In this episode, Roy Castleman joins the podcast to share insights from his extensive experience in leveraging AI to run multiple businesses. The conversation delves into practical applications of AI, the challenges of selecting the right tools amongst many, and optimising AI usage through clear communication and iterative prompting. Roy emphasises the importance of understanding AI as an enhancement tool rather than a replacement and details a method of using AI to form a ‘virtual board of directors’ for business advice. The episode also touches on various aspects of experimenting with AI tools and offers valuable tips on using AI effectively.
00:00 Introduction and Welcome
00:20 Diving into AI and Business Efficiency
02:15 Choosing the Right AI Tools
03:45 AI as a Critical Thinking Partner
04:59 Practical Applications of AI
06:39 Challenges and Limitations of AI
09:10 Setting Up AI for Business
16:34 Effective AI Communication
24:29 Final Thoughts and Contact Information
Listen on Spotify or Apple Podcasts
Connect with Roy Castleman on LinkedIn here – https://www.linkedin.com/in/roycastleman
Connect with Daniel Welling on LinkedIn by clicking here – https://www.linkedin.com/in/danielwelling/
Connect with Adam Morris on LinkedIn here – https://www.linkedin.com/in/adamcmorris/
Visit The MSP Finance Team website, here – https://www.mspfinanceteam.com/
MSP Glossary: MSP Finance Glossary Explained | MSP Finance Team
We look forward to catching up with you on the next one. Stay tuned!
Transcript:
Dan: Roy, welcome to the podcast,
Roy: Thank.
Dan: Roy. We’ve known each other for. I’m gonna say six or seven years now, maybe a little longer. And I’ve always really enjoyed our conversations when we meet up for a coffee or a teams, and our most recent conversation was no different. We got stuck straight into AI and ultimately the number of it is well. how does one person run four businesses? And today in the modern world, AI is the answer. So, tell us, tell us all about it.
Roy: Well, I. It’s a, an accumulation of different things and AI now just makes it so much easier, doesn’t it? I think there’s so much noise out there at the moment with ai and this is such a problem for people, you know, there’s, you know, a hundred different tools or a thousand different tools you can use and the clarity. We need to bring to it is what is the business function that we need to fix and how can AI help us do that? And you know, just really digging into that, you know, for your own companies, your own elements is where it becomes. A lot tricky, right? You know, what is the business function that I’m trying to fix?
You know, what is the problem with that business function, you know? And then what are the tools that I need to use to enable that? And I maintain that. Over the last six months, I’ve probably done 6-7 years’ worth of work. And having built you, I’ve had probably 20 odd companies in the last. Years, you know, from those that just totally didn’t go anywhere. You to a massive adventure center that I bought in thousand 18, which later failed due to COVID through the it, the three IT companies that I run. So I know what working looks like, right? And I know how much. How time, effort it took me to get things moving before. Very good idea. Now, what does.
Dan: I, I always enjoy talking about your various, in interests outside of the IT world as well. But, we’ll maybe have you back on for another episode to talk about some of those, but, but yeah, I guess, if someone literally is, entirely new to this, where do you begin?
Which, which products do you pick? Do you pick first or which, which business needs do you pick first? Which is the, where are we gonna get the low hanging fruit from here?
Roy: I think first of all, you have to understand. That this is a process that you need to go through, whether it’s ChatGPT, Claude, any of the large language models that you need to understand, you know, you need to go through that whole process of understanding how the tools work. So yeah, use ChatGPT in your own life, just. You know, ask it questions, you know, open the fridge, take a picture of the food in the fridge and ask it to make you a menu, you know? and just get to that process of understanding the basics of the tools. You know, we, a lot of people are very afraid of AI at the moment. You know, it’s gonna take my job, it’s gonna make me less creative, it’s gonna do all these kind of things.
