Latest / The Enterprisors / AI: The Conversation You've Been Avoiding with Guest Gerard Fogarty from Precision AI

Transcript

AI: The Conversation You've Been Avoiding with Guest Gerard Fogarty from Precision AI · The Enterprisors

• Tim Meadows-Smith: Welcome to the flip. Every successful CEO faces a choice. Keep doing what got you here, but no further, or flip to building what gets you there. Two 20 million pound businesses make different choices. 10 years later, one is worth 35 million, the other is worth 4.4 billion. That's not luck. That's not market timing. That's the flip. From founder dependent to enterprise, from hero to architect, from 35 million to 4.4 billion, this is the flip. Hi, I'm Jo and I'm here with Tim and we are talking to lovely Gerard Fogarty about the AI conversation we've been avoiding. Tim, how do know Gerard? I came across Gerard ⁓ through my business turnaround work and I'm happy to say that Gerard is now ⁓ working in one of the organisations that I helped to lead and is leading a new practice ⁓ helping mid-sized business owners understand how they create some more time and capture some new efficiency using AI. But I think the big thing that I'm learning from the many mid-sized business owners that I meet is it's now started to become embarrassing. because everybody's heard about AI, everybody knows that they ought to or thinks they ought to know something about it. And it's getting a bit embarrassing to ask what it can do in a mid-sized business context. Yeah, I'm sure. So, Gerard, help. So what is the elephant in the room? What do senior leaders, CEOs need to know about AI? What don't we know? so firstly, thank you guys for having me. I really appreciate the time to speak on this. I think the biggest key factor for CEOs and I think business leadership to know is that it's okay not to know. ⁓ AI is one of these strange technologies that came along where it feels like it's so ingrained in society and culture at the minute that what would be normal to say, I don't know how to rewire a plug socket or I don't know derivatives trading to say that you don't know AI seems almost like saying you know email, you don't know Microsoft Office, you're not proficient at it. But we didn't know email, we didn't know Microsoft Office a while back. So why are we embarrassed to say we don't know AI? I'm not too sure. I think because it is so culturally here at the moment and everything you hear about is about AI and everything you read about is about AI. It's well known that it's going to transform businesses. So for business leadership to not know about AI seems quite embarrassing. And I've sat in multiple rooms with CEOs and talked about solutions and even had a CEO on point of an AI company that I worked for describe things that AI can't do to a potential client and the client nodded along and said, ⁓ that sounds amazing. Let's do it. And you have to cut across the room as the technology expert and say, actually, I'm really sorry. That's not possible. But what we can do is approximate and get you halfway there. Wow. So even within the CEOs of AI businesses, I think there's a big difference between strategy driven knowledge and technology driven knowledge. Right. I don't expect a CEO to know ⁓ architecture. don't expect a CEO to know specifically what vector space means, but I do expect a CEO to know the difference between what AI can do and what it can't do. Absolutely. And the difference to know what it can do for their business specifically and how we're going to get there. AI is this incredible catch-all term. ⁓ And I know we spoke a little bit before we started recording, but ⁓ I think an analogy that my co-founder Aditya of Precision AI uses quite well, I'm going to steal it from him is AI is this incredible catch all term, it will be the same as if we didn't have a term for a car travel, train travel, sea travel, air travel. And we just called everything travel. And to come in as an expert in travel mechanics could mean that you're anything from a car mechanic all the way up to an aeronautics engineer. And there will be no differentiation for you as the person who's hiring me to know what that means. And I think what's important here is that even AI engineers have specialties. So to expect to see you know absolutely everything about every aspect of AI is more to ask them to be this all seeing all knowing entity rather than an entrepreneur who's gotten this far. And I think that's the biggest takeaway of this is that it's okay not to know. You just need to know who to ask and what to ask. We go back to the Henry Ford thing, don't we? We also go back to that ⁓ need that so many of us business leaders need, which is to be the smartest person in the room. thing that holds us back is attempting to be the smartest person in the room and guys and girls believe me in the presence of these AI engineers you're not the smartest person in the room. then as Derard says you don't need to be the smartest person in It's that whole thing about what you said but it's like I can I know how to turn a light on I don't need to know how that works it's like I know how to use the internet I don't need to know how that works so there's a whole it seems that possibly because AI has come so quickly and it's literally as we're talking, it's improving as it goes. We seem to think we need to understand the whole lot straight away. Exactly. To draw on that analogy you just used, it's the difference between knowing how electricity flows through your building and knowing how to switch on a light bulb and the difference between