136 | Scaling AI: Build your own custom GenAI Engine in three easy steps with Pam Boiros

0:00Hello and welcome to another live episode of the Leveraging AI podcast. the podcast that shares practical, ethical ways to leverage AI, to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host, and we have a spectacular episode for you today. And the reason it is spectacular is because when I talk to businesses about AI, and I talk to a lot of businesses, either through my workshops or the courses that I run, or when I speak on stages and so on, I get to talk to a lot of companies, Most of them, the place they start with AI implementation is marketing.

0:38And there are goods and bads to that. The first thing is there are a lot of AI use cases in marketing that are already good enough. So that's one aspect, which is great. The flip side of that is that a lot of people are really frustrated with trying to implement AI in marketing. And the reason for that for most of them is that it's very hard to go from a demo or a prototype to an actual operational working solution that's actually providing consistent business value. And that obviously is involved with the fact that your leadership team is putting a lot of pressure, start using AI, get to do this faster.

1:15We need to cut budgets. We need to get this there. And that combination of on one hand, you feel that you can do this, but on the other hand, it's very hard to make it consistently good and to work in the scenarios that you need is creating a lot of pressure and a lot of stress on a lot of marketing people. The good news is that our guest today, Pam Boros, has been in marketing for a few good years. And when I say a few good years, she held CMO and executive VP of marketing in several different large companies. So what she brings to the table is a really in-depth knowledge of how marketing in general should be done right.

1:57a very holistic approach with real life knowledge on running large marketing departments. And she learned herself and her team how to bring AI into that in order to infuse the actual processes that are happening in the department with AI to get them to be done faster. And she has developed a methodology that she's going to work with us, including the tools that she's using. So this episode is going to be very hands-on. Here's the tools that I'm using. Here's the process that I'm using. Here are the prompts that I'm using in order to get actual marketing business outcomes using AI. I find this to be really exciting.

2:37I know a lot of people are going to find value in this. And hence, I'm really excited to welcome Pam to the show. Pam, welcome to Leveraging AI.

2:49In 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.

3:29Hey Isar, so good to see you and be with you today. And same here. You and I spoke several times before. I was very impressed with everything you're doing. And as I mentioned, you and I think from that perspective are very similar where we came to the AI world, not, oh, I want to become an AI expert, but rather I'm a business professional who run really large marketing in your case and in my case, just business organizations. and we're just looking for ways on how to implement this new capability into all the stuff that we already know. And that's why I think you're so good at what you're doing because of that.

4:05It's not the AI, it's just a means to an end that you already know how to achieve. And so I'm very curious and I'll give you the stage and we'll take it from there. Awesome, that's so great. I'm looking forward to jumping into the demo in just a second. I just thought I would set it up a little bit. Thank you so much again for the great introduction. I'm going to be demoing a tool that's called custom GPT.ai, not to be confused with the custom GPTs in chat GPT. This is a standalone solution. And I think what it does, what I hear when I talk to marketing leaders and what I experienced myself as a marketing leader is there are three areas of concern for most marketers.

4:43One is hallucinations, quality of the content. The other is the content voice and tone, which when it comes out of one of the standard models can be very flat and very kind of plain vanilla. And then data protection and the file security of any of the files you upload to a system. So I was specifically looking for a platform. I'm not a technologist. I was certainly not going to build this thing. So that's why I was so thrilled to meet the folks from customgpt.ai, including some of my friends who are on the phone on the call with us today. So the dirty little secret is that team members in marketing teams, if you're a marketing leader, they are using these tools, no matter what they say or not.

5:23They're just too enticing. They're just too compelling. And that might mean they're using free tools. It might mean they're actually paying for the tools out of their own pocket. But you as a marketing leader have absolutely no visibility of what's going on. So there's a lot of what I'm seeing, random acts of AI. That's what I affectionately call it, random acts of AI, where a marketing team member comes to you as a marketing leader and is, hey, can I buy this book? And you say, sure, that's no problem. Hey, can I take this course? Sure, no problem. Hey, can I get a free or a paid license to this tool?

5:53Sure, no problem. So there's that kind of whack-a-mole of is what's happening. But none of that is going to let you build it for scale or build this capability within your team. So that was the vision for the training program that you alluded to, Isar, that I built, which is called the Marketing AI Jumpstart. We're not going to talk about that, but we are going to the fundamental core when I deliver the marketing AI jumpstart for a B2B marketing team is this custom GPT.AI and the ability to very quickly, very easily build your own custom AI engine. So that's actually what we're going to do. So I'm going to go ahead and screen share with you.

