56 | Transform your business with Custom GPTs ! A Step by Step Guide with Matt Paige of Hatchworks

0:00Hello and welcome to a live episode of Leveraging AI, the podcast that shares practical ethical ways to leverage AI, to improve efficiency, grow your business, grow your business, and advance your career. This is Isar Maitis, your host. And yesterday may have been one of the most critical events in AI adoptions ever. And if you don't know what I'm talking about, I'm going to explain. OpenAI, the company that just over a year ago gave us in the world, ChatGPT, which changed everything, just launched a GPT store yesterday. day. So some of you, I'm sure, are asking, what the hell is a GPT and why is that so exciting?

0:40A GPT is like a mini version of ChatGPT that already comes with detailed instructions and data focused on completing a very specific task. And that task can be anything you set it up to do. So it could be something in your personal life, something in your business, things like respond to a customer service query or creating a question for a specific job interview or analyzing reports from a specific kind or any something that is very specific that requires specific instructions and data can be packaged as this thing that OpenAI calls ChatGPT. Why is that exciting? And exciting, first of all, because it's making it very useful to people who don't know how to use ChatGPT and how to prompt and how to use the whole process because it's geared to do this one little thing with very simple instructions.

1:30But it's also really exciting because this may be the App Store moment of AI. If you think about everything we know about mobile phones today, it exploded when the App Store became available, meaning it suddenly had very functional capabilities, the apps that we all use every single day across everything we do, that made mobile phones a lot more usable for a lot more people, for a lot more use cases. and the GPT store may be exactly that for generative AI, which means it's really exciting times that we're in day one of the app store. And think about what you would do if you knew what the app store is going to be 10 years later.

2:16So I'm telling you now, this is where we are. Now, our guest today, Matt Page, is the VP of strategy and marketing at Hatchworks, which is a software company. But in addition, he's an AI coach and his biggest area of expertise is developing these GPTs. And he's been creating them for his business since they became available a few months ago. So he's an absolute expert on the topic and hence he's the perfect person to have this conversation with. And in today's conversation, we are going to review in detail, step-by-step exactly how you can develop your own custom GPTs for any need you have, whether for personal life or for your business, and potentially monetize it in the future through the GPT store.

3:04So I am really excited to welcome Matt to the show. Matt, welcome to Leveraging AI. In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you. I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

3:54Thanks, Ester. So excited to be on. And we really couldn't have picked a better timing for this. We were chatting about when we should do it. And we're like, I think it was that day they had announced the GPT store. And we were saying, let's do it Thursday. It may be live when it comes out. And it came out, what was it, yesterday? It's been a warring since. It happened, but so excited to dig into this and the potential here. And just for background, like you mentioned, I had a marketing strategy at Hatchworks, but early in my career, I worked as a research analyst and I worked with data scientists and it blew my mind.

4:30First of all, how smart they were, but working with all of these models and all this type of stuff, it was so complex and so difficult. And my main role was making sense of it and telling the story with the data, right? But it's become completely democratized over the past couple years with everything going on with generative AI. You no longer have to be a developer, a data scientist, all of these things to start taking advantage of it. So super excited to be on and to go deeper into custom GPTs. And you mentioned like an analogy earlier, there's two ways I like to think about custom GPTs. One is if y 'all remember the movie, The Matrix, and Neo's taken, I think it's the blue pill, blue or red.

5:15Somebody correct me if I'm wrong, but he's in there and Hank, one of the guys that's there, he plugs this thing into the back of his head and he spends 10 hours just learning all this stuff. And that's where you get the iconic line, I know Kung Fu. Like that is what a custom GPT is. You get to optimize this thing. You get to give it its own brain, its own knowledge base, the custom instructions, like you mentioned for very specific use cases. It's like a hammer versus a specialized. I'm just really pumped to talk about these a little more today. I don't know if there's a specific area you want to get into, E-Serve, that triggers any other points on your side, but really excited to talk about this today.