knowing not to pour water on it. Yes. Some of the knowledge has to be intrinsically in you and you need to know the difference between that strategy, but to know everything about it is not necessary. And I think what you're seeing at the minute is, especially on social media, whether that be LinkedIn, all the way through to TikTok, is AI influencers have a prerequisite drive to tell you about how they're experts in AI. They need to, that's where they get their finance from. They need you to know that every piece of AI is so important to your business and it's going to transform your business. yeah, we're selling. But are you going to the people who are selling you based on fear or are you going to the people who are selling you based on strategy? What's actually going to bring impact to your business. True, but of course at this particular stage of the AI journey we don't know, we don't know what we don't know, we don't know who to go to and so how long have you been in AI so to speak? So my background was in computer science so we've done a little bit around AI through college. ⁓ I would say that the majority of my learning has been experimental. ⁓ I was quite lucky at the time in which Chatch GBT really reached an inflection point that we were working with a quite famous brand on some AI pilots. When did GPD reach its inflection point? I'd say about three years now. So it hasn't been around very long. kind of, it's one of those tools that a lot of the software developers would have known that it was coming out quickly because it was so capable of doing coding and they would have tried it out so that it wasn't that capable in comparison to them and shelved it. And what you saw was that it started to be driven as a productivity tool more than anything else. And then as more and more people started using it and learning it and becoming quote unquote experts in that technology. That's where you start to see a lot of productization of these tools, a lot of new tools coming up very quickly and a lot of iteration over what was available and what was capable at the time. And I think that's why we're edging towards this, this future of quote unquote AGI, where the AI is going to become as smart as humans and as capable as humans. And I think we're quite a bit off in that front. And it's not something I really want to focus on because to plan for an unknown future is very difficult for a business leader. Yeah, and that's scary, but isn't it? It's the unknown future. So let's just part that. Perfect. Okay, so ⁓ I'm a CEO and you come into my business. Other things that I definitely do need to know about AI. Not about AI, but you need to know about your business. I need to know about my business. So therefore I need to know how AI can help my business. Not so much. You just need to know what is currently processified, what is productized within your business. What is set in stone? What have you documented? What processes exist? And the reason why is that when you look at the difference between a very founder led business and an enterprise, that's the main difference. You're looking at the difference between some one person being the bottleneck of business. and the decision maker and a business where everything is delegated fairly and everyone knows what is going on because it has been documented. ⁓ we've spent a number of years telling organizations that their data is worth its weight in gold. We didn't tell them what to do with it. So they just kind of hoarded it. ⁓ If you have data in your organization that is well structured, well tagged, well sourced, that is a perfect opportunity for you to come in and explore where AI might be beneficial for you. If you're still relying on a founder's gut instinct to lead your business, AI can't query your gut instinct. So you need to first look at how you're storing your data and what the structure is, and then work backwards from there. ⁓ help me with that. Absolutely. I can do great work in terms of taking unstructured data and making it structured. It can help to simplify that process. It can also help you identify where you're not doing that right now. And I think that's the biggest... For me, as someone who comes into organizations and ideates solutions for them and tells them how to build solutions around AI, typically I can tell within the first few minutes of a conversation how it's going to go, because it's the difference between a founder saying, ⁓ my God, that sounds amazing, let's go, and an enterprise saying, okay, but we need to get approval from X, Y, and Z. We need to make sure the data is sourced and is anonymized for you. The longer process and the time for an AI build typically tends to point out how successful it's going to be. ⁓ because every AI when you boil it down is an input, a transformation and an output. As complex as you want to make that, if the inputs aren't sourced properly, if they're not stored properly, if you have this incredible AI tool that's going to solve your problems, but it still requires someone to go into folders and search around and manually find things, no matter how fast or incredible the AI solution is, you're still relying on that manual process, which means it can only be as fast as your people can move. So let me just ask. If you have two business owners with similar size businesses in the same market, one of them brings you in now and says, I want to embrace AI, what can it do for me? And the second one says, Jer, I don't believe all this AI stuff, go away and does nothing for another year. What would the difference in the two