Leveraging AI
Dive into the world of artificial intelligence with 'Leveraging AI,' a podcast tailored for forward-thinking business professionals. Each episode brings insightful discussions on how AI can ethically transform business practices, offering practical solutions to day-to-day business challenges.
Join our host Isar Meitis (4 time CEO), and expert guests as they turn AI's complexities into actionable insights, and explore its ethical implications in the business world. Whether you are an AI novice or a seasoned professional, 'Leveraging AI' equips you with the knowledge and tools to harness AI's power responsibly and effectively. Tune in weekly for inspiring conversations and real-world applications. Subscribe now and unlock the potential of AI in your business.
Leveraging AI
28 | The AI Playbook: A Proven Framework for Implementing AI to Boost Productivity and Profits with Paul Bratcher, a technology implementation expert with 20 years of experience.
Could AI supercharge your business or destroy it? 🤯
Artificial intelligence is transforming businesses, but the path to success isn’t always clear. In this episode, we demystify AI adoption with business transformation expert Paul Bratcher.
Topics we discussed:
💡Why every business needs an AI strategy ASAP
💡How to get buy-in from stakeholders
💡Finding the right low-risk, high-value AI projects
💡Developing a framework for AI implementation
💡Creating reusable prompts for ongoing productivity
💡The people and culture side of AI adoption
💡Competitive threats from AI laggards
Paul Bratcher is a veteran technology leader who has been implementing AI, automation, and other emerging tech for over 20 years. He's helped transform businesses of all sizes with digital technologies.
Connect with Paul on LinkedIn to continue the conversation!
About Leveraging AI
- The Ultimate AI Course for Business People: https://multiplai.ai/ai-course/
- YouTube Full Episodes: https://www.youtube.com/@Multiplai_AI/
- Connect with Isar Meitis: https://www.linkedin.com/in/isarmeitis/
- Free AI Consultation: https://multiplai.ai/book-a-call/
- Join our Live Sessions, AI Hangouts and newsletter: https://services.multiplai.ai/events
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Hello and welcome to Leveraging ai. This is Isar Metis, your host, and I've got an exciting show for you today. Maybe the question I get asked the most, especially from business leaders, CEOs and other C-Suite and executives, is how to get started. How does a business, not an individual, go into the process of implementing AI in an effective way that really transforms the business? And this is exactly going to be a question we're going to answer in today's show with the assistance of Paul Bratcher, who has been helping companies implement technology for the past 20 years. As always, at the end of the episode, I will review some exciting and big news that happened in the AI world this week. And now let's dive into the framework that can help businesses implement AI successfully. hello and welcome to Leveraging ai, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This Isar Metis, your host, and I've got a special guest for you today, Paul Bratcher has been in the forefront of technology his entire life. He has been implementing digital transformation and emerging technologies. In anything from huge international, large corporations through his own entrepreneurship journey, as well as holding different fractional leadership positions in various size and types of companies. So he has over 20 years of experience in implementing new technologies in businesses as a transformative process. And I hope he's not mad for me dating him by saying it's 20 years, but he has a lot of experience in doing that. So it's not surprising that in this past year, he has been helping companies implement artificial intelligence in businesses because that's the new technology that's around. But for him, it's just. Another one of those technologies that he's helping businesses implement, which makes it very interesting because, he has a unique approach on how to do this right? How can and should a business look at AI from a more holistic approach instead of jumping into the tactics? Instead of that looking at what is the best use of AI for our business? How to figure that out. And this is gonna be the core of our conversation. I think this is incredibly important to any business today because more or less every business today is considering implementing ai and most businesses don't know exactly where to start and Paul is the perfect person to have that conversation with. So I'm very excited to have Paul as a guest. Paul, welcome to Leveraging
Paul Bratcher:ai. Hey, thanks man. So it's great to be here and yeah, I don't mind you dating me at all. I think myself and the other two founders of our company, between as we worked out, we have. Seven$50,000 of experience of implementing transformation in UK retail. That's incredible.
Isar Meitis:We've, that's a
Paul Bratcher:really big number. Yeah. we've done some things, and we've learned some things along the way. So hopefully some of those will, will come out in today's conversation.
Isar Meitis:Awesome. So I really think, let's just dive right in. If I'm a business owner and there's all this AI craze going around and there's a huge FOMO going on of I gotta implement AI because everybody else is doing it, and if not, I'm gonna lose the competitive edge and the, the competition's gonna eat my lunch. What are the first things as a business owner or somebody's leadership I need to consider and look at before I start diving in into
Paul Bratcher:an implementation process? So I think the first thing, and see CEOs ask me this all the time is why and where should I start? That's like the first question and the first kind of little bit of advice is if you are interested in AI and you've been listening to ai, then you'll build this mindset that everybody's doing ai. And actually, when I speak to a wide broad of companies, a few are doing ai, a few are doing it really well. A lot of people are trying to answer the, how do I start question. Yeah. And my first answer that is, you need to start, that,'cause that is step one of the process. You need to start. It doesn't really matter where you start, but you need to start somewhere. and this is just change practice. And the first thing I remind people is we've done change for 20 years from digital transformation, from cloud models, from software as a service. It's the same thing. It's just happens. The tooling this time is badged ai. Yeah. And the things we learned about change were. It's entirely about the people and it's entirely about having a clear vision, and it's entirely about being able to communicate that vision and the tools will take care of themselves. Yeah. So once we all take a breath from that. And then I like to explain about the overlap of changing technology and the next thing I will talk about is if you think about digital transformation for the last five years, what's really new? Like nothing. it's got to the end of, its curve of activity. Yeah. If you think about ai, everything's new every day. There's thousands of new tools, there's thousand things. So you've got these two overlapping technologies and the digital transformation age is basically ending, and we're now actually what we describe as the augmentation age, which is about how do you join. AI tooling to your workforce and generate, a better solution for everyone else.