6:28So while you're doing this, I'll share a few things. One, first of all, if you're joining us live on LinkedIn or if with us on Zoom. So first of all, thank you for being here. I see a bunch of people already there and I appreciate you spending the time with us. First of all, go ahead, introduce yourself in the chat so people know who you are, where you're from, what the size of the company you're in, where are you in the world. It's always really interesting for us to know that. And it also helps us answer the questions with relevance to the size of companies you work for and the type of business you're in.

6:55So please do that, whether you're on the Zoom or on LinkedIn with us. If you're not, what the hell are you doing? We do this every Thursday. Come join us. It's always at noon Eastern, always on Thursday. You can ask questions. you can network, and you can be a part of the conversation. You can do that with us. Like I said, go to my LinkedIn. There's a link on how to sign up for these, and you can get reminders every time we do this. So that's one aspect. The other aspect is I want to relate to what you said, Pam, as far as the biggest gaps that people have. And I agree with you 100%, right? So people are terrified about their own data, but there is ample statistics from the biggest kind of like consulting companies in the world, the McKinsey's and Accenture's of the world, that about 75 % of employees calling, they call it, you had another kind of like nice name to this, but they bring your own AI.

7:47So people bring their own AI to work. And even if you're going to block all your systems and you're going to block all access on the IT and you're not able to install, they have a phone. They can have chat GPT on their phone. They can have Claude on their phone. They can have stuff paste from their emails on their phones into ChatGPT or into any other tool. And so by putting your hand in the sand and say, I'm going to just block anything until this thing, until we figure this out, is not a good approach. And having a more structured process with great training and education for your employees is the right way forward.

8:26But with that, let's dive into your process, Pam. Yeah, I couldn't agree with you more, Isar. And that's exactly what we're going to show today. So we're in the custom GPT.ai environment. Everything in the custom GPT.ai world is called a project. So a project is a way for you to create a collection of training materials and basically a different use case. So I'm showing two that I use on a regular basis. Again, I'm a fractional CMO and a marketing consultant. So I have built a B2B marketing wizard, which has trained on all the latest and greatest thinking in B2B marketing. I also have what I call a report synthesizer.

9:07I try to read everything that there is about marketing, HR tech, which is the area that I'm in and AI. That is a heck of a lot of stuff to read. So the first thing I do is I upload and I'm sure you face this too, Isar. I upload all the reports and first I query, give me a summary because I want to know which ones are worth my time. So those are just a couple of the projects, But we're going to build a project from scratch today because I'm going to show you how easy it is to do it. And then we're going to we're going to actually play with it. So just go ahead and click to create a project. And the first thing you do is you just need to give it some kind of a starting point.

9:42So what we're going to start with is we're going to start with a website. You could start with a Zapier integration to data files you already have. You could link to Google Drive. We're just going to make it super easy. We're going to start with a website. So the company that we're going to build this for is a company that's called Beamery. So I'm a fractional CMO and I work with solution providers, which are basically vendors in the HR tech space. Now, just full disclosure, I have no official affiliation of any kind with Beamery, but it's a company that I know and that is always on the list of hot companies in HR tech.

10:18And I happen to know that they have a lot of great content that we could get even not being part of their ecosystem to build this demo. So that's why I picked them. So they are a talent platform that helps you hire candidates faster and develop the skills of the workforce. So they're really focused on employee retention. And again, they've won a bunch of awards. We're going to go ahead and we're going to build. We're going to start with their website. So it's going to validate the website. That just takes a quick second to do. And let me just say one other thing about custom GPT while this website is being validated.

10:56So what custom GPT.ai, I think of it like a Lego block, right? So it's a Lego block that snaps on top of one of the foundation models. And one of the foundation models, the primary foundation model that these guys use today is they use ChatGPT 4.0. Now you can choose a different model. They also have Claude, use Claude as well as an option. But what I like about these guys is that they have built it in a model agnostic way. We know that all the big companies have their model wars. I don't want a piece of that. I don't want to let the billionaires spend their billions continuing to raise the bar on the foundation models.

11:36But CustomGPT snaps on top of anyone their model agnostic, which again is what I really like about it. So it accepted our, I'm just going to get a, I've got a demo folder handy here and I am going to just upload something that makes my background and my avatar look better just so it doesn't look quite so general. So it's on brand, easily set those settings, very intuitive. So the first thing we're going to do, a couple other things about custom gpt while we're talking about it this is a rag system so this is retrieval augmented generation so the first thing you need to know about that you've not read about rag is it really helps with the solution the hallucination minimization this system will literally tell you sorry i can't help you with that which as a marketer i would much rather hear than a typical chat gpt response which gives you a very confident very wrong answer so i think that's really a key thing.