businesses be in 12 months time? So I would say there's, would split that into two. First is from an organizational perspective. When you have an organization that thinks AI is a lot of hype, it's something that is still taboo. What that does is it puts a very quiet, shadowy area in your organization. And that sounds quite malevolent, but within the scope of what is currently happening in your organization, people are already using AI. And we've seen that across the board. When I enter into an organization, there's a bit of a patchwork system going on where someone in sales is using AI to draft sales messages, sales emails. ⁓ Someone else in the finance department has got this incredible mathematical model that's running and they've developed it all from scratch and they know how it runs, but they don't really know how it runs. And then you have some people in the organization who have never even heard of ChatGPT. So what happens there is that as the folks in your organization are using AI and they feel as though it is a taboo and it's not allowed, there's a natural fear that if you're caught doing it, that there's going to be some sort of reprimand or some sort of idea that you're cheating the system, you're not doing the work that you've been hired to do. while that kind of shadowy force exists within your organization, you have very quiet data leakage happening. And that is salespeople who are taking contact lists and putting them into ChatGPT, getting them to segment them. Because ChatGPT is an open source, isn't it? open AI. not so much open source. When you're on the free tier, you effectively are offering that you're okay to be trained on. So when they come out with their new model and they're using sources for that data, you're effectively saying that it's okay for them to use that data. different tools have different tiers of what that looks like. But if you're an organization who is frowning upon AI, you have no insight into who's using what, no insight into who's paying for subscriptions themselves, and no insight into what kind of data is going out there. So by way of being an AI positive organization, you can effectively create a policy document that says, we know you're using AI, we're okay with it. ⁓ But here are the tools you're allowed to use, here are the tools you're not allowed to use. And if you're unsure, here's who to ask. And what that does is it just takes AI out of the shadows. It brings it back into the forefront where managers now know who's using what AI tools and whenever someone wants to use a new AI tool, they have to go through an approval process and your IT systems, your IT teams, your InfoSec teams can make sure that those tools have been vetted and they make sure that the data is staying secure within your organization. Are there any other ways of security leaks? 100%. ⁓ So within that kind of remit of putting in customer data within that remit of putting in your processes, there is also, I think, ⁓ a bit more of a security leak in this terms of. It wouldn't be your typical security leak in terms of data leaking out, but it is a bit of a security leak in terms of how operational your business is, is this idea of cognitive offloading. So I spoke to a friend of mine, she works as a manager for a lot of planning analysts within quite a large broadcasting company. And she told the story of these planning analysts who up until now had been using the same Excel formulas day in, day out. They knew exactly how to write them from scratch. It was part of their expertise. They started using AI because they could describe exactly what they wanted wanted to do, chatgbt would throw out the formula, they pasted in, everything worked great. Until you ask the question of, and what does that formula do? And they couldn't say it anymore. They couldn't tell. They hadn't written it. They had no idea of what it was doing. It kind of looked like it was doing the right thing, but they had no way of telling. And also those skills had atrophied over time. What used to be less than a thought to think about how to structure the right formula for Excel. Now suddenly was this ⁓ really deep thinking task they had to go through. There's both ⁓ a security risk in terms of data leakage from your company, and then also a security risk in terms of you losing the domain expertise. that exists in your company by way of it being offloaded to AI without you knowing. So you take control, you build your own AI agent, is that what they're called? I don't even think you have to go that far. Agent would be the right term, but it's not even about building your own systems. It's just about creating an environment where you know what's going on. Things are documented well, like any process in your organisation should be. Your AI policy should be documented as a first step and then you can go from there. You can have conversations with people, find out what tools they're using, why they're using them. You can find out the intended usage of these tools and then you can find out the unintended use of this tool whether that be through that cognitive offloading or through other manners as well. that all sounds a bit defensive, what about positive benefits from embracing the toolset? Absolutely. So that was on the organizational side. When it comes to the kind of growth side of a business, if you take a very AI positive approach and you start to sit down and actually work out a strategy around where you want to take your solutions, what solutions you want to build in AI into, at that point, you start compounding. It's that law of