Isar Meitis:I love what you're saying. I think I want, but okay. So I wanna summarize some of what you said. That it's, it happened before, like nothing is new. It's just another new set of technologies that is coming in that the world will have to adopt to and work with. So that's not new. I think the thing that is new and I, I'm roughly the same date as you, so like same production date as you. So I've been through a lot of it mostly as c e o of tech startups or in leadership positions in tech startups. And I think the biggest difference between all those other ones that happened before is speed. So I'll give a simple example from my side and then I would love your opinion on this. As somebody who's been helping companies through this, I first logged into the internet. In 93 or 94, like my roommate at the time bought a modem. He said, there's this new thing, it's called the internet. And I'm like, okay, what does he do? He said, I don't know, but we're gonna test it out. So he bought a modem, we connected, he went, one of those for a while. And then we got up like a prompt line. Those of you who remember Doss, that's what we had on the screen. Yeah. and I'm like, okay, so what do we do now? And he's I dunno. And that was the most underwhelming technological experience of my life because we connected to the internet, but we didn't really know what to do with it. So that was 94. The first time I got an email address is in 99 when I went backpacking in South America. So that's five years later. So from the moment the internet thing started to the moment I got an email address, forget about all the stuff we do today. Five years went along. And I'm an early adopter tech guy who loves this kind of stuff and geeks around it. What's happening now is within, November of last year ChatGPT launched. And it seems like you're saying that every new day there's new tech to learn. So the speed in which things are happening and the amount of aspects of a business I'm putting aside, our social life and personal life, the amount of aspects in a business it touches is almost everything. And I think that's the biggest difference between this cycle and every other cycle, at least in my
Paul Bratcher:eyes. What are your thoughts on that? I think that's very true. I think, we I helped to define people's thinking in there was, or there wasn't. So you remember what we always often remember when there wast internet and then there was Yeah, my sons sat behind me there. In indeed there was iPhones and then there wasn't, and they're distinct marks in technological time. And I think right now we're on the, there was an AI and now there is, it's a step, almost a binary step from the past to the next. Yeah. And the thing that's really fascinating that from a human change point of view, is the last 20 years have really been targeted at, automating a way, what you would braw describe as non knowledge worker tasks. but the new AI technology strikes straight into things like UX design, business process design, content creation, human relationship processing, CRF. It's deep into, ironically, the people who the last 10 years have been changing everyone else's job. Now it's there to have their job changed. Yeah. and it's quite interesting watching all these like project managers and business and developers, Grasping and struggling with the change impact. So I think it's a time to all, as with all change to be sensitive. So I agree. So to that end, I guess the question that leads onto, is, what is AI good for? Yeah.
Isar Meitis:Yeah. Because you said you, you said your first recommendation get started, but that's okay. How do I get started? What is this thing I get started
Paul Bratcher:with? So we ask companies to think about ai, in a couple of ways. The first is, and I'll come onto to business case. I've, we've got, we have a playbook we use for business cases. and it leans into our kind of ethical sort of social responsibility ethos as what our company's all about. But the first thing we start about is saying, if you start to read the news and you ignore the AI Cult of Doom Madness and click back and get the principle. It's basically you need to think about ai, with three, three in a sort of triangle of three thoughts. And at the moment everyone's talking about thinking AI as a brain, as a thought engine, either large language models or more excitedly protein folding if you're being, reading that kind of scientific research or, had conversation with a PhD in AI in about 92. and I think the first AI computer science paper was like 1957 or something like that. So it's not new, it's just accelerating. Yeah. So there's this talk about this one corner of the triangle, which is all about thinking. We'd like to say that, there's two others. To make a holistic thought about ai, you need to think about sensing or input. And that can either be things like computer vision. And a really good application for computer vision at the moment is things like, mountain to the above. assembly lines. Checking. Assembly. Okay. Yeah. there's, firms I know of that are using AI to record movements of self-checkout, to try and spot people who accidentally fail to scan things and then maybe encourage them to scan them, for example, and that's all computer vision orientated. So this, you think about, you've got input and sensing. How do I gather an understanding of what's going around me? Yeah,
Isar Meitis:it applies. Saw, I saw a very interesting use case, not in vision, but they actually use vision as well. But most of it is sound. They record machines. Yeah. Or they don't record. They listen to machines and compare it to the original recording. Yeah. And they can detect the beginning of a potential fault in specific parts of the machine based on a very detailed understanding of the sounds that it makes. And it helps you fix that component of the machine before the machine breaks and stops your assembly line or your production line or whatever it is that you're doing. So it's, there's really endless number of use cases for the sensing part that you're talking about, because it allows you to connect things in the physical real world into this machine learning world where you can then make assumptions and understand things and
