Drew Hart, your go-to guy for all things Amazon. Drew kicked off his entrepreneurial journey in 2018, launching two private label brands and now providing Full End to End Account Managment Services Agency that handles over 40 brands, Vinculum (VIN-CUE-LUM), which offers a full range of Amazon services, making him a one-stop-shop for Amazon success.

> Here’s a glimpse of what you would learn….

In this episode of the Ecomm Breakthrough Podcast, host Josh Hadley welcomes back Amazon and e-commerce expert Drew Hart. They explore the transformative potential of AI tools and bots, particularly large language models like ChatGPT, in enhancing operational efficiency and reducing overhead for e-commerce businesses. Drew shares practical strategies for utilizing AI, including creating custom GPTs tailored to specific team roles. The discussion emphasizes the importance of adapting to a rapidly changing landscape marked by rising costs and fees, and how leveraging AI can help businesses remain competitive. Key insights include first principles thinking, chain of thought methods, and practical applications for AI in strategic planning and problem-solving.

Here are the 3 action items that Josh identified from this episode:

Resources mentioned in this episode:
Here are the mentions with timestamps arranged by topic:

Special Mention(s):

Related Episode(s):

This episode is brought to you by eComm Breakthrough Consulting where I help seven-figure e-commerce owners grow to eight figures.

I started Hadley Designs in 2015 and grew it to an eight-figure brand in seven years.

I made mistakes along the way that made the path to eight figures longer. At times I doubted whether our business could even survive and become a real brand. I wish I would have had a guide to help me grow faster and avoid the stumbling blocks.

If you’ve hit a plateau and want to know the next steps to take your business to the next level, then go to www.EcommBreakthrough.com (that’s Ecomm with two M’s) to learn more.

Josh Hadley 00:00:00  Welcome to the Ecomm Breakthrough podcast. I’m your host, Josh Hadley, where I interview the top business leaders in e-commerce. Past guests include Kevin King, Michael Gerber, author of The E-myth, and Matt Clark from ASM. Today, I’ve got a returning guest in Drew Hart, and we’re going to be talking about how to build AI bots and tools that are going to set you apart from the rest of the competition. They’re going to help you reduce your overhead and also give you the strategies that you need to compete and stay relevant in the marketplace for years to come. In our previous podcast episode, Drew talked about the critical mistakes that business owners tend to make with wasted ad spend on Amazon. So if you want to go back, relisten to that episode from Drew, it is well worth the listen. This episode is brought to you by Ecomm Breakthrough, where I specialize in investing in and scaling seven figure ecommerce brands to eight figures and beyond. If you’re an ambitious eCommerce entrepreneur looking for a coach or a consultant who can help take your business to the next level, my team and I bring hands on experience, strategic insights, and the resources needed to fuel your growth.

Josh Hadley 00:00:53  So if you or someone you know is ready to scale or looking for that coach or consultant, reach out to me directly at Josh at Ecomm Breakthrough dot Com that’s econ with two M’s. And let’s turn your dreams into reality. Today I’m excited to bring back on the show Drew Hart. He is your go to guy for all things Amazon. Drew kicked off his entrepreneurial journey in 2018, launching two private label brands, and he now has a full services agency called Vinculum. And this is full end to end account management services. And it makes him a one stop shop for all things Amazon success. So with that introduction to welcome back to the show, drew.

Drew Hart 00:01:24  You’ve got a great voice, man. You’ve got that that buttery radio voice I love it.

Josh Hadley 00:01:27  Well hey that’s what that’s why I record the audio versions. It’s not the video versions. I’m not a TV guy. Right. It’s just the video. But thank you.

Drew Hart 00:01:35  Yeah, it’s great to be here, man. Thanks for having me back.

Josh Hadley 00:01:37  Awesome to have you on the show. I want to dive straight into this topic that we’re going to be talking about today because, man, it is a meaty subject. You and I actually set up separate calls to talk about this, and you were kind of running it by me at the time. And you’re like, do you think this is valuable? And it was my mind was exploding of like use cases that I could implement this into my own business. And look, I think that as Amazon sellers and just e-commerce in general experiences a lot of margin compression with inflation going up, right. You have tariffs going up, you have Amazon fees only continuing to go up, and Amazon penalizing you more and more for any overstocked inventory or under stocked inventory, like it’s just never ending in terms of the margin compression. So the most important thing that you need to focus on is like, how do I stay relevant as a brand for the next 5 to 10 years? And I think what you’re going to be showing us today, drew, is going to be something that people will be able to use and almost have, like a professional like mentor that almost like sees it all in their back pocket to be able to bounce ideas off of, let alone being able to maybe reduce some overhead.

Josh Hadley 00:02:38  And maybe you don’t actually need to hire as many staff members as you have, because you’ve got some AI tools and bots working on your behalf. So is that a fair enough segue into this, or why? Why are we even going to be talking about this today? Yeah. No that’s.

Drew Hart 00:02:49  Great. Yeah. So today’s today’s discussion. All right. And we’re going to share some content with you. And we’ve got some some free goodies for everybody to we’re going to get into the technical side of like how these large language models work. We’re going to specifically focus on ChatGPT. So if I say ChatGPT or large language model, that’s the same thing. Right. But we’re not going to do is we’re not going to do like a tool demo of like, how do you create automations and go from step A to B to C? That’s a separate conversation. What we are going to do is we’re going to understand, like when you automate that one step, how do you get amazing results out of these large language models that really benefit you? So we’re going to take a manual approach because we’re going to unpack all of this like complexity.

Drew Hart 00:03:21  And I’m going to give some give your audience like a really clear structure around one. How do you think about these tools. And then two, how do you interact with them to create, business tools, whatever those tools are going to be? We’re going to get into that.

Josh Hadley 00:03:33  So I love that. drew, this is going to be fantastic. You’ve got some slides that are prepared. So for our listeners, make sure you do come check us out on YouTube because you’re going to want to see the slides that drew breaks down. And again there’s a special bonus. Those that listen all the way through here. Drew does have some free goodies, like these templates that drew is about to share with. You guys are probably worth over $10,000 itself. Like I’m dead.

Drew Hart 00:03:55  I say sorry to be charging.

Josh Hadley 00:03:57  You need to start charging more for this stuff. Yeah. That’s funny. Yeah. Super valuable though. So come check it out. So, drew, I’ll kind of turn it over to you if you want to share your screen.

Josh Hadley 00:04:05  Yep. Sounds good. Dive into this thing.

Drew Hart 00:04:06  Let’s go for it. Let me know if you can see that.

Josh Hadley 00:04:08  Yes, sir.

Drew Hart 00:04:09  Okay, great. So we’re going to be talking about harnessing large language models right. Or GPT. And as we kind of get into this it’s like well what does that even really mean? when you start thinking about these tools, these new AI tools that are out there, we describe it as prompt development, right? So when you’re going to, when you’re going to work with ChatGPT, the default interaction is just a conversation that you you have back and forth with the default model that ChatGPT provides. But ChatGPT gives you these really cool capabilities. You can build what they call custom GPT, and inside those custom GPT is you’re actually going to instruct whatever that custom GPT is. You’re going to instruct it to give it context. And the most interesting thing to me about, you know, like programming or developing these custom GPT s is you’re going to be doing it, but you’re going to be doing it with the English language.