5:59Yeah. So first of all, before we jump in, we are going to be very tactical and really explain how to do this and share it with you. But Matt, let's really get started with how does it even look like? Where does it live? If I want to build the GPT, where do I go? Like, how do I get started? Yeah. And feel free to share your screen. For those of you listening at the podcast afterwards, we'll explain everything that we're seeing on the screen. It will be available on our YouTube channel as well, but we will be sharing a screen for the people who are here live with us. Yeah, perfect. Can you see my screen?

6:36Yes. It's showing up. Okay, perfect. Yeah. So I think this is a key element. It's like, how do you even get started? You hear about custom GPTs and that may trigger for some folks. Is this like a completely different tool? It's actually part of chat GPT. It's very simple to get to it. The one key thing, I believe you do need the plus plan. So it's 20 bucks a month. I've had it for so long. I couldn't remember if that was in the plan or not, but you do need the plus plan. Best 20 bucks you'll ever spend. It'll pay for itself a thousand times over. So all I got to say is just do it. I'll second that and I'll let you continue.

7:16And all you do over here is go to explore GPTs, right? So this is the new GPT store interface. And we can chat a little bit about this too, because this is brand new since yesterday. So there's some parts that are really cool. There's some parts that I think are a bit underwhelming. The biggest, I think, piece missing is the monetization element. Correct. So they launched the GPT store, but what they didn't do is they did not launch the monetization piece. And if you look at their blog, go to OpenAI blog, check it out. There's a small blurb and actually have it right here. So why don't we just look at it?

7:51So they're introducing the GPT store. You scroll on down. Here it is. Builders can earn based on GPT usage. So in Q1, we will launch a GPT builder revenue program as a first step. But U.S. builders will be paid based on user engagement with their GPTs. We'll provide details on the criteria for payment as we get closer. So a couple of things like bummer you can't monetize yet. And also a bummer that it's only launching initially with U.S. folks. And of course, we'll expand that globally, but starting with the U.S. The most interesting thing for me, though, is they're paying out on user engagement.

8:29And I think this was a big open question. Was it going to be you mentioned the app store? Was it going to be like the app store where builders can set their own price? How does it work? What cut does OpenAI take? But they're actually going to be taking a user engagement model. And something that seems similar was the old TikTok, I think it was creator fund is what it was called. So based on the user engagement, the use of your GPTs, that's how you're going to be paid out. And I think that's a very short-term thing, right? I think they don't exactly know how it will evolve. If we look at probably the most successful model of this in history is YouTube.

9:08If you're a YouTube creator and you can drive engagement, you can make shit loads of money. And yes, they sell ads. So it's a different business model than ChatGPT. But none of us knows where this will go six months from now, not to mention two years from now. And so I definitely think there will be a monetization mechanism. I really hope they will figure out a way to make it significant because that will really drive people to invest in developing better and better GPTs, just like we know from the App Store or just like we know from the content on the bigger YouTube creators. But with that, let's really go back to ChatGPT itself and look at, so if I want to get started, let's first of all explain, show where it lives, but then explain what are the components you need?

9:55What are the building blocks to build one of those GPTs? and examples of what they can do. Yeah, definitely. And just for reference on the GPT interface, the store, there's different pieces you can scroll through to see other ones. Great way to get started. See how other people are building GPTs. But if you want to create your own, right up here on the top, you have an option of create. So very simply hit that and it takes you to the GPT, custom GPT interface. And I'm going to walk through the couple of different elements here. But I think one key thing that's so cool about custom GPTs. If you think of back in the day, Excel, you were at a blank canvas.

10:37You can do tons of different things here, but what do you do? It's the same thing with chat GPT. Chat GPT is amazing. I can do a million different things, but I'm with this like blank canvas of not knowing where to start. Custom GPTs are like templates to help people get started for specific use cases. So the first point to call out at the top here, there's two different modes. There is configure mode, which this gives you full autonomy and control over your custom GPT, and you can completely configure it. And we'll jump into an actual example here in a second that I like to use to go through the steps.