Paul Bratcher:define actions. So then you've so got sensing, thinking. And then the last one is to take action. now in ChatGPT action is done by the human being and it's done using the most powerful part of ChatGPT, which is copy paste. it's the number one interface for ChatGPT copying, paste out vice versa. Yeah. But in the rest of the world, there's quite a rich world of, process automation, physical automation. In fact, Microsoft's Inspire launches last week, they've used a whole bunch of stuff around deep process mining using, AI to develop and define new business processes. So there's a whole bunch of interesting AI stuff in that space. And then of course you get into, robotics and drones and all of that kind of stuff. So when people start to talk to us about ai, our first point is, let's pick a place that would allow you to get innovation to generate value that's going to be additive to your business and is low risk to your existing business. Because if you do an AI project and it fails, as with every change project, every naysayer in the world will say, I told you it was rubbish. That's the lesson we've learned from 20 years of doing digital change, right? You want to get some wins on under your belt. And just to put all that into concept as to how interesting alternative use of AI could be. Ikea, the, furniture manufacturer having have released and designed a set of drones, which they release into their warehouses overnight. They map the warehouse, they take off, and they do the stock. They do the stock count. Interesting. Now, interesting enough, people don't like counting stock, and we're really bad at it'cause, we just get confused and it's a job no one wants to do. So that, for me is a really good example of a use case for a business which is AI driven, but actually it's not leapt into ChatGPT, it's trying to solve a problem that needed solving.
Isar Meitis:Okay, so let me do a quick summary and then maybe we can dive into either additional use cases or how do I know which of the three to start with?'cause you mentioned a few very important things. So first of all, you said you want the first thing you pick your, let's call it your low hanging fruit. Something will be additive that will actually add value to the business. And that there's always two sides to that coin, right? It can solve a problem or it can increase a potential. In both cases, it's additive to the business. It needs to be low risk because you don't wanna bet the farm on something that is a new technology and you not sure how to implement it. And you want to get quick wins. So you don't want to create a three-year project. You want to create a two week month, two months project that can show quick benefits and show a win because then A, you get the buy-in for anybody we need to get the buy-in, whether it's the board, the investors, employees, like whoever needs to be in, and b, It can by itself, then finance the next step because the company's now making more money and you can do more stuff. So I love all these points. I think they're incredibly valuable. How do I identify, so in, in my eyes, what you just said as a, me as a business owner, you just confused me a little more because so far I was thinking ChatGPT and I didn't know where to start. Now I got ChatGPT and I've got sensing, and I've got actions that I can take across different aspects. How do I know to pick that stuff that is additive, low risk, quick win that I
Paul Bratcher:need to start with? the first part of that is just to start with, get some good advice. Okay. Because if you do them, if you do the maths in the uk, in the 2021 centers, there were 109,000 people recorded their job as systems architect, solution designer, or business analyst. Okay? Okay. There's 5.5 million businesses in the uk, 1.8 million of which have more than a hundred thousand, more than a thousand employees. Even if there were a hundred thousand AI people, the chance of you recruiting or finding one is about 35 minutes is your share of the resource pool. Okay. they're just rare people. So the first thing you need to do is a strategy to either get a good person to help you or a strategy to grow some people Yeah. In your own organization that learn these tools. Okay. Which is another reason why you need safe places for'em to learn. Yeah.'cause you don't want them to become, unsuccessful. So let's just drift into trying to find candidates of what AI is good at and what people are good at. Awesome. AI is good at a bunch of things, but what it's really good at, if you think about was language models as a use case is dull things extraction and summary, transformation, translation, interpretation, sentiment setting, prediction, stuff that is just really basically what you would call, the utility room of computing. It's the dishwasher of your house. It's not exciting stuff. Yeah. No one cares if the computer does it. The thing that ChatGPT brings in, a lot of times it wasn't plug in. they radically lower the cost of doing those tasks in an automated way. Yeah. If you wanted to automate your email workflow before you had to do data transformation, you had to do this, you had to do that, you had to do process. I was all really quite hard. The language tools are really good at that general purpose language management, which historically is systems have been poor at, which is why those probably have been left alone. So we like to say AI is good at joining dots. As a simple metaphor is good at joining dots. So the opposite of that is, what are people good at? people are good at relationships. They're good at dealing with you, unexpected. They're good at dealing with your no, they're good at imagining a new way of doing things. So they're good at making new dots. So you've got these almost beautiful sort of symbiosis where something that lights, life is the wrong word. Something that cares not how many times he does a mundane job. Dos to dots is a thing. You could have people who like doing new interesting work, but not doing the mundane now have a way to be creative and be expressive. So it's about bringing these two things together to make what we describe an augmented AI experience where you are using these AI tools to essentially give your everyday work superpowers to your people. So you're going from trying to, Thinking about how can I get a productivity saving or how can I get, a reduction in effort. So actually how can I get to a significant value gain for everybody? and this is where we now stray into the how do you do a business case, the C F O, and we have a fairly simple pitch process for that.