And what I found for myself as an entrepreneur, as a business owner, is it’s opposite what I mean by that. Initially, if we had to go out and do something, we have to go and learn. You know, so we go through a learning process and we kind of go and get some books and we go watch some YouTube videos and we learn, we take this information in. Yeah. Now we don’t even have to do that anymore. Right? You know, we can actually just find out what the problem is, be clear what the problem is, and from there we’re saying, okay, this is the problem. Ai, can you help me with this? the pitfalls in this are quite often that AI is a yes man, right? And then you, anybody that’s got any decent experience and is gonna understand AI is a yes man. You ask it a question and say, oh yeah, I think this is a great idea. And it’s gonna say, yeah, Roy, that’s a magic idea. Yeah. Even selling bloody soap on the side of the road’s a great idea. Go out and do that. If you ask it, I need you to be my critical thinking partner. I need you to help me understand what opportunities I’m missing, what, biases I’m bringing to the conversation. You know, and really be critical on this, you get a totally different answer. So it’s understanding how to work with these tools that is, you know, gonna make them so much more powerful for you.
Dan: So the power very much is in the prompts rather than the imagination as to what you like. start small and but be very specific about how you ask.
Roy: I think the power is becoming less in the promise than it was before. but I think the power is very much an understanding that AI is a tool, you that you can use. To you enhance your human, you know, rather than AI is the tool that’s gonna do everything for me. Yeah. I for example, use AI for automation a lot. And yeah, so, so what I’m doing, if I do a podcast, for example, I’ll take the podcast audio, drop it into a folder, it’ll go through a whole bunch. AI processes and spit out and connect onto my podcast tool in the end, that’s fine. I couldn’t get to that stage without understanding how the limitations of ai. AI is a 60% tool. It’s not even an 80% or 90% tool. It’ll do 60% of the job for you and do that well, but you have to be very clear on what you’re asking it to do. Otherwise, you get the hallucinations, you get chat saying all sorts of odd and strange things that just, you know, don’t work. The tools are moving ahead at such a rapid pace. That’s what’s happening is what I could do on ChatGPT version four. Yeah, it doesn’t work now on ChatGPT five, you know, or it works in a different way. So you need to keep your finger on the pulse. You need to be consistently moving forward and yeah, you don’t need to. Up with everything. Yeah. In terms of don’t go and buy all the new tools, don’t go and upgrade. You know, what’s the business problem that you’re dealing with? ’cause if you can deal with the business problem, get the business problem solved, you know, use an automation to do that, you know, then you can, you stay on track. I think trying to do too many things for too many people is you really where AI’s falling down because it just doesn’t do it that well.
Adam: Roy, I just wanna pick up on that part around being a critical thinking partner, which I really like that concept and I really like interweaving that into prompts appropriately. you mentioned also around how it hallucinates and it gets things wrong. and despite the fact we’re on version five, the ChatGPT and version X or whatever else, we, it’s still making these simple mistakes because it seems to be a yes person and it wants to come up with answers.
And, you know, I just had an example over the weekend where I was trying to hook up, some legacy speakers to a new tv. and do it under the remote control that came with the tv. and I was pretty sure it wouldn’t work, but AI was telling me it would, and, but it hallucinated, it made up a particular, menu in the TV because like, it would do that.
So I was just interested to know, have you’ve got any tips out there for our audience around how to reduce those sorts of basic errors? Any particular wordings or prompts or anything around that to make it a bit more accurate?
Roy: I would be very clear. Yeah. the prompting piece Yeah. That we spoke about initially is important. It’s important that you set the context. Yeah. Generally. You’re talking to AI and AI is a thousand different experts in the room. So in your example, if you’re saying, right, I want you to be an expert working on the Samsung model X, Y, Z, you screen. I want you to also be an expert with this model of, sound equipment or speaker. Yeah. Now, yeah, help me with this. Be sure that you don’t give me any information that’s not active and live on the internet at the moment. You know? So go out and check the manufacturer website. Yeah. when I talk about AI being a yes man, what you need to understand is AI is trained on the human condition. AI is trained on how people react and how people think, and how people, you know, wanna react. So when we go through the process of meeting a new person, what do we do? We try and show them our best face forward. We try and make sure we become friends. and AI at some point is trying to do that. It’s trying to move forward and say, oh, okay. how can I tell this person exactly what needs to. I actually find at the moment that ChatGPT five is way, way more hallucinogenic to use that term than Claude ai, for example. Claude has just taken a huge amount of steps forward, so, so there, there’s. Tools for purposes that really, you know, you need to understand and you need to, but you, before you even get there, you have to get to the point of, right, let’s use this every day and see what it’s doing, and then take my human, you know, be the architect of the conversation.