compounding that happens with interest, happens with organizations as well. By way of taking that first step and creating an AI policy document that is AI positive, by making sure that everyone feels supported in their roles and feels as though they can use AI tools to help expedite their work, not replace their workflows. You get people who bring you solutions, who bring you idea to solutions. And from there, that kind of compounding effect of expediting what is currently human driven, that compounding effect of replacing automatable tasks with AI that can do the same thing, means that your organization can focus on the things that really take that human element in. They can focus on the areas where creativity and intelligence drive those kind of business outcomes. and can help move them into the future. have you worked with companies that have successfully made that transition from the black market AI of people doing their own chat GPT on their desk and going, this is the formula, this is the formula, can you do it? And this is my sales letter, can you write that? And can you make this a bit more professional? Have you actually managed, you've been with companies that have successfully taken everybody's experience of AI and made it company policy and everybody knows what they're doing. Absolutely. So within that kind of framework, when you work with companies, and we've done this for our own company as well, is make sure that we're practicing what we preach. What we found is that differentiation in terms of your skill level of using AI. ⁓ becomes a little bit easier to manage over time. You're not bogged down with all of these new AI tools that are being announced. You're not in a situation where you have kind of paralysis by analysis. You're in a situation where you know what works for your role. You know what can help you do your job better and therefore you know what to ignore. So it's the difference between signal and noise. What you're paying attention to are the signals. What tool comes out that can help you do your job better in the future versus what tool that comes out that is a bit of hype and therefore you can kind of ignore right now. I think it's the biggest change for our own organization, especially building out these kind of AI positive policy documents. ⁓ but that happens across the board in most organizations. Usually the really strong return on investment ⁓ AI work that happens is the boring stuff. It's not the incredibly flashy, mystical, magical elements. What do mean by the boring stuff? So an example, I'm speaking quite frankly, I don't know how strong the NDA is. So for now, it's a story about a luxury car brand out of Italy, ⁓ specifically manufactured red cars. Yes. them to ⁓ speak about AI and we pitched them a lot of different use cases. Digital twins of cars so you can manipulate the chassis and see what the physics outcome would look like. Digital twins of their entire manufacturing process, you could see exactly where the energy draws coming from and practically plan. What they chose was a chat interface, bilingual, ⁓ English and Italian, that ran on their policy documents. And compared to the things that we were pitching them, it's incredibly boring. But I think it shows the difference between how an enterprise thinks and how more founder led businesses think is that on the manufacturing floor for them, everything is made on site and is made by hand other than the windows. So as they're going through, if anybody has a question that relates to regulatory compliance, ⁓ procedures, documentation archiving, anything at all. They would stop everything that was going on on the floor, ring someone who knew what they were talking about. That person would let them know exactly what the regulatory statement looks like, and then they would return to work. The people who were in charge of keeping up to date with the regulatory and compliance frameworks were so busy answering the questions, they couldn't keep up to date with the regulatory and compliance frameworks. And every time that the entire floor has to shut down, you're spending real hours not doing what you want to do, which is manufacturing cars. So when we came in, that was the pilot program that we ran with them, was not to do anything incredibly flashy, it was to solve a real problem that was costing them revenue and do so in a way that didn't replace anyone's job. It just helped to expedite the workflows that required their input. That's fascinating. That's amazing because a lot of people are worried that their jobs are going to go from AI. That's the whole thing about we're all going to be replaced by AI and you're only going to need one person at top to make sure that AI is doing it properly. Is that a myth? I don't think it's a myth. I think there will be job loss because of AI and I think we'd be remiss to expect anything otherwise. I think what we're kind of seeing right now is this we're in the middle of a bit of a cycle. What I mean by that is that a lot of organizations are pushing to automate everything. the more minimal their headcount is, the more maximal their revenue is, and it sounds great on paper. But what you're finding is that a lot of businesses are taking things that are very human-led, human-driven and unique to their organizations and trying to replace that and automate it fully with AI. And what happens in that situation is that if someone special in your organization is special because of the way that they sell or the way that they talk to customers or how they connect with them, and you replace that with AI