Isar Meitis:So before we dive into this, I want to add one thing. I just came back from a great conference. It's called, it's the Marketing AI conference, and it was in Cleveland, Ohio. And one of the speakers was the Chief Decision Officer at Google. She's incredibly smart and yet has this amazing skill to take really complex stuff and turn it into words that normal people like you and me can understand. Yeah. And she was talking about the whole idea behind AI and where's it going, is it differentiates between thinking and thunking, which thunking is a word she made up. Yeah. but thinking. Is what humans are very good at. Yes. Is taking something that did not exist before and thinking about it and making this new idea out of it. Thunking is the stuff that you do once you've finished the thinking in order to make the thing actually happen. Yeah. Which a machine does a lot better and it does it a lot better because it has access to a lot more data. It has a lot more memory than you and I have, and hence it's very good at these repetitive data analytics kind of things. And so if you cut the line between thinking and thunking, which is basically what you're saying. Yeah. and I think the biggest difference, that's not a new concept. I think the new concept is that we've pushed the line of thunking. Way further ahead than it was before because there were tasks, not that they were thinking tasks, it's just tasks that computers did not know how to do easily until a year ago. And even that's not totally true. It was available to Netflix and Google and Facebook and these companies just wasn't available to us, the common people in an amount of money that we could afford. Correct. And now it costs you either zero or 20 bucks a month and you get access to an A P I that can do things that just were not doable earlier. So I love the way you framed it and now let's dive into your framework on how to pick
Paul Bratcher:the business cases. So we always start quite after that conversation starts with the C F O. Okay. Always someone says, how do I persuade my C F O? I'm like, okay. I've been in business 20 years and I've said in board meeting, after board meeting where we've done the strategy in the financial review for the next year. And it always starts with something like this. The CEO says to the. Chief commercial officer, I need you to save half a percent margin. And then they'll say to the HR director, I need you to save a quarter percent margin. And it'll go around and you'll think, if we can get another 1.5 or 2% margin in this company next year, that would be great for us. It's that straight bottom line profit. So it's all about margin. Okay. So I say, what if we think about your human resource spend, which is typically 70% of your cost, excluding stock, excluding property. So it's of your manageable cost, it's 70%. What if you could make those all 10% better? So that gives you a net 7% gain in margin. Would you like that? I haven't yet found a CFO who doesn't want 7% margin gain. Okay. Yeah. So I said, okay, let's use AI to get 7% there. And they're like, okay, that's a good idea. okay, that's quite nice. The problem with that is there's no one to do AI in your organization. You can't do your plan. What if there were some projects that were, on the to-do list, but had always been put off because they were a bit not a good return on investment. all those things like, social responsibility, wellbeing, all of those soft processy type things that, that, just never happened. But actually your shareholders are saying, what is your e s G agenda? What is your social responsibility? what is your what if we used some of that 7%, saving to do those products? You say, that would be good. actually I've got billion news for you. Actually, it's not 7%, it's probably nearly 20%. So you could have your 7% for free. So for your shareholders, they're happy. You are happy, you could take the next 10% and invest that into doing these long-term shareholder projects. Yeah, you could use AI to drive some of those out. You can use'em as train you around and that generates you. Set of people who are skilled in being able to do even more of this. And then I'll usually look at the HR director and say, just remind me again, motivated, well-trained staff, add what Productivity to companies. I think you'll find the numbers 60%. So you're gonna do these two things to get 20% unlocked from your business and you're gonna end up with a set of people that have just been doing projects that they like doing that were interesting, that retrained them to unlock a further 60%. At the end of the process, you're already at a hundred percent target. Well, 80% target, and actually 20% is a low bar number. If you look at the stats from the employment plans for the US for the impact of ai. Same in the uk, same from McKinsey. They're talking about 45 to 47%. Yep. Revenue gain. So if you want to unlock and have a change program that's gonna work. You've got ask the question, what do you wanna do in the business? Generate value. How are you gonna do it? We're gonna be a bunch of projects that are safe to learn on, that are low risk, to retrain our staff.'cause there's always gonna be a staff shortage. We're gonna end up with motivated staff. So you're looking to generate how we phrase it. Wouldn't it be good if they had generated value for everybody? For good? Why wouldn't you do that? And usually this by now, the C F O and the board are like, it seems like a reasonable plan. Where do we really start Now we're bought into it. we're gonna start, we need to start, we're bought into the idea of the exact where? Yeah. And then the exact where, we try and target a combination of impactful enough, safe enough and enough risk that there's an impediment to do it. So it's gotta solve a problem. It's gotta solve on it, but not crazy risk. Yeah. Like for example, if you're clinical practitioners, would you immediately let AI do, or your diagnosis No. that's just a, that's just a bad idea. However, would you let it help Feline the 60% of administrative time that once you've seen a patient that the clinical practitioner can't do because he's got 35 forms fully? that to me seems like a safer place to go than the other. And the same if you're a sales oriented business with a call center. If you've got five star customer services call center, leave it the heck alone. If you've got one star customer service call center, then maybe it is a good idea because it's not gonna make any worse. Yeah. So we try and sit down with the board and find these, itches, these challenges and say, okay. That feels like something you feel you want to do, something you feel will get value and something that is within your board's level of risk appetite. And every company, every board's risk appetite is a little bit different. But broadly speaking, the companies that are implementing AI now want to be the leaders as opposed to lag guards. So they tend to have a little bit more, appetite for risk, at a board level than the others. And also my experience is they tend to have a little bit more appetite for risk than the CIO or the IT department or the business themselves actually realizes if you can express it and control it in a sensible way.