Drew Hart 00:04:53  Like that is the new programming language for these models. And that matters because as humans will talk back and forth and we’ll use language like, oh, you know, what you should do is you should just go hire somebody and you should go create the job description, and then you should go post that on Upwork, right? To get the thing I use, the word should should actually means like a suggestion because I’m being nice. I’m being formed like I’m being informal with, you know, who I’m talking to will also treat that like you’re being informal. So if you want to like do something, you need to say you will do this, not you should do this. So even though we’re interacting with it with the English language, and we’re kind of treating it like it is almost a pseudo, you know, entity, like a being you need to be aware of, like the language that you use because that’ll matter. But the like the strategic reasons for why, like why we use it and how we use it, we use it to basically maximize the efficiency within our organization.

Drew Hart 00:05:39  So that doesn’t just happen at low levels, where we are enabling an analyst and giving them superpowers by letting them work with one of our custom GPUs. It works with the leadership team too. So we’ll we’ll go through a couple of these examples, where that comes into play. but having a bot helps you solve these really complex problems. You now have a sounding board to, at a minimum, go through, kind of like your first iterations of, hey, I’ve got something that’s really challenging. Maybe it’s logistics, maybe it’s product design, maybe it’s like trying to understand like customer problems or whatever those things are. You can work with that custom GPT to sort of unpack what those things are. Another ones like marketing strategy, which is we’re literally using this right now. Over the last two and a half months, we have built out a pretty sophisticated marketing program for Vinculum, and we launched it this week. Now we use ChatGPT. We actually built some custom GPT prompts, one for strategy, one for content creation.

Drew Hart 00:06:27  Right. So those types of things and then we get really sophisticated. We started doing things like if you use the out of the box language that ChatGPT, speaks with, it gets voice and tone. It doesn’t really resonate with like, our brand or like who I am as drew, because I want my content to sound like it’s from me. So we have a like a lot of like, cool controls. And we’ll share with you guys today, like sort of how we approach all that, but you can do it for a lot of other different things, like if you want to figure out ways to like differentiate your product or your service. you know, or, you know, things like that. and then constructing, like, you know, brand identity, like, we’re actually working on a new service. we have one that’s called Listing Compass, which is sort of like, listing strategy for, like, how do you approach the optimization for a listing? But we’re also working on one that’s like brand focused, which is really, really cool.

Drew Hart 00:07:09  And it’s something that we’re working on right now to to do so you can use it for brand identity and then, you know, the precision and control. Like once you learn how to create these bots, you can really start to dial them in. So there’s a ton that you can do. Like this is still I feel like scratching the surface in terms of like, really what you can do with these tools.

Josh Hadley 00:07:23  Yeah. True. You’ve, you’ve already got my head spinning here. I’ve already been feverishly writing down notes over here, because I’m already getting ideas of how I want my team to be able to implement this into our own business. So why don’t you go to that next slide? Because you do have some examples here. Yeah. but here’s kind of my here’s the thought that’s coming to my mind right now. Drew is with my team. And this is a question for you. Does it make sense for me to almost like tell my team. Tell my team to go out and create a custom GPT to help them in their own role.

Josh Hadley 00:07:57  So like for my supply chain manager, I want you to build a custom GPT that’s going to help you make your life more efficient and better help you make you know, the decisions for how much inventory are we ordering or not ordering, or how do we optimize our containers for maximum efficiency as we’re bringing them overseas? Right. I’m almost like thinking like, can I just like, provide these instructions, have them go watch this video and say, hey, here’s the scripts, here’s what I need you to go do. I want each of you to go build your own bot that’s going to help you be ten times more productive in your own role here in our company. Is that.

Drew Hart 00:08:28  True? Yeah, the answer is yeah. And you can even get meta on this.

Josh Hadley 00:08:30  So let’s hear it.

Drew Hart 00:08:32  You can have them go build a bot that its intention is to help them go build a bot, because you may go give them that directive and they’re like, okay, that’s great. Thanks for sending me this video and giving me these, you know, these kind of starting point documents to go build out a bot.

Drew Hart 00:08:45  They still may not know what it is that they need to build or given the complexity of the problem that you’re solving. So go build a bot that actually understands that that domain, that knowledge domain, and then use that bot to actually design another bot. we do this. We literally do this. We have a master bot. His name is Fabio. So we name all of our bots. We have a bot named Fabio’s. I don’t know why we picked Fabio. but Fabio is our prompt engineer. He that that bot literally creates the core constructs for all of our custom prompts. And then from there, we then start branching out. But one of the first things we’ll do, we’ll see this when we go through one of the examples here, when we build out like a we’re gonna build a customer service agent on this call. And one of the first steps is to create the job description for that customer agent. And that itself is its own sort of like thing. You’re building out a bot that will create a job description, and then you use the job description to go there and go create a customer service bot itself.

Drew Hart 00:09:36  Yeah. Yeah, okay.

Josh Hadley 00:09:38  I like I like this, so maybe, maybe we’ll come back to that and maybe that will be the encapsulating aspect of this is I’m going to ask you like, all right, so how do I go build this bot that’s maybe more meta. That would then allow my team to go like I could say, hey, I want you all to go watch this training in this video. That’s great. But also, here’s the bot I want you to use to go get ideas and to go construct your own bot that’s going to like actually reduce constraint in your own role and allow you to a10x your output.

Drew Hart 00:10:04  Yeah, I think, you know, like depending on your team members and their capabilities like to explore things. We’ll get into some pro tips here in a second too. But it’s all about the capabilities in their context. Like if you’ve got someone on your team that doesn’t really understand their knowledge domain very well, or they don’t understand like what the goal or objective is that they’re going to create, which can be a little confusing because this is all new tech, right? Like there really are no like, you know, socially understood like norms around like what are we building and how do these work and how do we use them.

Drew Hart 00:10:28  So if you feel comfortable, like letting someone go explore, have them go. Just build the bot. Like just have them go build like whatever you’re in goal is, but just realize that they may build something that is 70% useful, 60%, 80% useful, right? But it’s not really covering like all the little corner cases that might be important to you. That’s fine. Go rebuild the bot again after you kind of learn on your first iteration. I think for us, like my marketing content, bots, I think I’m on version three, Josh for that. But version one was like, damn good. It was really damn good. So we just got more and more advanced around, like, how do we want to dial it in with voice and tone? you know, you’ll see things like one of the dead giveaways on AI content right now is like hyphens, like ChatGPT loves hyphens. So, you know, we want to control for those types of things because we don’t want people thinking that we’re creating inauthentic content, which I don’t think you are, by the way.

Drew Hart 00:11:16  Like this. This philosophical topic of like, you know, is it is it really our content? I am the mastermind of like, what the content is going to be and the fact that I use a digital copywriter or I use a human copywriter really didn’t change how I go about the process or like fundamentally, it just changed, like who I work with in order to execute on that process.

Josh Hadley 00:11:31  So makes sense. All right. Let’s dive into so there’s a way that I’m envisioning using this in my own team. But tell me, you know drew as you work with other Amazon sellers go back to that slide and let’s talk about what are the different ways, people could use bots in their own brand. Let’s go to slide three. Yeah.

Drew Hart 00:11:47  This one right here. Yeah, yeah. Sorry. So, yeah. So I think, you know, the traditional use case is the ones that I see a lot like online, right. Or with like YouTube videos, people really get focused on like the marketing campaigns.