11:14The other mode is create. And this is exactly like interacting with chat GPT. So it starts off by saying, Hey, I'll help you build a new GPT. You can say something like make a creative who helps generate visuals for new products or make a software engineer who helps format my code. And you literally just interact with the GPT and it starts to configure and build based on your interaction. So it will start to fill out the configure pane here. For those of you who are listening and not watching on the podcast, It's just two different tabs that you can, or button and modes that you can switch to.

11:52One is more beginner's mode. That is the create, which is more of a natural language. And the other one gives you more menus and options and things to fill out, which if you're a little more advanced, it's really not that hard. And we'll go through it right now. Like literally anybody can do this. Yeah. Great call. It'll be a little more descriptive for the podcast guests. So in this pane as well, one thing you'll note is the right pane of the screen over here. This is essentially your preview mode. Think of this as like how you test your GPT as you're building it. So it gives you a way to interact with how the GPT will work as you're building it.

12:28So to start with a pretty basic example. So I used to be in deep in product management. And one of the biggest and most important elements when you're working on the agile side of things is developing good user stories, right? So this piece in and of itself creates so much rework when there are bad user stories. So think of a user story essentially as requirements for what to build, right? and there's a very standard framework within Agile principles of how to create a user story. And more times than not, it doesn't get followed. A lot of times it's pushed off to a junior BA to create and you'll have the user story.

13:11The developer will create the feature and it's done incorrectly. There's something lost in communication. And a lot of times it's due to a poorly written user story. Let's create a GPT that helps you craft the perfect user story. And what's cool about this is a literal tool that anybody on this call, podcast, whoever could build this. And it provides direct value in terms of cost savings, less rework, especially with developers, because those are expensive folks. So the first thing we do is we give our GPT a name. Let's call it our user story pal. Let's give it a description. so when you think about the description this is publicly facing so in the gpt store this is what people are going to see when they're on your gpt so think of a really strong kind of hook and value prop when you're creating here and you can see on the right pane or testing pane it's starting to fill this these details out so you see the name there you see the description and just to tell people how simple this is because the description needs to be two things a it needs to be descriptive It needs to tell you what it's going to do, but it's also a marketing thing, right?

14:26Because now it's a store. Maybe you want people to pick yours. And you can use chat GPT to help you write the description. Tell it that it's going to be on the GPT store. Tell it that it needs to be very detailed, but yes, attractive and irresistible to people. Use whatever adjectives you want, and it will help you craft the description. And it's relatively short, but it will help you craft the description in a way that will be attractive to people to actually test this out. Yeah, I'm glad you mentioned that. That's a big unlock that sometimes people forget. It's like use chat GPT as you're building your GPT to improve it.

15:01Have it as your sidekick as you go. It's a really important concept. up. So the next major piece is called instruction. So these are detailing your GPT, how to interact, its persona, what it should do, all of those things. And the greatest thing here is you're building this into your GPT. So the user in a blank chat GPT interface no longer has to put in all this detail. It's built directly into it. So I'm going to add some detail here. We'll go through it. All right. Let me pull it up here. I'm doing a nice copy paste to make a better use of time here. Sure. Okay. So let's pop this in here. All right.

15:46So you're an expert user story writer with 30 years of experience writing perfectly defined user stories. So really important to set your persona. So what role does your GPT play? Another key piece, give it a very specific role that's well understood. Don't give it like five different roles that may confuse your GPT. All right, the next step here is going through detailing the process of how the user will interact with your GPT. So this is where you're instructing your GPT how to interact. So what we have here is when a user engages with this GPT, follow the below process. So the first step, ask the user to provide a copy of their user story or provide details about a user story they want to create.

16:31Step two, after the user has provided a copy of the user story, give it a grade of one to five based on the quality of the user story. So you're giving instant feedback to the user, which is pretty cool. And you can use scoring type of ideas like this. Number three, provide a clear explanation for the grades. People want to know why they got a certain grade. We're going to give them that explanation and then provide clear instructions on how to improve it. And then this is where the interaction comes into play. We want our GPT to ask clarifying questions that will allow the GPT to provide an improved user story.