Be the leader of the conversation. You know, don’t give away your thought leadership. If you give away your thought leadership, you’re in for a hiding a hundred percent of the time.
Dan: So it sounds to me like, the different, tools actually have their own, characters and personalities, like people as well. Like you were referring to, you. One, one is like this and one is like that. and that’s why you are advising actually. You’ve gotta, get to know them in the same way as you would in a human, circumstance. Go networking.
Roy: yeah, a hundred percent. I’ve probably tested 350 to 400 different AI tools at the moment, from the, you know, the avatar creation pieces, to the audio creation pieces, you know, to. Things like Descript, which you, which I use for editing.
Uh, you know, I’ve got, a range of different, large language models from, you know, Genspark, which is an agent type, unit, uh, Claude, ChatGPT, Gemini, you know.
So they all have strengths in certain places, uh, and they change quite regularly. As they, they’re consistently going through this upgrade, upgrade, upgrade phase so quickly and so rapidly, you know, their properties change. So you need to keep an eye on that. And, you know, in some respects, again, if you’re gonna try and keep up to date with that all the time, you know, you’re gonna spend, I spend 10, 15 hours a week just playing with AI tools and doing different things. Yeah. And I do that so my clients don’t have to. I need to be able to sit there. After this, I’ve got a call with my AI community where I go through, I sit down and I say, right, okay, you know, what are the problems you’re trying to fix? And I try and bring them back into, listen, just fix the problem, but fix the problem with, you know, the clarity that you’re leading the conversation. ’cause if AI says This is a great idea. I quite often use ChatGPT here and Gemini here, and I’ll put the things in all three of them and see what comes out. Now a lot of people don’t have the, well, they believe they don’t have the budget to be able to do that, but when you’re paying for each of these a month, you can then use them to crosscheck each other and cross-checking each other.
is a very good idea. I mean, the other thing, you know, we spoke about you. How do we actually get AI to do this more effectively? Get it to site you where it’s getting things from you. so, so you can go in and check the sites because quite often chat will say you, the location where you’re getting information from, it’ll come back and you click on the links and there won’t even be anything there. so so you really need to, you know, own the thought leadership you and have AI as their thought partner.
Dan: really fascinating. Having a panel of ais and, not necessarily pit pitting them against each other, but, but certainly being selective about which ones? well, like, like, like building a team within your own business in terms of right, right person, right seats.
And, the, the same is true of, of ai as well. who’d have thought!
Roy: Yeah. And I think the other big, play that people are missing as business owners at the moment with AI is go you, if you’re a small business owner or you’re even, you have a team of people in the business that you’re working with, you don’t need to. Now just rely on those, that team, you can have an entire board of directors working for you, you know, at the level that you wanna. I I mentioned earlier that AI is like having a thousand experts in the room, and if you tell AI who it’s acting as. It can very much act as that person. I want you to be a top 1%, you know, a marketing specialist, you know, in social media. I want you to be a marketing specialist in email. I want you to be a marketing specialist in whatever that is. I need you to be a, you know, a CEO. I need you to act as a CTO or a COO, you know, and you can then have. the input that higher level of intelligence would give you from a person. Uh, and you can then take from that and say, okay, yes, that makes sense. I’ve had for some customers, I’ve set up an entire board for them, right?
So there’s a prompt for this for each person. Now that’s the board. And then I have, okay, so this is a NED. I want you to act as a ned and I want you to review what the board said. It just gives you more information to work on. And part of that is the problem because it can give you way too much information. I dunno, how many of you have got a thousand threads, you know, on your ChatGPT and, just trying to keep up with where that is, you know, is working itself. So yeah, bringing it down to, right, this is what we need to, I think is.