Drew Hart 00:11:57  They focus on content creation because, you know, we want to be everywhere for everybody, etc.. doing customer service is a relatively simple one, which we’re we’re going to do that one today and like doing things like listing optimization for those who are really focused on Amazon getting your titles right, you know, putting the right keywords in there, even understanding like which keywords need to go in there, right from lists and things like that. You know, ChatGPT has got a pretty fair amount of knowledge and best practices to understand, like getting you to that. I would never say like GPT always produces like the final result, but it does a really damn good job of giving you the draft that you need to review and then finalize and then go ahead and launch into production. That’s true for any traditional use use cases.

Josh Hadley 00:12:30  Yeah, I love this. And I think you just touched on something that I think people really need to understand a little bit better. with how do you utilize AI? There’s a lot of people that just think like, well, I supposed to go do all of it for me, right? And that’s why there’s an And apprehension to to a maybe use.

Josh Hadley 00:12:44  I but I think the best case and the best use of I at this point in time is to do the following. And it’s called the Steve Jobs principle. It’s the 1080 ten principle. Okay. And Steve Jobs, whenever he like when he had the vision for the iPhone he brought, he did the 10% work up front to like, visualize and mentally think through, like I’m envisioning, bringing a device that can have music stored on it and maybe even a phone, a photos, and maybe even that cell phone. Can we merge these three things together? That’s my challenge for you, right? And I think these are the ways I would I would conceptually think about it and use it. Then he gives that to his team and he kind of sets it up right, to then allow his team to go do the 80% of the work. So he’s already done his first 10%. His team’s doing the 80% of the work. Now, that’s a big bulk of the actual work. Then he’s going to come back for the final 10%.

Josh Hadley 00:13:31  He’s going to review the output from his team. He’s gonna be like, this makes sense. Maybe not so much this. Maybe. Let’s refine that. Maybe let’s start these I’s and cross these T’s over here. And I think that’s the exact same way that I think people should be using AI. It’s like, let it do the bulk of the work for you. Then you use your own human intelligence to be like this. Makes sense, but only if we modify it this this way, or change this just a little bit. would you do you like that principle? Would you agree with. That’s the way you’re using.

Drew Hart 00:13:56  100%, like we use. I mean, we’re using AI in so many different ways, and they’re embedded in integrated into like our workflows, like our formal workflows, SOPs. We use Clickup right to actually like automate the workflows and the tasks themselves, follow along in SOP. And then in the middle of those stages, there’s work that needs to be done. That’s done by human work, that needs to be done.

Drew Hart 00:14:14  That’s done by a bot. the 1080 tens, right? We give 10% load up, frame it up, give it the context. 80% is done by the bot, and then the final 10% is done on review. Refinement. Like truly create that final like finished product on the outside of whatever that process is, then it’s complete. We never rely on the output of a bot on its own. It’s not always accurate. And in fact, like when when you kind of start building out your own bots, you’ll be somewhat disappointed in the precision that comes out of it. If you become really good at prompt engineering and prompt design, you will dramatically improve that final stage, that final 10% of that review process or the 80% of the bot completes. But when you’re first starting out, like you can’t expect the bots to just completely automate all those things unless it’s just something that’s like low value, right? If you’re writing a transactional email and you’re having the bot kind of like do the draft, yeah, you don’t have to put a lot of overhead into that.

Drew Hart 00:15:03  But if you’re going to go build a serious workflow within your business, you’re going to need to take an iterative approach, and it has to have a human. It has to have a human, you know, quality control component to it for sure.

Josh Hadley 00:15:11  Love it. Okay. Yeah. Brilliant ideas.

Drew Hart 00:15:13  So even when you’re doing your own personal bots, it’s the same thing, right? Like like we’re talking about the strategic consultant. That’s a bot. That’s for drew. That’s not a I mean, sure, I’ll give it to my team to go use, but even my interaction with that bot, I have to take it with a grain of salt and depending how good I am about actually communicating the context and instructions that I’m actually asking of that strategic consultant, it’s either going to give me like, like good results or it’s going to give me great results. It’s always going to give me results by the way. Like he tries really, really hard to, like, look amazing.

Josh Hadley 00:15:40  Yeah, I like that.

Josh Hadley 00:15:42  What are some other advanced use cases here then? Yeah.

Drew Hart 00:15:44  So one of the first ones that we did was an innovation workshop. we actually and this was like the first kind of like cool use case for like getting really sophisticated with the layering that you can set up within GPT. So the innovation workshop is you basically tell it that you are an innovation workshop facilitator. And I got this off of a medium article initially like somebody else’s idea. And then we started just kind of iterating on it. but it was like a menu driven system, and you would kind of like go through like an innovation kind of workflow within all within the context of ChatGPT. But it was really cool because you can give it the inputs like, hey, listen, I’m trying to problem solve, how do I diversify my manufacturing sources so that I work away from China? Right. Because tariffs are coming, something like that. It’s a big bold like strategy kind of idea. The Innovation Workshop bot will then start coming up and start working with you, with asking you creative questions.

Drew Hart 00:16:31  You then have to give it the answers. Then it comes back with like alternative considerations. And then as you move through, it’s each of its phases. It’ll eventually come down to like, here’s like the top ten possible outcomes that you you have shared with me as the human. Here’s what you’ve shared with me that are like, viable for you to go explore further.

Josh Hadley 00:16:47  Very interesting concept.

Drew Hart 00:16:49  When you when you want, when you want to do that first draft like like we have a leadership team, right? We have five people on our leadership team. I may not want to go do that innovation workshop initially with them. Now in the past I might have, because that’s really the only way that I’m going to be able to go brainstorm, you know, qualified ideas. But now with GPT, if I give it enough context about my problems and my situation and understands, like what we’re doing, I can go and have that first round of innovation. I can then go into a conversation with my leadership team, get a better utilization of that time, and prime the discussion with some, you know, like some direction that’s already been developed before anybody’s even, like started talking about the topic.

Josh Hadley 00:17:23  Drew, maybe a question that I have that’s come into my mind here is if your team members are going to be like, let’s take your five leadership team members as an example, do you provide them each with their own ChatGPT Pro account write this is the subscription account, not the $100 a month one, but the cheaper version of it. do you pay for that for each of those team members? So as they have these conversations, it creates a memory and it understands who they are, the challenges that they’re working on specific to them. Because I think that’s one of the biggest power advantages is like I’ve been using ChatGPT for two years now and like like I could go ask you like, hey, what do you think are my weak points, my weaknesses, and what are the weaknesses in my brand? And that’s all I’m giving it. And because I’ve peppered it with so many different things over the years, like it knows exactly the brand I operate, it knows I have a podcast, it knows the challenges that my team has and things like that, and even myself personally.

Josh Hadley 00:18:16  So tell me how you like structure that from like team to because it’s not I would assume it’s not a shared account to access these. It’s like they each have their own. Oh we do.

Drew Hart 00:18:24  We do use a shared account, but they they have their own two because we pile on. We have like all of our marketing managers, use one shared account. Okay. Got a couple of different features. So let me, let me kind of go through like what you, you were just asking. So I think the default model for ChatGPT is it’s got memory because you’ll do exactly what you’re describing. You’ll have this, you know, subscription model where it’s like, it’s my bot that works for me. It’s got memory about me, and it’s going to remember that. And that way I can have quick non custom GPT conversations. And it’s going to maintain some context around me so that it’s interacting with me. You know based on what it knows about me. Right. It’s memory okay. But GPT has the concept of projects now.