immediately, even if you you try to copy exactly what their cadence is and what their voice and their tone, how they sell is, you're effectively gray scaling your business. You're taking something that is colorful and unique and dynamic, and you're just evening it out. You're using the same AI model that every one of your competitors is using to do the same job that every one of your competitors is trying to do. And what you'll naturally find is as we go through this cycle is that we find that after we've gray scaled everything in our business, we then have to reintroduce the color back in again. come by humans, it will only come by their creativity, from their intelligence and from the lived experience that they have, which means that when I try and sell you something, typically I'm not a very good salesperson. I don't try to sell you something because I know what it can do and I know what problems you're having. typically if I'm trying to sell you something, I'll tell you, here's my solution, here's everything it does, here's everything it doesn't do, if you'd like it, here's the price. If not... We're probably not the right fit for you. Usually when I make sales because of that, it's not because I'm a good salesperson. It's because they needed the solution that I was selling. And I think that lack of empathy that AI is going to inevitably have, that lack of lived experience, that lack of being able to dynamically switch how you're talking to people is that kind of natural cycle that you're seeing at the minute. We're about to reach the top and then slowly but surely we'll come back to that human led Yeah, I think so. Because we as humans want that interaction. We are pack beasts, we don't want to be isolated and left to deal with robots. I, for example, get really pissed off if I have to go through an automated telephone system to get to somebody. So there's going to be a level, if we have to do that on every single interaction, we're going to go, no, we'll go to the company that's got nice little people that we can talk to, can have proper conversation. Even though those people are more likely to make mistakes than the AI. person at the other end, but we crave that, I agree. So do you think there's an opportunity there for job creation with AI in your company as well? 100%. I think for me... So about two years ago, I was diagnosed with a tremor. And it's a tremor that will naturally get worse with time. And there's no way to kind of plot how worse it gets. It's no way to figure out where I'm going to be in 10 or 15 years. But for someone who actively codes, the idea of losing the functionality in my hands is something that's quite difficult to deal with. I'm really sorry. ⁓ thank you. In the two years since that diagnosis, I haven't listened to the doctors. I haven't limited my caffeine intake. I haven't gotten enough sleep. But my anxiety around that kind of loss of function has quite the gone away because now I have incredible accessibility tools that were built as productivity tools but have a natural knock on effect. I don't need to spend time writing documents because I can speak them and the AI can transcribe. Brilliant. I don't need to spend time writing a lot of code. What I can do is I can use my expertise in the field to properly manage AI coding agents to make sure that they're doing well and not creating liabilities that I have to take care of later on. There's been this real push for productivity in the AI space, but by naturally pushing for that, you've created a lot of accessibility tools. And something that I used to have as a pet project that I was trying to get sorted out was a way to use AI to read what was on screen as someone who is, might be visually impaired to be able to access the internet the same way that someone who is, who has that kind of vision can do so. And we watched AI do that overnight. We see AI driven browsers now. where realistically you could go on as someone who has a visual impairment and can have the AI read out all of the product reviews, describe the product in detail based on the image, and you could go through an entire transaction without ever having to see the screen anymore. And I think where you see a lot of the discourse around job loss, I think that's real. And I think as you see more and more jobs go away, it's very easy to paint AI as a villain. But AI is a capability. It's not a solution. It's like saying, ⁓ I have an AI solution. is akin to saying I have an electricity solution. It's not the enemy. It's the way that businesses are ⁓ integrating AI into their business, into the flows. And I think it's the knock-on effects of that, which I think we need to an eye on. Brilliant. So tell us bit about... the situation that a lot of mid-size business leaders would be in, which is that they've got roughly 50 people, all of whom are fully overloaded, couldn't take on another project. They've got a raft of projects that they'd like to be doing but don't have time to execute. And you're coming along and saying, you know, I think my AI is a more important project than the list you've already got. Am I right or is there something that you could use your services to help solve that problem of overload? Yeah, 100%. I think it's one of the biggest red flags if someone comes into your organization and says, my solution takes precedence over all of your solutions. Because firstly, again, as I mentioned before, AI is a capability, it's not a solution. So if I come in and say, I have an AI solution, it's going to solve all your problems, that should throw up red flags immediately. I