine the AI initiatives without embarking on an expensive multi year replatforming effort? I think what we hear, Elon, from our customers is, oh, we know we have data silos, but it is so expensive It'll take us 10 years to redo it. What have you seen in your experience? Yes, that's true. So to me, the most successful angle that we have seen is when we are looking on federated access. You don't want to move your data around. Like you said, it's like a three years project. It's unreasonable to accomplish. So you don't move data. What you want to do is you want to query it where it lives and then you want the AI system to patch together the answers to produce the insights. And when you are picking the first workflows that you want to augment with AI, really you want to pick the ones that especially like the first one. Right. The more mature you are, it's it easier it becomes. But you want to start with 2, 3 systems. So for example, with one of our clients, we connected to confluence, to Jira, to internal GitHub operations. It's not too big, but it allows us to run a successful prototype within weeks and not quarters. And if we look on data silos, really the hard part is, is not the technical one, it's the organizational one. Because in many, many cases what we see is that the person who owns the use case doesn't really own the data and then it's really hard to accomplish. We haven't even really talked about the change management of people, of process, of tools. It is real. And I think you just hit on it that as workers like me, the ability to leverage AI to do my job, there's some change management that has to go about it. Not just enabling, but really understanding. To your point, I might be the business owner in the workflow owner, but I don't own the data at this point. So really understanding that. I love that you hit on that. Let's talk about maturity. You know, executives hear it all the time and we hear it as well. I've just had a C level executive come out of a board meeting and they say, listen, we have to do AI. Let's talk about maturity for a second. So in your AI maturity framework, you focus on technology, organization and governance. How can the C suite executives assess their company readiness for AI? Are there tips and tricks? Where should they begin to invest to ensure there's an impact? Yeah. So as you said, we are measuring it on three dimensions, mostly technology, organization and governance. And usually from what we see, technology is actually the easiest one to accomplish. Right. The technology is there, we're all aware of it. Really where the challenge begins is on the organization, like we just mentioned, and on the governance. And if we look for a second on the organizational readiness, we kind of go back to who is the A accountable for the Success of these projects in the vast majority of organizations, there is no one person yet that says, my job is to make sure that AI produce good outcomes. And I'm measured by how good the AI efforts within the organizations are and how do we expect a successful AI transformation, if you will, when the person who owns the data is a different person from the person who owns the outcome and the KPI and is different from the person who owns the AI itself. And on the governance side, then I actually see that the biggest impact right now or the biggest challenge that is coming right now is actually in Europe and right now with the EU AI act coming into play. I think, August, if I'm remembering correctly, people need to start to document, you know, how they approach risk, how they approach compliance, how they monitor these things. AI generally is something that is very hard to understand in terms of how it does what it does. So we need to start to play to this role very quickly, especially around data and data access. Right. The other thing that we hadn't even hit on is budget. We've seen out from some of our prospects and some of our listeners have actually shared their stories that IT data management doesn't have a huge budget, but yet the AI executives and the people focused on the projects have all this money to go do AI, but they're not fixing the data problems at the back end. So I think you just described that, that not only is it a change management process, but really understanding who does what in the AI world within your organization and how to help the executives understand what success looks like. So it sounds like that's something that you help your customers with. Last question. So many organizations struggle with aligning their AI projects with their broader business strategies. Again, we have executives that come out of a board meeting. They say we have to do AI, but sometimes they're not aligning to business strategies. Can you tell us what you've seen that you consider main blockers that executives should look for when assessing AI readiness and how can they prioritize their actions for the most significant results? Yes. So it's a great question and I love that we already see like a pattern. It's always like this disconnected mandate. And the way that we tend to approach this, especially at the very beginning, is we have like a test that we're pushing our clients to go through. It's a very simple one. You know, you need to write one, one sentence. If this AI effort works, this specific matrix, whatever it will be changes by this amount in this timeframe. And if you cannot write that, don't start. It's too early for you. You need to get that right. And we see this again and again that the people who the vast majority of our clients are the tech people. So we talk to the directors of AIs, the CTOs, they get to have fun with the project, which I love. But eventually in many of the cases, the CFO own the kp. And the most successful projects that we had, like the example with Keylook, is that the person who owned the KPI, the head of Churn, is part of the effort. It's not disconnected, which made the effort successful. Right? So AI transformation is really like we don't want just transformation. It's a big world. What we want is the ability to pick a workflow, to augment it, to rethink it. With AI, not every workflow just need to stay the same and improve. With AI, start small, you walk your way outwards from that and see results and measure them. Yes. Amazing. Elon, thank you so much for your time today. I really enjoyed this conversation for our listeners out there. Please don't forget to like, share, follow download us wherever you get your podcast. Feel free to download us there. And until next time, thank you very much and everyone have a great day. Thanks, Elon. Thanks Amy. Stay tuned for more illuminating discussions until we meet next time. Keep harnessing the power of data and AI to bring transformative outcomes to your life and business. Make sure to click subscribe so you don't miss any future episodes. And tell your friends about us too. On behalf of the team here at Informatica, thank you for listening.