Drew Hart 00:18:58  So they look like folders on the left hand side okay. Within the project now. So for instance, like we’ve got projects for sales, we’ve got projects for marketing, we’ve got projects for like seller operations. And then if you go in there, all of those conversations are in and around those specific topics. So like our sales one, which does mostly like sales copywriting. So like respond to a client or a brand, maybe they’re asking some questions about a service. And I’m still going to architect the response. But I’m going to go inside that project to share my, my version of like what I want to say. And it’s already got all this historical context around style and tone, and it’s got history in it that I can then pull out and use within the context of sales.

Josh Hadley 00:19:35  Okay. So that in that folder or that project, it keeps the memory just of that project and not the whole that’s the design.

Drew Hart 00:19:41  That’s what that’s what I designed with the concept of those projects on purpose. Okay.

Drew Hart 00:19:45  So it’s like a level up, right? Like again, we’re talking about here is sort of like casual interactions, right? With GPT where I just I fired up, I hit new message or new chat and I start talking to it. And as I move along, it’s keeping context about what am I talking about, either in that project or with the memory, if it was outside of one of those projects. The third one, though, which is what we’ll talk about here, is the custom. GPT is you can put all that memory and all that context in the GPT. You can tell ChatGPT exactly like what you want it to know in terms of the context. So like when we build our bots, depending on the bot, we’ll give it context about its role, like its job description, right? We’ll also give it instructions around our brand voice and tone. We’ll also give it instructions around like our services. We’ll also give it instructions around like our client type. Right. Just like what what what is our client? It’s an e-commerce brand that sells on Amazon.

Drew Hart 00:20:31  You know, whatever the details are. And that then gets used with its main purpose, right? Writing, marketing copy, helping me with the strategic consultant. It needs to know that maybe not the brand voice, but like it needs to know, like who is my client? What is my service? So as it’s helping me out with, you know, a strategic conversation or even an innovation workshop, it’s got confines and context to understand. Like, you know, how will it start to respond to me, you know, based on whatever conversation we’re having?

Josh Hadley 00:20:55  Awesome, I love it. All right, drew, well, let’s get into the meat and potatoes of all of this. Yeah.

Drew Hart 00:20:59  Okay. So I just want to kind of tee this up real quickly. Let’s do it. I’m going to go open up ChatGPT and do some programming here. What we’re going to do is we’re going to do a walkthrough around the conceptual components of how to bring like, you know, snippets together to go build a bot.

Drew Hart 00:21:11  And then for the audience, you guys are going to get all of these like technical documents that would go into a custom GPT. You’re going to need to look at them. You’re going to want to like sort of review them and unpack them. It’s all English. So you’re not going to struggle reading any of these documents. but for now, we’re not going to actually go and do a walkthrough with ChatGPT on how to create a custom GPT, or if you need that, honestly, go ask GPT.

Josh Hadley 00:21:32  Yeah, and a lot of this stuff, it’s not as complex. You know, when you talk about programming, coding, things like that, it’s not as complex as you would as it seems. fact and point being that I created I’ve created my own custom GPT, and I am about as far away as from being a coder or developer as one humanly can be. So if I can do it, you can do it. It’s basically what I’m saying.

Drew Hart 00:21:51  Yeah, totally. And and I’ll just start diving into this.

Drew Hart 00:21:53  What we’re really going to be talking about is like, how do you sequence and and put the components together to get good results? That’s what we’re talking about. So when we when we use this word programming, we’re not really programming like at all. Like we’re not doing programming like in a traditional sense. All we’re doing is I’m going to share with you guys our practices, what we consider to be our best practices for how to do prompt design, use it, change it, explore it, like play with it. But this is just how we do it. So when we look at the structure of creating a prompt. So we are actually going to ask ChatGPT to build a custom prompt. And to do that we are going to give it a prompt. So we’re going to say hey pretty please go build a strategic consultant. But when you do that and you say, pretty please, what exactly is included in all that? And so there’s sort of four parts that we look at when we’re asking ChatGPT to start to produce a bot.

Drew Hart 00:22:38  Now keep in mind we’re not interacting with anything. We’re building a bot at this point. We’re not actually interacting with that bot. So the first step is you got to give it some instructions and you got to give it some features. So you’ll see here we want to create an interactive bot that can assist with internal teams. With customer service it gets cut off. But that’s what we’re going to do there. The red section that says prompt structure, we actually are going to tell ChatGPT to produce output that is in a structured format. And the reason is, is because then we can review it, we understand how it’s organized, and then we can make customizations to the prompt in a very structured way. The final piece is this output formatting. And it’s a technicality we want at least for pendulum. We want our output to be consistent. This may seem a little over the top if you’re first doing this, but what will end up happening if you adopt like you start building out your own bots, you’re going to build out not one.

Drew Hart 00:23:24  You’re going to build out 20 bots in some of those bots, you’re going to want to swap parts and share them across each other. So for instance, if we talk about like our client personas, we will go in this first bot and we will go create a client persona section. When you create a third or fourth or fifth bot, somewhere along the line, you may want to pull in that customer persona component, but you don’t actually have to go and recreate that. You can just go open up, you know, bot a copy the customer persona section and paste it into bot, you know, K or L or whatever one you’re building. Does that make sense?

Josh Hadley 00:23:53  Make sense?

Drew Hart 00:23:54  Okay. yeah. Josh, anything I’m saying is like, not clear. Just like, let me know and I can reframe it. Okay. So this is what we’re going to do is we’re going to build a bot that’s got these components. And you guys are gonna get the swipes for this. So you will get instructions, features, prompt structure and output formatting.

Drew Hart 00:24:07  So these are little screenshots that you can read. don’t bother trying to like replicate those or anything. You’re going to get the actual codings that you can use to do this. So when we actually build out a prompt, there is sort of a process that you need to go through. When we do this, we do it from what I would call like a clean slate approach, the first step for me to build a customer service bot is I want to get a job description, and then I want to sort of like create the main bot. It’s just going to be a, you know, a like a customer service agent. So they’re going to get questions about Amazon products and they’re going to formulate answers based on, you know, answers that I’m going to craft. But I want the bot to reframe them. Right. So the main bot is going to include the instructions the features. And then this structure and output that we talked about in this slide. And it’s going to give me a it’s going to actually give me a bot.

Drew Hart 00:24:51  But that first version on step one is probably not going to be good enough. But I need to start somewhere. And I don’t want to try and tell ChatGPT build me the perfect bot in one step. So the first main bot is just give me a generic kind of customer service bot that meets some main criteria that I’m looking for. Then once ChatGPT does that and you like the output and you’ve modified it slightly to meet your needs, you’re then going to want to add in some specific type of feature or detail, like your brand invoice. So you may not want to put your brand voice instructions in the main bot. You may want to do that as a feature detail. And when you do that, you go develop your brand voice with ChatGPT using this stack of instructions, and it will produce an output format that you can then insert into your master prompt. That’s what you’re doing with additional features in step three. You’re then going to take part one and part two and say, bring these two things together and make it a cohesive bot.

Drew Hart 00:25:41  That’s the general process. Each of these things are done in a separate chat instance. I’m not creating my main bot then asking it to add the feature, then asking it to bring it all together. I’m not doing that one conversation. I am breaking those up into three different conversations.

Josh Hadley 00:25:56  Okay.

Drew Hart 00:25:58  You may or may not like that and you can play around with it. We do it because that clean slate approach gets us really clean results.

Josh Hadley 00:26:03  Yeah. So you’re basically taking each of these one at a time, opening up a new chat thread to. Yeah. Exactly right.