think what we try to offer is as a first touch an AI strategy session. So sit down with you, figure out what are your growth goals? What's your USP? Figure out where you want to be in a couple years time and figure out what are your priorities right now and then work on an AI solution that would solve one of those problems in a real way. So whether that be automating through some of the tasks in which your team are being bogged down with and have to repeat day in day out, that's something that might be a really first project to work on. Whether it be a case of just providing you with AI policy documents, tools that would make your life a little bit easier that you don't have to buy from me, but I can tell you what tools are going to be useful and what are going to be worth the subscription. I think it's super important because that might alleviate just enough of that burden to free up some time to think about solutions that they might want to create as well. I think when you have these AI experts come in and tell you that their tool can do anything, nothing can do anything. When they tell you that their tool can do anything, ⁓ and that it's the most important thing in the world and that it's going to save you all the time in the world. I think that's when you have to start having a serious discussion about is that a tool that you want or is that a tool they're trying to sell you based on fear? And I think when you sit down and have an actual discussion around strategy, when you really understand what a business wants to do and where it wants to go, that helps us figure out should we be actually providing you a solution or should we be telling you no, don't talk to us, come back to us in about six months time when you solve these structural issues within your organization and that will give you a good foundation to build upon. So does that take the overwhelm out of the whole conversation? Because if you are running ⁓ a business and you haven't got much AI in place, the thought of having AI is just like, where do I start? Where on earth do I start? So to start with the basics for you to go, actually, it's not about having AI, it's about let's look at your business and see where AI can support you rather than you have to have AI and then you have to run your whole business around the AI structure. Do you have a lot of CEOs just taking a breath and go, oh, thank goodness you're here? 100%. Because what we try to build into strategy sessions is an element about education as well. You don't have to be an expert in AI to have a discussion with us about AI. That's why we're the experts. Yeah, absolutely. So when we sit down with you and we do a little bit of education around what AI is good at, what it's bad at, specifically where you should implement where you shouldn't implement and get a feel for what makes your business unique. That gives us a good remit to tell you what your strategy should be moving forward. And specifically, yes, we're the strategist. also the builders. But the reason why we start off with the strategy piece is that if you want to take that strategy document and not implement it tomorrow, or you want to take to someone that you know who builds AI, we would prefer that. Business is built on trust and AI is also built on trust as well. To come in and you that you have to build it with us, another red flag. Also to tell you you have to build it right now is me selling you on fear and what that's going to do to you and your business is if I tell you you have to build a solution right now and you believe me and we run ahead and build it. If the AI model provider shifts how their models perform, if they shift the pricing, if they shift anything around how their AI models are iterating, what they're capable of, that completely shifts the solution I just said to you. So if we try to push you into a system where you're running ahead and suddenly the foundation starts to shift and you realize that foundation is not built on concrete, it's built on sand, then you're never going to want to build a system where you're to have a bad reputation. So by way of coming in under the strategy piece, giving you the to walk away and go with a different provider at any stage. Our hope is that that provides some level of trust so that you feel there's enough trust to have that conversation and admit what you don't know so that we can tell you what you need to know and put you in a position to move forward. That leads me into another big fear. Go on. And that is that how do you pick out an expert AI solutions provider or advisor? Because it seems to me that many of the people out there who were selling services first heard about AI about the same time as I did. I would say the biggest. did you? when did you? Oh, a couple of years. Because it's been around since 2015. It's been around for years. There's a fantastic book that I read. can never remember the name, but I tell people about this, but it tracks the history of AI the whole way through the initial perceptron stages, which is like tiny little gateways all the way up until the first image recognition models, which used to be incredible back in the 90s. But if you had an image of a llama and you changed a single pixel, it would suddenly tell you it was an image of a spoon. It was built on those pattern recognitions. So AI has been around forever and majority of the services that you use today are powered by AI. Where ChatGBT came out, hit its peak and we started to refer to everything as AI is what's known as generative AI. It's just a sub-discipline. Sorry, so I