12:39The second thing, we're going to upload some data in a second. All the data is protected. It stays within the custom GPT environment and its data is fully encrypted. These guys are SOC 2 compliant. And if you upload any files to a foundation model, you don't always know if that file is going to be used to train the model, further train. We know all the foundation models are desperate for data. They're doing synthetic data. So any data they get, they're going to very happily use to train the model. With Custom GPT, anything you upload here, if you upload a customer list or any kind of proprietary information, it stays secure.

13:15And then very importantly for marketers, the style and tone, your voice and tone. We're going to do that in just a second. So we've created our new project. I entered the URL. We've got a couple things that set our avatar. Now, technically, you could start using this today with that one data point, which is just the website for Beamery. Now, I want to give it a lot more data, but basically, you're honestly ready to go with the ask me anything as soon as they upload everything. But we're going to upload a lot more data. So what kind of stuff would you upload? So if you're a marketer, you might want to upload your company case studies, you might want to upload your ideal customer profile, you might want to upload your sales decks, you can upload anything in a variety formats, PDF, Word, Excel, MP3 audio, MP4 video.

14:05So I'm going to go ahead and just upload some items here. Sorry, let me just upload some data. I've got a pile of data over here that I downloaded that all publicly available stuff from Beamery, from their website and from other sources. i'm just gonna grab those things and carry them over give me one second yeah so again to iterate on that it's consider your business or a segment of your business this could be just for hr or just for marketing or just for sales or just for b2b sales or just for b2b sales in north america like each and every one of these projects could be very specific to whatever needs you are going to use it for.

14:54So think about this tool and similar tools, right? And again, just to be clear, Pam or myself are not affiliated in any way with KachimGPT.ai. It's just a tool that Pam is using regularly with actual clients and seeing a lot of success. That's why she's choosing to demo this. They're not paying her anything to do that. But the cool thing here is that you need to understand that once you have a tool like this, you can use it not once. It's not, okay, this is now the repository for my company. No, you build these custom projects for each use case because the data state might be different and the instructions might be different and the look and feel might be different because it's serving a different purpose.

15:34And that's the cool thing about this tool is that you can do that and do as many of these as you need. Yeah, it's a great point. So as I said, Isar, all of my clients have adopted this tool because they see the value when they go through a demo like this. So in the couple of seconds to talk through that, these files have been uploaded and now they are part of the training set. Now, again, I could do a really fancy integration through Zapier or some of my other things or Zendesk, like if you have a customer service repository. The only other thing I'm just going to add here is I'm just going to add their YouTube channel.

16:10So give me one second. I'm just going to copy that over as well. Now, obviously a robust company or a company that has a robust YouTube channel. Think about all the amazing value that you're going to get from that. So we have now added that channel and that's going to proceed. Now where I think the, what makes I think custom GPT super interesting to me as a marketer is this idea of giving the agent of persona. This is really cool. So basically we know what we want to do. We want to create a, something that's useful for marketing. So I've got a something I'm just going to copy and paste in here.

16:48So you can just type in here, or what I love about what these guys do is they have a persona generator. So basically you tell it in plain language, what you are, and it's going to generate a persona in the kind of style and approach that makes the most sense for the custom GPT environment. So it's going to clarify things. It's going to give it guidelines. It's going to give it special instructions. And all we need to do is just copy and paste. It looks like code, but believe me, it's not as scary as code. Can I close that out? And by the way, code is not scary if you don't have to generate it, right?

17:24If the machine generated for you after you gave it a few words in simple English, then I don't care. It's a great point. And you said it reminds me of something else. We think about all the years that we've all been learning, not me, but other people have been learning programming languages. Human beings were taught to learn the language of computers. With generative AI, computers now speak human language. So that, I think, is the real differentiator. And if you treat it like that, treat it as a tool, somebody, a really smart intern that understands language, you are going to be very successful with it.

17:59So we have one of this. Again, just not to jump to a different subject, but I started writing and creating little pieces of software that do small little things for me. I've never written a line of code in my life. I don't understand a single word that's in the code. And yet I can do this and I can troubleshoot it because tools like Gemini or Claude or ChatGPT allow you to write code and allow you to troubleshoot code and so on and so forth. So not even going to like the real coding tools like Cursor or GitHub Copilots or stuff like that. just within the tools we use anyway, you can do a lot of stuff.

18:32So don't be afraid of that. If that's something that you're considering doing, if it's not too complex, just play with it. It's a fun experiment. Yeah, absolutely. I think that's part of it really like tinkering the word tinkerer. I don't consider myself that, but I've become a really curious, interested tinkerer when it comes to these tools, because you don't really know the full capacity of what they can do until you start playing with them. So this persona is now in here. So now this thing knows what it is, we're going to save those changes. And basically, this is ready to go. Now, this is almost like a cooking show, Isar.