Adam: And can you just expand that a little bit into the practicalities of that particular scenario? we’re talking about setting up a number of NEDs for a small business in MSP, for example. how would you actually construct that to make it, Well, to bring some continuity into it, to bring some structure into it, some order, some search so that you can, you know, use these NEDs on a regular basis.
They remember what they said before, they’re aware perhaps of what the other NEDs have said. Perhaps, you know, how would you actually go about constructing it? is it, for example, projects in GPT or is it something else?
Roy: Well, between, you’ve got GPT, you’ve got gems, you’ve got a range of different, approaches there. I like the GPT approach, you know, because the GPT allow you to set your criteria, upfront. Now, you know, the GPT don’t allow you to remember the conversations as effectively. They can remember one conversation, remembering that within any large language model, the window, important thing. Is 4,000, 8,000, 10,000 characters. What? Whatever. It’s that They’re talking about 200,000 characters and some of them that the reality is you give, and you’d have noticed this within with any of the large language models you’re using. You go in, you start having conversation. You know, by the time you’re having the 10th or 15th iteration of that conversation, suddenly it’s forgotten what you said at the beginning. Right. So, you know, understanding that if I’m gonna go and do any complex prompting of this nature, I’ll go in and say, right, this is the prompt, and I’ll copy that prompt out, and then every fifth or sixth, prompt, I’ll put that in. Just reminder that this is the prompt, and just bring it back on track again. So in the, in direct answer to your question, there. The tricky part, with running out NEDS and things, unless you’re creating the full large language model, which is possible now is to just make sure you’re very clear on what that role is. You know, the same as you would with your staff. You, your NED can be, you can literally take. In the GPT, you can take a couple of different books. You can take different websites, you can put various things in there. Put all the information you wanna work on as the storage. Yeah. And then just have the prompt saying, I want you to act like this in cases. And then they’ll keep back and it’ll give you consistent answers for that subject. Does that make sense?
Dan: I think so. And. and again, the more I listen to describe the situation, the more it becomes apparent that you do have to have, you there’s no. There’s no AI pill that we can take that’s gonna get us straight to a level of mastery. you’ve gotta be familiar with and have used, used all of the different tools to build up in your own mind that comparison. Almost like, we’re, we are becoming a, an AI tool ourselves. and almost we need prompting to understand how to prompt.
Roy: I mean, the other big piece they are, that I talk about a lot is around communication, right? And what we believe we’re communicating as humans. You know, we believe we’re being very clear. We believe you know, that our communication is crystal clear. And as business owners and as leaders, quite often, you know, this is so far from the truth, it’s not even funny. And you know what I learned very clearly and I’ve been on this AI journey for two and a half, you know, three years, however long has been out. And, you know, you put sh1t in, you get sh1t out. Excuse the French. Yeah. and this is so true, you know, of ai and it’s very true of staff as well. I used to have a, I got a staff member in many years ago, George. and George was you, he was gonna come in and do some cold calling, so yeah, he was doing 50 calls a day, and so I came in, I was like, Hey, George, what I want you to do is I want you to, you get hold of the clients, give them a call, build some rapport, and then you can start going to these, you know. This the sales process and we went through this entire piece and a week later, George is just not doing what he needs to be doing and there’s just no response coming out. So I sat down with him and I listened to some of the calls. And George, you’re not building rapport. Well, what I don’t understand what you mean by rapport. Yeah. and so I wasn’t clear enough in my communication to make sure that was, you know, very much a, a set of guidelines. And this is really just amplified by ai and it’s amplified to such a degree, because ai, if you are not a hundred percent clear, you know, setting the context, right, setting the scope, right?