took away from the question that Tim asked, which was... How do I spot somebody who isn't just a page ahead of me in the book? Yeah. is guaranteed outcomes. Okay, you will guarantee outcomes? I will never guarantee after having done our strategy you'll do the strategy but you won't go and say this will get you this, this will get you that? No, because AI is a probabilistic tool. Any AI solution is built on the back of a lot of complex math, a lot of complex statistics and a lot of complex modeling. It's not a deterministic solution. You can never guarantee that 100 % of the time it's going to come out with the same response. So to tell you that I can guarantee you an accuracy rating of 100 % is inherently false. In humans, we wouldn't expect it, but in AI we do. And when you sell someone on that guaranteed outcome, that this will definitely work, this will definitely get it right 100 % of the time, and this will not cause any liabilities is where that biggest red flag comes in. I think when you look at AI as a solution, realistically, the reason why we have to build trust with the people that we talk to is because every solution that we create is a bit of an experimentation. We know the science behind it. We know the technology behind it. We know what it's good at and what it's not good at. But in every single iteration, every single solution that we provide, there's an element of experimentation and iteration that we have to do with you as the business owner and with your staff as well. If we want to create a solution for your staff, we have to understand how they are working. We have to understand what their workflows look like. We have to make sure that they know how to use the tool effectively. ⁓ No one would expect to just go into an organization that's been running on Microsoft Word this entire time to drop Microsoft Excel into the middle and to take two steps back and just assume they're going to figure it out. Most people get that kind of education as they go along, but it would be remiss of me to drop an AI tool into an organization and not tell them anything about it. So in terms of discern, ⁓ and where we're looking for an AI expert, we want someone to say something along those lines of, it's not guaranteed, we'll build with you, we want to understand your business, we will tell you how it will work in your business, we'll help people understand it and use it rather than this is your solution. Exactly. Now there is a bit of a differentiation between product providers and solution providers. Most product providers know what their product is well, they know what it doesn't do well and they can sell you based on that. It's that usual routine of say, that idea of selling a pen. You create demand and then you sell them to that demand. Where AI solutions come in is that it's typically a boil the ocean situation. I know what AI can do, it can do a lot. But if I try to sell you on the fact that it can boil the ocean, it means nothing to your organization. So I need to know how your organization runs. I need to know where the holes are. I need to know what the data is stored and I need to know what processes and workflows exist within your organization before I can tell you what my solution could do for you. Otherwise, it's just guesswork. So Gerald, I happen to know, but I don't know what it means, that you're one of the few people around who's trained a large language model from scratch and a large language model, LLM, is essentially what we're buying from the public service providers in what we call AI tools like ChatGPT and Claude and Gemini and... some of these others. So what does that mean when you come along as a supplier to a mid-sized company? So for a lot of them, typically means that I'm too expensive to work with. That's a bit of a misunderstanding. When it comes to AI technologies and tools, there are a lot of LLMs that are openly available. So the likes of OpenAI, can use an API to get responses from without having to train your own model from scratch. Similar, there's a lot of open source models that you can even take and host on a server to have a nice and secure environment. So that as users are using that connection to the language model, your data isn't going anywhere. Is that co-pilot? No, this would be effectively if you took open AI and you put it onto a server box in your organization and whenever someone talked to it they were just talking to that box. So it doesn't go anywhere else? Yeah, the actual organization doesn't see that data, they can't train on that data, they can't see what's happening. It's a much more secure way of developing and integrating AI into your business. There's a lot of open source models out there that can be picked up and put onto those servers so you can use them. A step further than that is the idea of fine tuning. So you take your data and you fine tune the model. So it gets a little bit better at identifying what, what makes your business different or where the processes might be different. The analogy I use is the difference between an MVP for me. My MVP is a minimal viable product. A sports person might be Michael Jordan. That's a very big distinguish. if I asked for an MVP from an LLM. So it's the idea of just customizing down that data, the processes, the answers to what makes sense to your business and what makes sense to your organization and your industry specifically. And then the step further than that would be effectively training a language model from scratch. So you take, yes, you take a full data set of text. In my case, was text. I was training a