19:06So we have put this in the oven. Now we have given this a lot of stuff to absorb. So I think it's probably ready for us to interact with. But just because we're on live TV here, I'm actually going to go over, I have another project, which is exactly what I just built here, the Beamery Marketing AI engine. And this is ready to go. And I know this is ready to go because I trained it several weeks ago and I've already been playing with it. So it's just same thing. That's in the oven. We just brought the beautiful turkey out of the oven and we're going to interact with this one. It's the exact same thing, all the same training, all the same prompts.

19:45So let's get started. So there's a bunch of things you can do in marketing. And I would love if anybody on the call has something they want me to try. I'm a long time, I always say I'm a recovering software product marketer. So usually the first rule is always know what you're going to demo in advance. But what I've found is that the prompts or ideas that people suggest on a conversation like this are actually super interesting. And I'm always happy to give them a try. But let's go ahead and let's just start with the first thing. So you're going to interact with this prompt box, just like you would any other chat.

20:18GPT or any other tool. So a lot of things that marketers have to do is write a blog post. So I'm going to ask it, please write a blog post of 700 words based on

20:38Zalando, which is a customer case study that I know is in the training set because it's one of the things we updated. So imagine the repurposing. So you've written a case study, you have gotten approval from the customer to publish it, and now you want to put it in all these other places. So the ability to take this and get a pretty good draft right away that is through the style and tone of Beamery, which hits on Beamery's key product attributes, which has the key findings from the customer's success story, again, extracted right out of that case study. Now you could decide this is great. So those of you who are not watching, we just got a nice 700-ish words long blog post that takes into account all the information that we gave it that is related to Zalando's project that is part of the training data that we gave it.

21:38So it's not something you can get from ChatGPT because it doesn't have that specific data behind the scenes. Yep. Great point. And I just refined it to say, okay, I really want to make it 500 words. So it just did that on the fly. Now, of course, are we going to copy and put this on our blog? No, we're going to take it to a document. We're going to validate what we have. We're going to maybe have stakeholders reveal it, right? So we want to be careful always with anything that comes out of generative AI. It's only as good as the training data and we still need to check it. But if you get 85 % there rather than starting with a blank page.

22:12Pretty good. So that's the first one. I want to pause it just for one second, because there's a bunch of questions coming in. So the first question, which I think is a great question because you alluded to that, but we didn't explain why or how. So there's a question from Sarah on LinkedIn, and she's asking, how is this different than creating a GPT on ChatGPT? So basically, the custom GPT in ChatGPT versus custom GPT.ai, the tool, what are the differences and why is this better. Yep. So you could certainly do that. Of course, what I think is you've built a more general purpose tool that lets you do a whole bunch of marketing activities with the same training data set.

22:51And with that kind of consistent persona, I think the other thing that I like about it, it's not demoing the way that it works here, but all these are me, But what you get is you get actually the visibility of what all the people on the team are doing with this environment. So as a marketing leader, you can coach people, right? So I have this situation with my other clients where I can see what the folks inside the client are doing and what questions, what prompts they're asking. And I can coach them and say, oh, did you think about this? Did you think about refining it further? How about using it for this use case?

23:26You only seem to be using it for LinkedIn posts. Maybe we should be using it for our webinar program. So as a marketing leader, I can coach and get all of their skills up. With a custom GPT, that's not possible. So that's a great point. A few, two other points that I can think of. One is the amount and types of data you can put into it. So you can put in significantly more data here than you can put into a custom GPT. Custom GPT is actually relatively limited with the amount of data you can load to it. The other thing that I'm not 100 % sure about here, but in a custom GPT, you're also limited with the amount of instructions you can put into it.

24:04So I've hit the cap and I think it's either 8 ,000 words or 8 ,000 characters that you can put in when you're creating a custom GPT. And I hit that more than once. And now you need to start making compromises that you're like, okay, I got to cut this off. I got to make this shorter. And then it's not exactly what you wanted because you're limited with the amount of instructions you can give to it, which I don't think is the case here. And then the next thing is really the integrations, right? So you just showed us that there's all of these pre-existing integrations, like to your customer service system, probably your CRM, these kinds of things that don't exist on ChatGPT.

24:41Now you can do it in custom GPTs on ChatGPT because it has that API section at the bottom that almost nobody's using, but you can have ChatGPT write your code to connect to an API of a third-party platform. that's way above the pay grade of most people who are going to use these tools. And so if you have custom built integrations into existing tools that you're using in your company, like Zendisk, like Notion, like tools, like that's a huge time saver. And now you can use them in specific projects back and forth. And then the last thing that I would say, and this one I'm guessing because I didn't use this tool, but it's the same in other tools, is that you can embed the search bar, basically the chat bot into anything very easily.