Drew Hart 00:26:09  Yep. And specifically we just do it on tabs. Right. So number one will be tab one. Number two will be tab two. Number three will be tab three okay. And just to kind of share with the audience like the reason I’m suggesting that is your conversation you’re going to have about the main bot is probably going to get lengthy, and you’re going to be talking to it and giving it context and modifying it and looking at the output and then saying no, yes, no, yes, I like these things, I don’t.

Drew Hart 00:26:32  When you finally get to step two, there’s so much noise in that discussion, it’ll get lost on what you’re trying to accomplish with feature detail. So you want to try and add in that voice and tone. You’re like, oh, like, what are you doing? Like you’re not doing like what I thought you wanted to do. That’s why you want to start that new chat instance and just start from a clean slate. That’s the idea behind it. If you don’t think you’re gonna run into that problem, it’s fine. Go ahead and mix the steps together. Right. Sometimes I will actually do that, Josh. But to keep it really clean, if you’re if your first time doing this, your conversations with ChatGPT could get lengthy and it gets noisy.

Josh Hadley 00:27:00  Yep, a great no I agree. And it’s going to keep it a lot more clean. So good. Good tip.

Drew Hart 00:27:04  Yep. okay. So we’re going to build a customer service agent. I’ll go through this really quickly because anyone who’s on the, you know, or watching this that you guys can kind of read through this, but we’re going to I name my I name all my bots.

Drew Hart 00:27:14  I just do that because when you get into the coding piece, it makes it really clear who’s who. If I say the word you in the instructions, who are we talking about? So Kathy Cathy is our customer service agent. That way Cathy knows the difference between herself or itself and clients and you and potentially somebody that you’re going to be talking to as a shopper. Because if it’s a customer service agent, you might actually have somebody who’s named Cathy you’re writing a letter to.

Josh Hadley 00:27:36  Yeah. Do you have, do you have a prompt as that in kind of one of the, the swipes that you instruct the bot or you name the bot, whatever it is and you say your name is now officially Cathy, whenever I refer to Cathy, I’m referring to you as the guy.

Drew Hart 00:27:51  When you get the swipes, we give you a variable name called bot name. And then as you’re building the bot, you will you will maintain that variable bot name. And then when you’re done, when you finally get to the point where you’re going to implement this, you just do a find, replace the bot name with Cathy and then that’s it.

Drew Hart 00:28:06  You’ll see when you see the swipes it’ll be really clear because the variable names are like they’re bracketed, they look like code, they’re bracketed. And you’ll just be able to swap out those. Okay. So Cathy helps team members helps team members. So Cathy is an assistant for an internal team. Cathy is not your customer service agent. Cathy is your customer service agent subject matter expert who helps craft messages. She handles common queries. She doesn’t do any of the tricky stuff right. You can try and throw tricky stuff out and see how she does. But the intention with this design is that she handles all the common things. So 80% of your customer inquiries, 90% of your customer inquiries are going to be the common ones. We’re going to implement a menu based interaction, meaning Cathy is going to ask like how? Like what style do you want me to be in right now? Do you want me to be like assertive? Do you want it to be casual? Do you want me to be polite? You know what? What is that interaction style going to be.

Drew Hart 00:28:47  And then she’s going to generate predefined speaking styles because depending on what the customer query is you either need to be like really formal and polite, right. Because they might be it might be a shopper who’s like really bitchy, or it might be somebody who’s like, I love this product. And you just kind of want to be casual because that’s your brand voice and tone. And then, yeah, then she’s gonna implement some brand specific policies. So there’s a lot here, right? In terms of like what we’re going to build when we go back to this structure, this one, two, three, you’re about to see how do we put all that together. So this is a sequence I did actually build the bot. And I put some time limits on like how long it took me to do this. And I didn’t exaggerate. I didn’t do it because, like, Drew’s really fast. These are Drew’s timelines. I try to give you guys some guidelines around, like, what I think it would take if you did deep work.

Drew Hart 00:29:26  You didn’t have distractions. It’ll probably take you anywhere between an hour to two hours to build Kathy from scratch without the swipes. This is like from scratch. So the first step.

Josh Hadley 00:29:34  I think like but just a double click into this topic right there. Right? Like I love that you gave people like an understanding like, look, you’re probably looking at two hours of time. One of the most important things is like doing deep work. Look, if if our listeners aren’t setting aside enough time during the day to where you can sit down and have two hours of deep work, which means uninterrupted work. Alex Hormozi is the biggest proponent of this. Right. Like Alex Hormozi says, he basically locks himself in, quote unquote, a dungeon, right? It’s like four walls with no other distractions, no phone. And he’s just like, he even puts on silencing headphones. And he’s just like, he’s locked in and he will do ten x the amount of work in one day that somebody else will do, because he’s doing deep work and he’s not interrupted by, oh, I got this WhatsApp message.

Josh Hadley 00:30:15  Now I’m to end this train of thought for WhatsApp. Like I have to do that with my phone, you know, and drew, we’re in a lot of those same conversations. Yeah, I’m blown away as I look at the engagement that happens just throughout the business day in general. And it’s always like there’s always a consistent number of people engaging. And I’m like, do these people not like maybe, maybe everybody’s just on their breaks. And when I’m seeing it, I’m like, does anybody actually work around here? Because I don’t know how you do that. Like, I have to be able to process things like even for my text messages, like people will reply like I’m not replying within five minutes, right? Like it’s more of like, I’ll get to you in maybe 24 to 48 hours when it works on my time schedule, and that’s how you actually move the business forward. So I just think it’s like a really important principle. And like if you invest two hours here, I think the other principle of this is it will allow you and the rest of your team to run ten times faster.

Josh Hadley 00:31:04  So invest the time upfront. And this is going to compound week over week, month over month and year over year in terms of efficiencies in your business. So I just don’t want people to sleep on the fact of like, this is a small investment for massive returns that will compound over years.

Drew Hart 00:31:20  I cannot agree more. It’s like my biggest mistake as an early entrepreneur is that I did not believe. I did not understand the trade off between deep work and like, no, I’ve got all these busy things I’ve got to go be doing. There’s a book, Buy Back Your Time by Dan Martel, that it’s super easy to read. It’s fast. You don’t have to read it cover to cover. You can read chapter three, then chapter seven, then chapter one. super straightforward advice on what Josh just talked about.

Josh Hadley 00:31:43  Yep. Obviously I’ve read the book.

Drew Hart 00:31:46  I’ve got it. I’ve got it back here, I love it. It’s a good book.

Josh Hadley 00:31:48  All right, well, let’s get back into it.

Josh Hadley 00:31:50  So two hour process. this is a normal timeframe we can expect about. Yeah.

Drew Hart 00:31:54  And it’s iterative, right? Like, you know, in the first 30 minutes, you can get your main back going, right? Like you’re looking at what I’m describing here is like the full process for this. it might take you longer because you’re obsessive compulsive or you’re, you know, like, your standards are so high that you’re just like, no, it’s not good enough. No, it’s not going to fine, you know, like, do your thing. all right. So let’s take let’s walk through this. So the first step is to create a semi. Now for people who don’t notice me as a subject matter expert, I’m generalizing this because right now we’re going to build a customer service agent. But so our semi is going to be a customer service agent. I actually need to create that. So what do you want in a customer service agent. This is just like building on a job description.