text to text model. ⁓ You collate everything down. You tag everything effectively. You go through this process of tokenizing everything. ⁓ So I think a big misunderstanding with using tools like Chach-GPT is that you think you're talking text to text. You're not. The first thing that that model does is take all of your text and turn it into numbers. It takes those numbers and it tries to figure out where the best route to get to the right numbers for your response are. And it does that through next token prediction. So effectively figuring out what's the most likely next part of the word that you're looking to hear from me and serving you that. So it's why if you ask Chachi BT how many irons there are in strawberry it can't tell. It doesn't see the word strawberry it just sees a series of numbers. It's trying to figure out what the most likely output to your question is based on the numbers that it knows and the numbers that you've just told it and then it tries to give you numbers back and transforms those back into text as the last step. within the... I I was talking to a real person. No I didn't. They can't understand anything. Which is why again knowing how to use the tool and knowing how to use them effectively is quite important. If you ask an AI how many letters, how many hours there are in strawberry, can't tell you. If you ask the AI now to create a Python script that will break the word strawberry down into every single character, count them, and then give you the output, it can do that. So it's understanding what are the limitations of the tool itself. And I think you get a lot of that through going through that process of experimenting, of training a model from scratch, seeing what works, seeing what doesn't work. And it also gives you an appreciation for enterprises who come in and request for you to train a model from scratch because as someone who's had my own cloud bill shoot through the roof because I've left a model training for a bit too long, you only make that mistake once. So what happened? Effectively as you're training these models, you need so much compute power and so much usage of graphics cards and so on. So typically you would use something like AWS or Azure to go off and train your model because you don't have that kind of compute power lying around. Effectively you rent the server space from... Amazon instead. As you go through that process, if you don't switch it off at the right time, if you let it run for too long, even if you host a model for too long, that compute bill that bill that you get at the end of the month for how much you've used tends to creep up quite a lot. ⁓ For an example, I left a language model that I hadn't trained. I just sourced open source. I placed it onto a server. I used it for a bit of querying. I left it over the weekend. I came back after the weekend and remembered I hadn't turned it off and I got a 5,000 pound cloud bill. And as someone who was doing that as an experiment, I can't bill that to a client. I can't bill that to anyone. I just have to pay the bill. So it gives you an appreciation for how much revenue has to go into training these models. from scratch and gives me a really good appreciation for knowing when that's not the right time. When the right time for you might be the different time for another enterprise. But it's about the appreciation of again, where your business is going, what you want to focus on, where you want to grow from. And then I can make the recommendation based on that. speaking of running out of time, think we just have... have to carry on because I don't... Me too. I don't know enough about that. Let's just leave a final message for people who have yet have heard of AI, but have yet to embrace it within their organization. Can you just give us a quick one minute checklist of what they should be thinking about doing within the next quarter? Absolutely. I can break it down for four weeks. Week one, visibility. Talk to people, get a feel for who's using tools, what tools they're using, why they're using them. Week two, create a very basic policy document. And that is just, again, that checklist of what tools are okay, what tools are not, what to do if you're unsure. and who to ask if you want to implement another tool. ⁓ Week three and four, start to figure out where AI could benefit you from a solution perspective. So maybe one or two use cases where you're using a lot of the same day in and day out, step by step formula to get whatever you need to get and figure out are there use cases that will actually get you some return on investment or if there are use cases that ⁓ are more flashy than they are useful. And then from that fourth week, start to iterate. Go back through your policy document talk to people, figure out is it working, is it not working, start to talk to people about those use cases, speak to professionals, understand what's the lift involved, how much will it cost, will it actually get you a return on investment. Once you've gone through those stages of the visibility, the policy document and started to iterate over use cases, you'll very quickly figure out what is snake oil, what is genuine useful technology and get an idea of who's using what in your organization and how you can best support them. Fabulous. I think we should leave it there. Thank you. It's such a privilege to be in a room where you're not the smartest person in the room. Absolutely. mean, I know we've run out of time, but I could carry on asking you questions all afternoon. So thank you so much for time, you very much. I appreciate it.