17:06So the GPT has the context of the initial user story. Now we're saying, okay, GPT, based on what you know, ask some clarifying questions so you can now improve the user story for the user. Next up, after the user answers the clarifying questions, provide an improved user story following the correct framework detailed in these instructions based on the user story the user provided. If the user is, if the user story is too large in scope, provide a recommendation on how best to split it up. And that's a key point too. That's a big issue with user stories. People will make them too large. So great.

17:43We have our custom instructions here, but I mentioned follow this correct framework. What is the correct framework? So this is exactly easier. Like you mentioned, let's go to chat GPT and have it craft this for us. Let's go to ChatGPT and we'll mention, we'll give it the persona again. So we're going to say, you're an expert user story writer with 30 years of experience of the same persona. Then we're going to say, provide a detailed framework for how a perfect user story should be crafted, including the necessary sections in the user story and the detail that should be included in each section.

18:24So we're having ChatGPT do the work for us. Now I'll add one thing as we're waiting for ChatGPT to write this for us. In this section, and later on, you can also attach documents and maybe Matt, I'm stilling your thunder here, but you can add additional content. So if there's a book or an article you really like, or a best practices used in your company that, again, not many people actually follow, you can take that PDF document or whatever format you have it in and use it in here as the instructions for the GPT on how to A, grade and B, feedback, and three, now write an updated one that will follow whatever guidelines you want your company to follow.

19:10So again, it could be an internal document, but this could be a best practices from a book segment that you really like. Yeah, no, that's a perfect example. And back to our matrix example, right? That's the detail that we're plugging into the back of this GPT's head so it knows what to do. And we're actually going to do that in a way here in a second, again, leveraging chat GPT, but great reference. If there's a great article on how to write user stories, a book, whatever it may be, those are great things to add. So let's check out the output. It mentions a title. Yes, we need that. User role persona is a very important piece.

19:45We want to know what role that's interacting with this system. What's the role that they're playing? The user story statement, those that are familiar with it will be very familiar with this. A typical user story framework is as a, whatever the user or persona is, I want to achieve some goal so that explain what they're, the benefit they're going to get from that. So very clear framework there. acceptance criteria gets forgotten way too many times. It's what's the definition of done. How do I know if this work is complete? Gets forgotten way too many times. Any dependencies or preconditions, non-functional requirements, data requirements, lots of good stuff here.

20:28So I'm going to take just the first bits here. I'm going to take down to, let's go down to like dependencies and preconditions. Some of this other stuff we'll leave out for the purpose of this exercise. All right. So we'll come back down here and we will pop in our details for the framework. Let me just add in this bit of detail. Another thing you can do is you can break things up using different things. Again, for those of you who are not watching this, what Matt is doing is he copied the input we got from the regular chat GPT on what are the best practices and framework to develop a user story into the instruction sections of developing the GPT and is creating a visual separation like with a dotted line or any visual separation that you want to create because it helps the GPT.

21:27And by the way, same thing with just regular instructions in chat GPT to understand the different segments of what it is So the beginning was general instructions. And now he's adding the actual framework into it and saying, okay, this is a framework I want you to use. Yeah, perfect explanation there. So it adds that separation. Now we're getting into the framework. So I've added detail around crafting a perfect user story involves meticulous attention to detail and understanding the project requirements and stakeholders. Here's a comprehensive framework for creating the ideal user story. So I've popped in the key pieces I want there.

22:03So now it has a reference for the detail that should be included in the user story. So the other piece I want to mention is the knowledge base. And you hit on this. Let's give it some context for what a good user story looks like. Again, instead of me coming up with examples and ideas, let's just ask ChatGPT, right? So we'll say, we'll take the framework down to where we selected. And I'm going to say, give me five examples of great user stories using the below framework. All right. Pop that in. And again, in ChatGPT itself, for those who are listening, in ChatGPT itself, we're asking ChatGPT to give us examples of what it told us to do.