Drew Hart 00:32:25  Now you know in your at least I know because I designed this slide deck here. Like I’m I’m purposefully going to limit this customer service agent to the criteria that we described here. Right. It’s going to only handle common queries. This isn’t supposed to be some super magical bot that does every single thing for customer service agent. So I’m scoping the subject matter expertise down into something that’s pretty straightforward. I’m just gonna ask GPT for a job description. No formatting or anything, but I am going to read it and I’m going to massage it. And I’m going to make sure that like like I’ve got what I want. Once I’ve done that, I’m then going to use the structure. I’m going to take instructions, which is go create a bot, and then I’m going to add the features, which is the job description. And then I’m going to insert literally copy and paste the prompt structure in the output formatting. So now you see like all these four parts it’s really the features and the instructions. The instructions are pretty lightweight.

Drew Hart 00:33:09  It’s the features part that you’ve got to focus in on. That’s the context that you want to give ChatGPT. So when we start talking about the main customer service spot, that’s the part you want to focus on is the meat and potatoes of like, what is this bot going to do? Then what we’re going to do is we’re then once we kind of get a bot working, I want to I want to wrap this around an internal workflow. So I’m going to take talent that’s relatively inexpensive. Who I don’t want to go hire 4 or 5 bodies to handle my customer service, and maybe a bit dramatic, but I don’t want to hire 2 or 3 bodies to handle customer service. I still want to only do one, but with this bot, I’m going to be able to really expand their capabilities, maybe able to take an employee who maybe isn’t, English first. Maybe they struggle with how to respond to certain challenging conversations, like potentially like angry shoppers. And the bots can handle all of that tricky stuff for them.

Drew Hart 00:33:54  All they’ve really got to do is execute on it. So this interaction protocol, again, you guys will get the copies for this is essentially a menu system. And Kathy is actually going to ask your employee or your team member and say, hey, you know what? Like what are we about to work on? Is this going to require me to be sort of formal, or do you want it to be informal? Like, what are we going to do? And then that’s it. So you’re going to as you’re building out your bot, you’re going to build out the main purpose of the bot. Then you’re going to go create the interaction protocols, a secondary step. Again, you’re going to create that in a new, in a new window. And you’re going to take your instructions, which is I want to add an interactive menu and you’re going to say, what do you want to have in there? And then you’re going to do the prompt structure in the output formatting along with the mean bot edit into it.

Drew Hart 00:34:30  And it’s actually going to create the section. When you guys get this prompt structure, you’ll see that there’s a section designed in the prompt structure for an interaction protocol. Now we’re doing this for a customer service bot. It’s designed that way. If you don’t want an interaction protocol, you don’t want a menu driven system. Don’t put it in there. Just get rid of it. So next then we would then add in the brand invoice. And I would do that in a separate chat. And I would try to I would work with ChatGPT to describe what my brand voice is. And I would even ask ChatGPT like, I need you to be a brand voice expert and walk me through an exercise on how do I define and describe and build out a brand invoice. Like if you don’t know how to do that, GPT to be that little, you know, bot for you in just a simple chat and it’ll start to produce a brand voice and tone for you. You then add that in to your main bot as a section under speaking styles, which again it’ll come in the swipes that we we give you guys.

Drew Hart 00:35:16  And then finally, like if you’ve got any brand policies like, you know, for instance, like you may not offer returns, maybe your, your, your product is super expensive or super bulky and you can’t afford to do those types of things. Fine. What you would do then is put those policies into your final, custom GPT and just say, like whenever Kathy produces a response, it never does this or it always does this. And then those are your brand policies, so that Kathy doesn’t do things that you don’t want your customer service person to agree on or not agree on.

Josh Hadley 00:35:42  Makes sense.

Drew Hart 00:35:43  That’s a quick walkthrough. of sort of like what that process looks like. Josh, any any questions or do you want me to expand on this that this makes sense? Is there anything that you think I should cover further?

Josh Hadley 00:35:53  No, I think like conceptually, I think this makes sense. Right. And I think like you just got to start doing it, which I think is the magic. And drew, what I love is that you’ve already created like those swipes for people.

Josh Hadley 00:36:03  And look, this is one of those things that like, you’re going to get better and better at it over time. And so it’s one of those things, like the best way to get started on all of this is just simply to get started, right? Just go fumble around with it. And drew, you’ve already got these amazing swipe files for everybody.

Drew Hart 00:36:16  Yep. Yeah. And again like we we kind of set this up like this isn’t intended to be like an explicit like coding walkthrough on everything. This right now it’s a little bit abstract I know because you guys are getting this presented to you just kind of graphically and everything. Get the swipes, plug them into ChatGPT, start exploring them. You won’t struggle figuring this out once you kind of get started. Now I’m showing you like a six step process to go build a bot for customer service. I will just tell you guys for the most part, if you just did one and two and five, you’re fine. Like you’re going to get like almost everything that you want out of this thing just by inserting a job description, using the swipes and then saying, I kind of just want you to be casual with your speaking style or formal, because otherwise, like GPT can just sound like an enterprise company and you’re just like, this isn’t good.

Drew Hart 00:36:58  So change its voice and tone. So that tends to be like one big thing.

Josh Hadley 00:37:01  Awesome.

Drew Hart 00:37:03  Okay, so some approaches to prompt engineering. So you guys literally are looking at like how prompt engineering works. it gets more advanced than this. You’re getting the under the hood view of this. folks that put together agents, like, like, large language agents that go do, like, really sophisticated things. They do the same thing that you’re seeing here, but they break down the steps into little tiny components, and then they throw them into tools like make or integrate, task aid or in A-10 or any of those other tools. But fundamentally they are interacting with ChatGPT in a similar design style that you’re seeing here. So even when you get sophisticated and you start looking at like, well, how can I compartmentalize and automate some of these interactions, this is how you do it. So because you’re going to be working directly with ChatGPT and actually talking to it, I put together a couple of best practices for prompt engineering.

Drew Hart 00:37:47  We already talked about the first one, which is like one feature at a time. The concept here is that, look, if you’re going to have a really lengthy discussion about building out a bot or like interacting with ChatGPT, it can get lost over time. So the idea about one feature at a time is to compartmentalize your conversations with ChatGPT so that you’re really just focused on like, what are you trying to accomplish with extracting this feature or building a feature out? Pull that out, integrate it into your main bot, and then move on to the next one. Clear, simple and specific language is really, really critical. We kind of touched on this again earlier where I use the word should ChatGPT needs to be told what to do. It’s like a five year old. Like it benefits greatly from really direct and clear conversations that it knows, like where its boundaries are. So differences in things like, like words like always, all, any, never. Those are absolutes. ChatGPT does really good with those.

Drew Hart 00:38:35  Like, if you if you never wanted to do something, then use the word never or the inversion, right? Never do X or say always don’t do Y or always don’t do X. You know something like that. It’s a bad example, but you guys get what I mean. you’re going to need to test and iterate. So I didn’t cover this. as you’re moving through each of these stages, you want to test your bot out? I’ve got it down here at the bottom. It says test. At each stage. Don’t just move through these steps like go like when you create your main service customer service bot. Go throw go pull 4 or 5 actual like customer service questions that come from real humans. throw it at the bot and see how it’s performing, and if it’s directionally where you want it to go, then you can proceed. Otherwise you got to make adjustments to it. specifying the output format. I touched on this before. I highly recommend you can choose your own formatting or your own structure.