Setting, you know, the expectations, right? Setting the output, right. You’re not gonna get. The same thing, out that you’re expecting you’re gonna get something entirely different. Yeah. you put in some bias into the conversation, oh, I really like this and I really like this and I really like this. I think this should be at work. AI’s gonna agree with you. It’s gonna say yes. that’s just such a great idea. But if you go in very clearly in your communication and you set it clearly, you get much better out. And I think this is such a big play for why people are not using ai because they start, they go in and it just mirrors their ability to communicate. So it goes in, they say a whole bunch of stuff and it puts out what they want and they say, this is rubbish. And it’s like, yeah, it’s rubbish because put rubbish.
Adam: And in terms of that prompting, how do you recommend or what are some of the optimal ways of doing it? Are we talking about iterative successive prompts or are we talking about, you know, an optimum number of characters in that initial prompt? is there, are there some tips around optimising that approach?
Roy: I think the first step is to, if the output is wrong, you need to change the input, right? So if the output is wrong, don’t look to the tool and say the tool is wrong. You look to yourself first of all. So that’s, you know, just really reiterating what I said before. the next piece is to set. Who you’re wanting the AI to act as. This is getting better now, you know, but you know, the AI has the opportunity of being any one of these, you know, a thousand experts in the
room. So, set there, set the expectation of what the, the, AI is. Acting as, then set the context. The more context you can give it, you know, the better it’s gonna be at performing. And then be very detailed in what the problem is. Now, quite often you don’t even know what the problem is, you know, but you can say, this is the experience I’m having. You know, I don’t know. Really what the problem is here. Perhaps you can ask me, are you, and I normally go and say, act as this, you know, perform this action.
Ask me any questions until you’re a hundred percent clear that you can gimme the answer. Right? So get the, you know, get the AI to come back to you and say, right, I don’t really get what you’re saying here. Yeah. And that gives you, you know, the opportunity to say, right. yeah, this is what I actually mean.
Adam: Yeah. No, I really like that. I think I have heard that before, but it’s really great to hear you remind. Remind us of that actually. You need to use it to, to help you to get the right prompt to give it. So there’s a bit of toing and froing going on. definitely. No, that makes a lot.
Roy: would liken it to you go and you employ a graduate, right? And this graduates got all the information that you, they’ve learned from all there. They don’t have a real world experience of it. you’re gonna go in and say, right, you know, I need to go out and do this. And then they’re gonna say, well, actually, what does that mean to you?
How does, how do you want me to do that? And, you know. Just, just engage in the conversation. I like, chat’s way of talking to and from as well becauseI like the, you know, when you go through that conversation with chat pt, you can actually meet to this. When we hear it and we speak it,we,digest information in a different way. So, you know, going out, you know, the speech to text is very effective. Speaking and receiving is effective. and as humans, we. We act in different ways in, in, when we have different communication style styles. So using those communication styles to learn more about how the communication works. I mean, I’ll get up in the morning and I use both of these.
I’ll get up in the morning sometimes and say, listen, my brain is just so full. I’ve never been good at journaling. I just, yeah, I just never really worked for me, but I’ll use. AI as a right. I have this problem in the next two weeks. I need to. Then I’ll say, right now I need you to help me brainstorm, and I’ll put the audible side on, and I’ll put it on my whiteboard and say, right. Help me brainstorm this. Yeah. And then we back and forth and we brainstorm. so I’m using the tool to get me to my outputs, but I’m using it in a way that allows me. As an engineer, as MSP, uh, pieces, you understand this as an engineer. I can sit there with a problem myself for five hours trying different things. I put one other engineer in that mix and we start having a conversation and suddenly the problem’s fixed in five minutes. Even though the new engineer may not have given me much, it’s just a different mechanism of the brain is activated. Same with ai.
Adam: it’s fascinating, isn’t it? How when you have a conversation with someone and you are forced to articulate what’s in your head into words, you learn. In fact, you often find the answer to your own problem in just articulating the question better, don’t you? And
at the end of it, you say, oh, actually, I don’t need you.
Now. I’ve just worked it out. Thank you. You can go now. Yeah.
Roy: Yeah, and that’s been something from on the engineers that we’ve always had to do, is have that door open that they can talk to someone completed.