Isar Meitis:I love this. I want to touch on two really interesting and important points that you touched. One is, at the end of the day, decisions in a company are based on the financial outcome. Absolutely meaning. Yes, it's very important to get the buy-in from the employees and train them and get them motivated and whatever, but at the end of the day, the CFO slash CEO will make the decision based on, is this gonna make us more money or less money to ourselves and to the shareholders? And so the framework you shared is brilliant because I've never heard anybody talk about this. People talking about how do we get employees trained and how do we do this, but not about, okay, we can do all of that, but nobody will ever do this if it doesn't really make sense from a financial perspective, because that's what a company does. Yeah. The second thing that I really is the idea behind the ethical aspect of this. and to be fair, I would assume, and I've seen this, so it's not just an assumption, let's call it an educated assumption, that this will cost some jobs. Meaning if you're saying, okay, I have a. Ex-employees, 30 employees that are doing this task and now I can do this task 30% more efficient. you have two options. One option is to say, I'm gonna let go of 30% of the employees and get immediate savings to the bottom line. That's the less ethical and long-term, I think, less profitable option. The other option is to look at other parts of the organization, which are connected to that 30% and saying, can we push them 30%, that now we can sell 30% more or 50% more because of these capabilities with the same resources that we have today. And now everybody wins because now you have, like you said, happier employees. They're doing stuff they care about. It's more ethical because you didn't just let people go and the company grows, you're making more money, you're growing market share and so on. So I think I. The truth in most companies will depend, a, on the economical situation they're in, and B, somewhere in the middle between these two things. I think it's gonna be very naive to think that people are not gonna be let go because of this new capabilities. But I think the right way to look at this from a leadership perspective is can we push the boundaries across multiple aspects of the business? So instead of letting people go, we grow the business even more and use the same
Paul Bratcher:resources we have today. I think when we have this conversation, and it obviously occurs, is the first thing is you need to consider the fixed labor follow-up. The fixed labor fallacy. Okay. Okay. And a fixed labor fallacy is, there's a finite amount of work to do. Okay. If you go to any business and say, show me your business strategy, how many projects of that business strategy on the are only gonna be done this year, and what's your backlog in number of years of projects do you have, if you go to a major I did some work in a fus hundred, retailer in the uk and when I joined they had a backlog of projects, assuming the same IT resource going out for 11 years. Okay? There was no shortage work. What there was, it was a cost barrier that made the projects too expensive to do. Today, what these new tools do is they lower the cost barrier. So actually those projects that would've been efficiency gain, it would be nice to have after just become accessible. And, I think McKinsey was saying that there's a growing width between a leader and a laggard, and it's about 35% market growth. So if you're a leader, you are likely to be gaining, you're likely to be gaining more market than a laggard. And what happens is that gap doesn't narrow over time. The laggards never catch up. They actually get further behind. Until, and there's this, there's a very sad situation happening in the UK where a large retailer is in the process of going bankrupt and they're going bankrupt because 10 years ago when they had the opportunity to pursue a digital transformation strategy, they chose not to. Yeah. Okay. There will be companies right now choosing to not pursue an AI strategy or to use AI purely for cost reduction. In 10 years time, we will be remembering who they were. We won't be visiting who they are.
Isar Meitis:I agree with you a hundred percent only. I think it's in three years time, but that's a whole
Paul Bratcher:different story. I only have data to basic it. yeah. I, which is currently, it says 10 years to go on business.