22:54So he told us to write a title. Okay, give me five good examples for a user story title. You told me to provide acceptance criteria. Give me three examples of what a good acceptance criteria. And we're going to use these examples in the GPT to give it examples. And one of the most important things when crafting any prompt, but also instructions for chat GPT is examples. And in the professional language, it's the difference between zero shot and few shot prompting. But basically what it means is it means you're going to get much better results if you give it examples of what is good and what is not so good or what is bad.

23:35And you can give examples for both positive and negative things as long as you explain what they are. And if you do that, then that helps ChatGPT and in this case, this particular GPT to be a lot more concise with its answers because it will try to align with what you told it is good or bad. And you can use very long examples for anything. And if I generalize this, let's say you want to write a proposal and you have a proposal framework you wanted to follow. You can actually paste in a proposal framework. Let's say you have a script for salespeople. So take your successful salespeople have a transcription of what they use in the actual calls and use that as an example to create a new framework for your salespeople.

24:22Each and every one of those things that you have something that is successful in your business, you can use as an example. And like I said, you can also use bad examples. Like when we did this, the thing failed, whatever that thing is. So you can also give bad examples. So don't follow this because this failed every time we tried it. And now ChatGPT can have a much clearer framework or if you want guardrails to work within. Yeah, that's a perfect explanation. And a key point you hit on the whole notion of garbage in, garbage out. Same thing goes with ChatGPT, generic in, generic out, right? You have to be very detailed in your prompt and what you're looking forward to do.

25:03So what I've done here is I've just copied the five examples it gave me, popped over to a Word doc and pasted it in there. So now what I'm going to do is file and just download this into a Microsoft Word doc. I'm going to come over to our knowledge base. So for those that are listening, one of the features is the knowledge base. So you're able to upload files. And the beautiful thing here is you can upload Word docs, PDFs, PowerPoint presentations, Excel files, images, HTML files. There's so many types of files that you can upload into your knowledge base to give it context. And I'm probably forgetting several.

25:44I don't know if there's any others you can think of that are allowed and acceptable. And I think any file you're going to have can be converted to one of these things, right? So even if you have a file in a different format, if it's a database, you can probably convert it to a CSV. If it's a document from a different source, you can probably convert it into a PDF and so on. So really all the major file types you can upload here. And again, these file types are sources of information for the GPT to do the specific thing we're going to ask it to do. So I gave an example earlier. If you have a internal document that explains how to write user stories in your business, load that document in here.

26:29If you have a style guide and you want this thing to create images for your PowerPoint presentations, then load the style guide in here, it's going to use it and so on and so forth. Like whatever use case you have, the more background information you're going to give it, the more accurate the results are going to be. Yeah, great points. And the one thing I did add, again, another line break, I added a note, just giving the GPT reference in my instruction saying the attached document in the knowledge base includes five example user stories for your reference. So now it knows how to use that document in the knowledge base.

27:06Very important. Otherwise, ChatGPT is just going to figure it out on its own. And 80 % of the time, it'll probably be fine. But 20 % of the time, you may get some weird hallucinations. Better safe than sorry. I want to add two things to that. One, if you have more than one document, then name the document. So document this and that is this and that information. So again, ChatGPT knows what information to find in each. Another thing that I'm doing in mine when I use these kinds of things, not this, but even more specific, let's say you want to create a GPT that is your employee guidebook. So you're going to load all the employee guidebooks that you have in your company into ChatGPT.

27:46And then people can ask questions like, how long in advance do I need to request a vacation? How many vacation days do I get if I'm just starting in the business? Who do I need to call to get a new employee card? All these things can be in this document. What you want to add in the instructions is say two things. One is only use information in this document to answer the questions. If you do not find information in the document, please say, and then put in quotations, whatever you wanted to say, information not found because otherwise it will give you an answer. It will make up an answer because you ask it for an answer.