Drew Hart 00:39:18  Like you can just take what we give you and then you guys can run with it. But I would highly recommend that you stick to a consistent output format. Again, if you’re building this for your business and it’s your first bot, you won’t see the value. But when you build your 20th or your 30th bot, you’re going to get a lot of value out of always having the same output format, because you’re gonna be able to copy and paste between your bots and that will that really will come into play. role playing, just, you know, giving this is a pretty common one. I think most people know about this if they’re even remotely into LMS, but you want to give it a role. So like you’ll see in the swipes, you are a customer service agent for a, you know, multinational brand that focuses in, in around industrial products. It needs some context on like what its role is going to be. And then it benefits greatly from examples. And they have a bunch of fancy language for what this is called like zero shot, one shot, multi shot, two shot, things like that if you provide it.

Drew Hart 00:40:06  Two examples, it’s called a two shot. If you provide one example, it’s a one shot. If you don’t give it any examples, it’s called zero shot. but it’s a handy little term to to know about. but if you can give your customer service bot or whatever bot you’re building a couple of examples of what you want it to look like in terms of output or voice and tone or style, that’s going to help. That’s greatly going to help ChatGPT understand. Exactly. Kind of like, what are you looking for? Okay. Any questions on that? Those are all kind of.

Josh Hadley 00:40:30  Those are good tips.

Drew Hart 00:40:32  These advanced techniques I would highly recommend you go look these up. Asked GPT about them. Literally you can use these words. Noun phrase consistency is one of the most important ones though. So when you build advanced bots, I don’t consider this customer service bot to be very advanced. but when you do like a strategy bot or you’re doing a bot just to your point, like how do I optimize a container? You will want noun phrase consistency throughout your prompt instructions.

Drew Hart 00:40:53  Noun phrase consistency. You can actually ask ChatGPT to identify it. Like like you can use ChatGPT to clean all this up. When you get noun phrase consistency, you are removing the ambiguity of what words you chose to instruct this bot, and you are making them now consistent. So it’s a fancy way of saying, when you write your prompt, you probably aren’t going to use the same language in the top part versus the bottom part of some of your instructions. And ChatGPT is going to interpret the whole thing, and you won’t keep track of it. You won’t even necessarily know that you’re talking about or using a noun phrase differently in one context, and another one within your document. When, when you get closer to finalizing your prompt, ask ChatGPT to analyze and then refine and make recommendations for noun phrase consistency. And what that’s going to do is now when you’re talking about like a shopper, that you’re referring to a shopper in a consistent way all throughout your instructions, it’s just a way to make things really clean.

Drew Hart 00:41:45  Yeah. first principles. So this is a really interesting thing that you can do. You can actually just ask using a first principles approach, you know, help me understand, like, how I would break down a process to build a bot to do XYZ and ChatGPT will help you with the design process for how do you actually construct a bot? It’ll actually tell you like, here’s the components that I think that you would want to consider. And if you tell it to do it from a first principles perspective, it’ll get really down to the core aspects of what you want to get to quickly. And the last thing is chain of thought. And this chain of thought is like what some of these more advanced models are doing, like the one O and the three O models. But you can actually ask your own bots to do this. We do this. We have a strategy bot, and we actually have it use a Socratic method and chain of thought to then go and analyze its own output, create its own questions, answer its own questions, and then finally give me the thing that I’m looking for.

Drew Hart 00:42:39  So chain of thought is like a really powerful thing. I won’t get further into this detail, but basically what you’re instructing ChatGPT to do with something like this is tell me how you got there. Like, take me through all that and you. This is cool because you can use it for something like a strategy bot, right? Because the strategy is like, okay, well, if I was gonna approach this problem, like I would approach it by going A, then B, then C, then and you’re like, oh, that’s interesting. That’s going to get you thinking after you kind of read what ChatGPT is telling you. But the other thing is, if you use chain of thought, like in some of the design aspects of like say, a customer service agent, you’re going to actually get some insights into how ChatGPT comes up with its answers. And that’s going to be a feedback loop for you as the human or the prompt engineer or designer to actually learn. How is GPT even interpreting my instructions in the first place? And what does it even considering as it produces its final output for me? In other words, it lets you look under the hood.

Drew Hart 00:43:28  So those are some really good tips. These are the main ones. There’s a ton of them. There’s like 17 other things that you can do that are like really mind blowing things, but they get complicated. And I don’t think that they’re for most use cases.

Josh Hadley 00:43:37  Yeah. Drew this has been super impactful in as we’re running up on time. Is there anything else that you need to share that we haven’t shared yet?

Drew Hart 00:43:44  Not that I can think of, but you know, I’m, you know, I’m busy and stuff like everybody and like Josh was talking about, I’m really focused on deep work. But, you know, if this is something that you’re really working on and you’re interested in talking about, we don’t do this as a service like Josh introduced us like we’re an Amazon agency. We’re not a ChatGPT service. This is just what we’re doing. So if you guys get value out of enough, that’s great. But if you’re working on a really challenging problem and you’ve got like a legit business use case, I’d be interested in helping with those.

Drew Hart 00:44:05  if you’re working on trying to, like, figure out the customer service agent and stuff like that, look, you’re welcome to message me. I may struggle to get back to you with time, but if you’ve got a really challenging business case that you’re working on, I personally would probably be interested in helping somebody work on something like that.

Josh Hadley 00:44:18  Awesome. Drew, I love this. And you have the swipe files we’re going to we’re going to leave this as the cliffhanger for the very end, right? I think you have a QR code to share. Yeah.

Drew Hart 00:44:26  I’m not going to share. I’m gonna let you share it because it’s not wired up yet to the folder that I don’t know, like where.

Josh Hadley 00:44:30  You can share. You can go ahead and share it now because you’ll be able to set this up later.

Drew Hart 00:44:34  Okay.

Josh Hadley 00:44:35  So let’s let’s share the QR code where people can go get these swipe files. And drew, by the time this thing goes live and anybody seeing them you’ll be able to.

Drew Hart 00:44:42  What I’m saying Josh though, is that, this QR code is not wired up to the right folder like that QR codes wire to my calendar link.

Josh Hadley 00:44:50  okay. All right. Well, we can, what we’ll do then put it in the show notes. Do you want to have a link or if you want to just have a link that you want people to go through to access that, that’d be perfect.

Drew Hart 00:45:01  Yeah. That’s perfect. I’ll give you guys a link.

Josh Hadley 00:45:03  Okay. All right. Well, that’s that’s easy enough. All right, drew, as we wrap it up, I’d love to leave the audience with three actionable takeaways from every episode. So here are my three actionable takeaways from this one. Number one, your most important job as the CEO and leader of a business is to be able to think strategically and critically, and being able to sit down. Number one is like, can you carve out enough time in your day and in your week to do deep work, because it is your job to cast the vision and the mission for your brand, and that’s what everybody else on the team needs to be able to follow. So I want to challenge everybody to drew.

Josh Hadley 00:45:33  I love that you have a strategy bot, right. And I would challenge everybody like create your own strategy bot because that can be your, you know, maybe copy and paste version of yourself that we all would love to clone ourselves if we could. Right. That’s the closest thing to it that you can kind of program that way. And that will allow you to think into the future. Do the 1080 ten principle action. Item number two is I would highly encourage each of you to take this podcast and send it to your team members, okay? Especially your leadership team. And I would have you challenge them to think through. Hey, how what is the biggest constraint that you face in your role? And how could we leverage the use of AI and a bot to help move you through those challenges more quickly? Okay. To be able to level up your output, my third and final action item is if you want to really take it up a notch, is do what drew said and create your own bot that then you give to your team a custom GPT that says, hey, use this.