Isar Meitis:I just think things are moving so much faster right now that your ability to win or lose market share, Because of these new capabilities are very different than everything we've seen before. Nothing else. And so the ti the timelines of everything is just gonna be much shorter. And I don't know if it's three years or four years or five, but I don't think 10 is the number, even though if it was in the past. I wanna switch gears because there was something we talked about in our early call that I'm really interested in. And you said something that I personally agree with a hundred percent. And I think it would be very interesting to share. You said that prompt engineering is not a thing. And I agree with you. It's prompt engineering is a UX problem. It was created because these tool was released. When nobody was ready for it. And so people had to figure out how to engineer a prompt. I even today, there are tools out there that will do it for you. So I think that's not a thing that anybody should pay attention to. I think it's a skill that's easy to learn, and I think it's gonna be a skill that's even not gonna be more or less required in the very near future because the models will do it for you. So what you do need to understand is what needs to be in the prompt, not from an engineering way, but from a what is the results that I'm trying to get? And I know you have a framework for that, so I'm very curious to hear what
Paul Bratcher:it is. Okay. let's leap into, we've chosen you a chat based implementation of something. Yeah. So the game of the game, the idea behind generating a large language model based ChatGPT or Claude or whoever, is you want to unlock productivity. Okay. that's the key to the thing you want to unlock Productivity. Productivity is un is unlocked in two ways. It makes a thing quicker. That's nice. It makes a thing quicker, repeatably, so you don't need to invent the way to make it quicker. That's really nice. Okay, prompt engineering is a, as you've said, it's a UX problem. It's a feature of now. Yeah. There's an interesting quote, and I wish I could remember who said it first, but we use it all the time and it's that every problem can be solved, providing you can create a good question. Okay. And a prompt is basically a good question if you get it right. And we use, three, the three Cs to describe how to put together a good prompt. And the first one is the C of curiosity, which is you spend a little bit of time. Toying with the opening sentence, the first question, and we encourage people to try open questions and closed questions. Open questions tend to generate ideas and concepts. Closed questions tend to derive answers, but swapping between the two with ChatGPT can often trigger you to get to a phrase that gives you exactly what you're looking for. Now, the problem with asking ChatGPT any question is, we all know 18% of the time, it just makes it up, it just hallucinates it, right? So the next C is critical thinking. Okay? So you've got to, when you get a response from you, you need to ask yourself a bunch of questions. My first favorite question is, did it actually answer the question I just gave it, or did it answer something similar? Or did the answer it really earnestly and really well. But actually having read the answer, I realized my question was a little bit less good. Okay. So a good example of that we use is imagine you're planning a weekend in London and you just say to ChatGPT, tell me some things to do in London. It's gonna give you a massive list of things to do. If you say to me, tell me some things to do in London on Thursday near Covent Garden. I don't wanna walk around. I like art museums. I like coffee shops. I don't wanna go to the cinema. I don't wanna go to modern swords. You've now created a much better prompt. Yeah, you've asked a much better question. You're gonna get a much better, much more concise answer. But of course, you've gotta be mindful that the data in the model ended at a point in time. So it might just have recommended you to go to a really great coffee shop that's closed. Because since September, 2021 and today it's closed, or it's closed on a Thursday. So there's a really good tool I like to recommend people to use in conjunction with ChatGPT, we call it Google. You might have heard of it. Okay. And that comes down to we cash, we steal a phrase from the cryptocurrency world, which is you need to do your own research. So when you've got the answer, you're finding you've got this critical question, you've got a good question, it's gonna be a good answer. You just need to look at it and say, do I believe it's truthful? Do I believe it's fetched, accurate? and there's one trick I often recommend people use to help with that process in complex prompts. And this were, a good example of this is if you're, if you are, imagine you want to create five recipes. I said, I'd like you to invite me five recipes, one for each evening meal next week. I want them to take 30 minutes to prepare. I'm allergic to salmon. Rather than just pressing go. If you then say, give me the recipe titles and then ask before you proceed, it will just give you the titles and then you can immediately check from a small amount of human input i e reading, whether it's called, salmon with raki sauce. So you know, if you get that, you've got a bad response. Yeah. So if you can engineer a two step prompt, you can use that first summary step to catch hallucination or mis facts, false, fake news, maybe, as a catchall. So we recommend that, be curious, build a good question, refine your question, and then be critical of the answer. that's the first of the two Cs. Then the second C is consolidation. So I'm gonna explain to you my typical journey of how a prompt building goes. I'm gonna see from your face whether this resonates with you. So I start off at the beginning and I've got an idea of what I wanna do. I type some stuff in. It, gets some stuff, types some more in, and then they get that feeling of I'm getting somewhere, but I'm feeling a bit lost. I'm not entirely sure how I got here now. But regardless, you carry on and then you get to the, okay, I'm here now, but I genuinely am not sure how I got here. I'm, I've actually lost. I've got the answer, but I dunno how I got it. When you get that feeling inside you, that's the time to stop. Okay? Do not prompt on, don't go down what we call the prompt hole. call it there. Scroll back up the page and think. How can I consolidate this process into a fresh starting prompt? Okay. And that's really important because, you may be carrying, bias or context from your false answers from your journey into your prompt this moment in time. and the longer you speak with ChatGPT the longer will our context gets. The longer the context window is extended, the more chance you have of hallucinations or false data. So when you get that slightly uncomfortable feeling of I'm getting your Billy smile, that's when you should consolidate down, rebuild the prompt back to one cohesive prompt, and see if you can restart it. Perhaps in a new chat window, I can get back to exactly where you were. If you've got a prompt that does that, then you, you've achieved the ultimate goal, right? You've got a prompt that's done a piece of work for you. The only thing you need to do now to achieve full productivity is write it down. Not ChatGPT, build a playbook, consolidate it into like I use Notion. I've got a friend at work who uses Word and we literally, I literally have a page that says, this prompt does this, here's the prompt. And then next time around when I wanna do the same things I've done before, whether it's a market assessment or summarization or I want to check, see whether the article I've just written is in my tone of voice. I've got those prompts re canned, I just copy past and I've not had to do that whole 45 minutes of exploration. I've just gone to instant productivity.