Adam: And so in your sort of rhythm of using AI in this way, are you talking to it more than you’re writing to it Just as a medium. Mm-hmm.
Roy: I think there’s a few things. Yeah. it’s, again, it’s what is the function that I’m trying to achieve you. If I know I’m, I, my brain’s very busy and I’m dumping, I will talk generally much more. I have found, however, that ChatGPT, whisper text is way better than anybody else’s. So it’s able to understand my docent African tones and, you know, in, in a better way. You know, I can’t use, you know, Gemini to the same degree. So I’ve actually enabled ChatGPT whisper. So anything that I talk to on the pc, you know, I use ChatGPT Whisper and you know, that’s that. Me talk to texting has been a, has been real godsend. It allows me to get things out. My brain thinks way faster than my math, uh, talks to my hand rights, you know, so even slowing that down you is a challenge. Uh, but making sure that I’m more concise, more clear, as I talk you allows me to do way, way more than I was able to.
and, just thinking about how, how I type. like we have two, two fingers just about that, that is a limit to the, to the bandwidth that, tinter web can, can help me with. Yeah.
Dan: okay. So, we’ve covered quite a bit. I think it’s probably time now to pull all of this together.
and sort of summarise for either those that are, are entirely fresh to this and not currently engaging. or those that have perhaps tried and had a full start and, and have ma maybe, yeah, may, maybe back to square one. if we can break this down into a very simple prompt, a people prompt. so, I’m, I love the idea of triangulating, different, tools. So if I can pick free, which would I pick?
Roy: At the moment, I’d probably pick, pick Claude ai, is a very good one. Um, the Claude is about to come out with an agent element as well, which will be great. When it gets out. I’d pick Genspark. So Genspark is an agent tool which uses multiple different large language models and things to, to dig into. and then, yeah, over and above that ChatGPT as another option. You know, it allows you to do more at a lower cost, in terms of the free versions, in terms of, you know, what you need to do. Just start using it every day. You use it for your holiday planning. Use it for your meal planning. Use it for your wellness planning. You know, because if you don’t start using it, you’re never gonna get to understand what the limitations are. Limitations. Don’t give. Understand if the right,
Dan: Yeah. really great tip. And, those dumbbells that sit in my garage haven’t helped me grow, my arm muscles. But, it’s not the dumbbells’ fault. it’s this dumbbell’s fault.
Roy: for sure. Right. Sure.
Dan: really interesting. so, yeah, and, been as, as always, it’s fa fascinating and, very educational talking to you.
I, if any of our listeners have enjoyed this and want to wanna talk further, or get to know you a bit better, how best to get in touch.
Roy: I think LinkedIn is probably a key channel for me. So it’s Roy Castleman and I’m on LinkedIn. Otherwise, my, one website is, www.allthepower.co.uk. So have a reach out there. And yeah, let, I’m happy to discuss these AI things with everybody. It is very much, you know, a passion of mine. But then also, you know, we spoke about the strange things I do.
So getting in ice baths and, you know, breath work and other things, you know, all these things. Yeah, I think, you know what I do, I help business owners learn ai but also help them with wellness and then grow and scale their businesses and, you know, it’s a fascinating life. Live the life you love and, you’ll enjoy every day.
Dan: In Indeed And, yeah. Great. a great way to, to approach things. And, yeah, we de we’ll definitely have you back very soon to talk about some of the, some of the other interesting, activities and, and maybe have that, that, that debate about which is best an ice bath or a Matey bath. one will win? that’s a great.
Roy: In Bedford,
Adam: For that.
Dan: I’m, I’m a little bit shy, I think, especially if it’s cold water.
Roy: it’s definitely cold water. Yeah, we’ll keep it at five degrees so it’s not too cold
Dan: Okay.
Roy: cold.
Dan: very good. There we are. So, we’ll leave all of our listeners with that mental picture in their mind that they now cannot unsee. And it’s been a pleasure talking to you. Thanks ever so much.
Adam: Yes,
Roy: Thank you guys.
Adam: thanks.