Josh Hadley 00:46:27  And as you walk through this, it’s going to help walk you through mental models to help you identify what is your constraint in your role, to then be able to identify what type of bot you should be working on. So that’s going to be the challenge for myself that I will be working on before I roll this out to my team. But I do believe that this is one of those things that yes, it does require work upfront. This is not an easy button solution, but I do believe that this is one that will pay dividends consistently and your investment will compound month over month, year over year. drew, anything else to say on that note?

Drew Hart 00:46:56  No, I think, I mean, this is where the future is. so anybody who’s, like, interested in invest in it, invest the large language models themselves. There’ll be some there’ll be some tech changes here in the next, like two, three, 4 or 5 years. People will start automating them and making it easier for them to interact with.

Drew Hart 00:47:09  But if you understand these fundamental principles, what you do through these exercises, you’re going to have a huge leg up in understanding, like the inner workings of these models. That’s what you’re going to get from this. It’s going to make your innovation even that much more innovative.

Josh Hadley 00:47:19  I love that. drew, final three questions for you. What’s been the most influential book that you’ve read and why?

Drew Hart 00:47:25  so I’m reading these books. The 100 million offers, 100 million leads. awesome book. If anybody is reading that, that offers book is by far like for econ people. Look, you’re the leads. One’s interesting for like how you can use it with the customers and how to spike sales and things. But the offers one is the one that I think is really interesting because Alex is so adamant about value being value oriented. Value stacking is the phrase that he uses. the light bulb has gone off for us on this, and everywhere. You’re adding value with what it is that you’re delivering, whether that’s a product or a service, that that’s that book has been really, really.

Drew Hart 00:47:56  And he writes like a kindergartner. So like you’re going to understand like what he’s talking about.

Josh Hadley 00:48:00  So drew, on that note, how does an Amazon seller better enhance their offer on Amazon itself? What have been the ideas that have come to your mind that way through that? Well, it.

Drew Hart 00:48:08  Turns out yeah. No, we have a service for it. We literally have a service for listing optimization, which is a really boring term. Everyone thinks that that’s creative work and it is getting amazing. Stunning visuals is super critical. But it’s, we talked about this before the call. It’s about dialing in those emotional connection points. Like what are the emotional triggers that you can identify. And I think there’s just a ton of sellers, us and non us that just missed the mark. They’re really focused so much on facts and figures. That’s like that’s the table stakes information. But like how are you actually affecting someone’s life. What problems are you actually solving. this unique selling proposition super important. But like how does it affect me as a buyer? Like, what are you what are you helping me accomplish with your product? And that stuff gets ignored if they don’t show up on listings or the one the listings that do show up, they stand out like sellers know it too.

Drew Hart 00:48:48  Yeah.

Josh Hadley 00:48:49  Love that. Great takeaway. I 100% echo that as well. All right. Question number two. What’s your favorite AI tool or software tool ChatGPT prompt that you’ve been.

Drew Hart 00:48:57  Used to because we’re just talking about that. yeah, of course. ChatGPT is fantastic. It’s amazing. there’s a lot of discussion around deep seek and everything, but, my take on the deep seek stuff is that it’s the same tool as ChatGPT. Their infrastructure just happens to be less expensive to operate, but it’s not a different service per se to the user. But a really cool tool that we’re playing around with right now is called Gamma App. And there’s a couple of different services like this, but it’s like an auto presentation, kind of like document creator. It’s pretty badass. I was really impressed with that. But there’s like 4 or 5 other ones that we’re looking at. But that one’s really exciting. what else we’re really starting to mess around with, like, image creation and doing things like that. we have someone on our team that really, the images are way different than, like, what we were just talking about.

Drew Hart 00:49:36  But we’ve had someone who’s been really, digging into that and trying to understand, like, how to control the images better. I think the bottom line is like, go use the tools. Don’t try and do it directly with Midjourney or something, because it’s it’s so complicated to try and dial it in.

Josh Hadley 00:49:49  Yeah. What other? Great.

Drew Hart 00:49:51  They’re great. I mean, they’re just like. It’s amazing what you can do.

Josh Hadley 00:49:54  Yeah, I’ve heard, I’ve heard of gamma. Give me a couple other ones that you mentioned. Like we’re looking into, like, what are some other, like, just exciting ones we don’t have to have like, good use cases for them. Just throw out some interesting ones that have sparked.

Drew Hart 00:50:04  I mean, the automation side of it now. So like make an Integra. That’s what we use. That’s our main one. And it’s because we inherited it before even ChatGPT came out. Like we didn’t use Zapier. We use make. So now it makes got APIs to go into ChatGPT in a in is the next one that we’re working with, which is another automation platform that integrates with the large language models.

Drew Hart 00:50:20  and then task eight is the other one that I’m looking into. But I don’t think we’re going to pull away from make I think we’re too embedded in there. But some of these other tools, they’ve got some really special use cases that we might split some of our projects out into some of these other platforms. but those tools are just like continuing to evolve. The other ones like like going back to gamma. gamma does like document creation. So document creation is pretty generic. It’ll produce like slides. It’ll produce PDFs. but it’ll also produce sites like actual websites. And it does it like in a really nice way that lets you integrate it. So like you can use gamma, but you can use ChatGPT to go produce marketing content. You can have it all spot on with voice and tone. You can then throw that into gamma to go build a landing page for a special promotion. You’re running for some seasonal events and that workflow when you go play with it as an owner or something, or one of your like high powered team members, you let them go figure that out.

Drew Hart 00:51:07  You can create like a system in like we literally can create content that is like fully polished, on spot brand voice. the, the coloring, everything is like just really spot on. We can do it in like an hour. It’s insane. It’s insane. And then all we’re left to do, the 1080 ten rule is we’re left with like the 1 hour or 10 minute, 10% up front, ChatGPT gamma, go do the heavy lifting, and then we do 10% of spit polish and make it perfect.

Josh Hadley 00:51:34  Oh love that. Beautiful. All right Jrue final question. Who is somebody that you admire or respect the most in the e-commerce space that other people should be following and why?

Drew Hart 00:51:41  Yeah, we I mean, I follow a lot of this like all the same things that like you do. Right. Like you, you’re like, honestly, you, Kevin King, Brandon Young, like we talked about MDS is really great, but to be honest with you, like a lot, just because of the amount of time that we spend doing this, it’s our clients.

Drew Hart 00:51:54  We learn so much from the brands that we work with, they want and look for experiments and new ways to try to, like, come up with creative ways to grow or use data to make insights, to take action on those types of things. We’ve got brands that we work with that are that are smart. Frankly, I’m gonna call them challenging, but they’re not challenging in a negative way. They’re challenging and sort of like, how do we continue to stretch and innovate and grow and like push things further? and we learn a ton through those conversations.

Josh Hadley 00:52:16  Yeah. Awesome, I love that. True. If people want to reach out to you, they want to learn more about you. How can they do that?

Drew Hart 00:52:21  Yeah. That’s, actually, that’s is what that that’s what that QR code was all about here. they can contact me here. They can reach out at. Hello. I think thinking about that goes to me. if you want to book an appointment with me and learn more about what we do or talk some about some of your goals and challenges, we do free consults so they can book some time there.

Drew Hart 00:52:37  otherwise you can check out our website and our YouTube channel, which is, just about to start out.

Josh Hadley 00:52:41  Awesome. Sure. This has been a pleasure. Thanks for joining us and creating custom slides just for us.

Drew Hart 00:52:46  Sounds great. Thanks, Josh.