Isar Meitis:So I love everything you're saying. I wanna pause you to add some tactical, practical tidbits that I use myself and that I teach in my courses and give to my consulting clients. So the first thing, there's a tool called perplexity.ai, which is basically ChatGPT connected to the internet, which solve some of the problems that you talked about in the beginning. Go check on Google if it exists. The benefit with perplexity is that it actually shows you the links of where it got the data from. So if it gives you the link to that coffee shop, you can click on it, it will take you to the coffee shop's website. You're like, oh yes, this really exists. They actually are open on Thursday or not. So that's small tip number one. small tip number two, as far as potentially shortening the process or giving it less context when it's not relevant. When you go start a chat with ChatGPT and you keep on going and going the first time, it gives you a bad answer. You have two options. And by the way, most people don't know these two options. Most people just keep going and ask the follow up question or try to correct ChatGPT. But next to each prompt, there's a little. Pencil kind of button. Yep. That you can press and edit the prompt. And the difference between editing the prompt and re-asking the question is exactly what you said. If you edit the prompt chatGPT will not remember what was in the original prompt. Yes. So if you give you an answer that is not aligned with what you were asking for, you want to edit the prompt rather than ask a follow-up clarification question. And the reason for that is ChatGPT will remember the wrong answer that he gave you earlier, and he will use it as part of its memory when it gives you the future answers within the same chat. So what are the steps that you can do instead of just starting a new chat? Is just edit the prompt that gave you as soon as it happens, this is not exactly what I wanted. Instead of asking a clarification question, go and edit that prompt that you can ask something completely different or just slightly different until you get what you wanted. And only then move on. So that's tip number two. tip number three is with regards to a prompt database. What you're saying is critical. Critical, and it's as critical to make it available to the rest of the company. Meaning use a tool that everybody has access to, define some rules on taxonomy and how we write these things and how we define what they do, and maybe tagging, mechanism that people know what to look for and how to look for so we don't keep on reinventing the wheel across different people in the organization. And the last thing that I use a lot, and I've used it way before, ChatGPT, but now with ChatGPT and Bard and Claude, et cetera. I use it even more. there's a browser plugin that's called Magical. Yeah. What magical enables you to do is take a long piece of text, paste it in there, and give it a shortcut. And the shortcut could be any combination of words of characters, right? So it could be an actual word, which doesn't make sense, and you'll understand why in a minute. But once you put in that shortcut, it will put in the full, long text that you put, that you have pasted into the tool, which means if there's a prompt to use regularly or a segment of a prompt to use regularly, such as to prime it as you are an expert on this and a topic you have done this, and that. Here are three examples on how to, don't put it in a Word document somewhere, because then you gotta open and find the word document and copy and paste. If you put it in magical and you give it a shortcut saying prime one, when you know what that means, you type prime one and all this tech shows up every time. So that's again, just another very simple tactical way to implement some of the things, that Paul is talking about, but all are incredibly valuable points.
Paul Bratcher:I have kind of like a tip on the tip. Okay, awesome. when you write your prompt page to share, if just in the top of it you just stream of conscious what you used it for, then you should do what we did, which is we then wrote a prompt, which looks at all of the prompts and extracts into C S V tags metadata. So you can then search on metadata for what the prompts are about really. So we have a search for like research prompts, language prompts, that kind of stuff. Awesome. and if you use Notion, you can actually set that as an AI automatically generated field. So whenever you add the page, it will just, magically create the tags as a CS V. So that's how we, re-index ours. We let the AI do the dual indexing work.
Isar Meitis:That's awesome. That's a great tip. Okay, so now we're missing only the
Paul Bratcher:third C. Third C, which is, consolidate, oh, that was the third C. That was it. Just get it into a playbook. use, copy, paste. I mean there is a, a bit of a moving window on integrations for, chat G P T and we say there's three ways of thinking about ChatGPT and apps and plugins. So the first is you've got ChatGPT straight, which is you're using ChatGPT. You then have ChatGPT with plugins, which is where you're trying to put your app into ChatGPT. and at the moment the apps plugins in ChatGPT is a bit like, remember when the app store first came out and there was like 10,000 apps that just made fart noises? at the moment, the plugin store's a bit the same, right? It's 10,000 apps of which five do really cool things and 9,000 do stuff, which has a use case of the developer who wrote it. It's, it has no value to any, anyone. like Wolf from Offer is a classic go to link reader, ask my p d F, the ones at the top of the list when you do most instore. There's the ones to play with. And then the third thing, which you're seeing more and more of is where people are putting large language models in their app directly. Yeah. So that's where you're getting, and I think you'll see a lot more of that. And particularly in enterprise with low code, what you'll find is that essentially put a simple form in front of a large language model as part of the process to measure. I think you'll see more and more of that as the enterprise sort of solutions, mature and low code solutions automation fits in behind. So I think you'll see this move away from plugins I think they're a short term solution, and I think you'll end it with essentially, using the tool directly itself, or it'll be fronted by some form of data formatting tool to make it a more repeatable process. So I think that's how it will consolidate from a tool set point of view. I
Isar Meitis:completely agree, Paul. This was really fascinating, extremely helpful, I think to anybody who is in a leadership position and thinking how to get started. I thank you so much for taking the time and sharing this information.
Paul Bratcher:I am, I listen to your podcast most commutes when I'm going to from the office. I even, with all my experience in doing things, I still pick up tips on the way I think, oh, I really need to have a go at that, which is why I'm, deep into N eight N at the moment, having a play with that. I heard that on one of
Isar Meitis:your podcasts. So one of my favorite toys right now. That's a whole different podcast itself. Paul, thank you
Paul Bratcher:so much. No, thank you. Have a great day. Take care.
Isar Meitis:Great conversation with Paul. The silver lining through everything we talked about, it's something I highlight frequently to the customers I am consulting to is that implementing AI in a business is first and foremost a strategic business process, and only then an AI implementation process. So it's not different than doing any other big changes in your company? It requires figuring out strategy and processes and risk, et cetera, and only then the AI part of this, but it's definitely worth doing because the positive impact is very dramatic. Now to some exciting news from this past week. The first big news I wanna share with you is that OpenAI, the company behind ChatGPT is on path to making over$1 billion in revenue in the next 12 months. Now, this is significant because of several different things. First of all, because they projected they would make around$200 million this year, and now they're on path to making 1 billion. that means that they grew dramatically faster than they themselves anticipated. The second thing is, all of last year, they made$28 million in revenue and now they're making$80 million a month. This revenue comes from two main sources. One is people who are paying for the premium ChatGPT service. And the other payment comes from people who are using the A P I. So every a P I call costs you a fragment of a dollar, and sometimes fragments of cents depending on how big the API call is. But these obviously accumulate over time through the usage of multiple people and companies. What does that tell me? That tells me that despite the fact that there's some people that are saying the hype was overrated and that there's been a decline in traffic to their website, which is true, it's still in extremely high demand. I really can't think of many companies in history that in less than a year of releasing their product and much less than a year releasing a paid product got to a pace of$80 million in revenue per month. So that tells you that there's something real behind it and that there's real need and real usage and real value because people are willing to pay for it. Speaking of large language models, Baidu, the Chinese tech giant, just released Ernie Chatbot, which is based on a large language model also called Ernie that they have developed. I. It is the first of its kind that's being released in China. As you expect in China. It. First had to go through the scrutiny of the Chinese government, and first of all, had to show compliance with China's generative AI guidelines, which they released a few months ago, which we shared on this podcast. But in addition, as you would expect from a Chinese chatbot, its views are limited to the views that are allowed in China. And as an interesting example, if you ask it, where was c COVID-19 originated. It will tell you that it originated amongst American vape users in July of 2019 and later only made its way to Wuhan China through American imported lobsters. So whether you want to believe our side of the truth or their side of the truth is up to you, but it tells you that there's top down guidance to what is the reality and the truth that this chatbot can share, and it has to be aligned with what the Chinese government will allow. I'm totally not surprised by this. This was expected. The interesting part about this, it's the first time that a large language model had to comply with government regulations before it got released. And yes, it's China, so they have better chances of doing this. But I would really like to see that kind of approach to releasing large language models all over the world where there's clear guidelines that you have to comply with before you can make the large language model available to the public. The next two pieces of news I wanna share with you this week has to do more with the graphic side of things. The more important one is that Google just released a new digital watermark. It's called Synth-ID that works with its AI suite of tools. It's a development by Google's DeepMind team, and the idea is that it's a watermark that cannot be removed, that is not visible in the actual image itself, but is easily detectable by a detection tool. And the idea behind it is to eliminate the opportunity for DeepFakes, meaning it will be easy for anyone to know whether a picture is a real image or was it actually generated by ai. This is obviously an extremely important move because from my personal perspective, the most immediate big concern about AI is the ability to create DeepFakes, meaning the ability to manipulate the truth in the digital communication across everything that we do on the day-to-day, whether it's personal communication or business communication or general news. There is really no way today to tell the truth Between AI generated content and real content, unless such tools are being widely implemented. So I that this move by Google will be adopted either with their technology or any other technology across the board as something the government will impose, which will allow us humans to know what is real and what is not. And speaking of deep fake images, the ability to generate selfies of yourself that are not real is becoming more and more popular Across multiple platforms and apps like Lenza as an example. They all work kind of the same. You upload multiple images of yourself and then you can generate really cool, sophisticated images of yourself in different outfits and different setups of your hair and different environments and so on. Now, this is really cool and useful and the piece of news from this week is that Snapchat just created a feature like that built into Snapchat itself, which means it's gonna be used a lot more by people because of all the existing Snapchat users. Snapchat calls this new feature dreams, the interesting thing is that it's a paid feature, and you can buy packs of these cool AI generated selfies. If you enjoyed this episode or the podcast overall, please recommend it to people that you know that can benefit from it. Share it either directly with'em on dms or on social media, and please rate and rank the podcast on the platform you're listening to that helps us reach more people. which means you are helping more people learn from the podcast as well. And I really appreciate if you would do that. And until next time. Play with ai. Try different things. Don't be shy. Reach out to me on LinkedIn. Let me know what you think of the podcast, if you have any ideas or topics you want me to discuss. And until next time, have an amazing week.