Jeremy Cabral on Scaling With AI Systems, Not Just ChatGPT

Jeremy0:00
Thanks for having me again. I'm loving all these BoA events and they're awesome. Yeah, guys, thanks for joining as well. Yeah, for me, I, I love talking about a lot of topics, but for me, one in particular has been really fascinating is AI and how you can use it to scale your company. Yeah, from my standpoint, I think it's still early in this whole journey. And so, yeah, there's a lot of learning to be done and a lot of sharing needed for all of us as businesses to really make sure we're using it correctly and to its best potential. So today I'm going to share some of my learnings. As I've been going around the traps over the last couple of years and seeing what's working, what's not. And yeah, hope you find it valuable. But firstly, just want to give you a bit of an overview. Let's get that working about myself in case you don't know me. So yeah, one of the co-founders of Finder. Finder is a financial comparison website here in Australia. It's used by over 35 million people around the world actually. So we're in the US, UK, and Canada. And yeah, honestly, it's Been a crazy journey. We bootstrapped all the way to $100 million plus in revenue, which is a pretty wild story. Happy to share more of that on a future one as well. But yeah, even raised $30 mil at a pretty big valuation at the end of December 2021. And I led the expansion into over 20 countries for Finder. So yeah, pretty massive journey. I was actually in a chief operating officer role with the business, ran product design, tech growth, and international expansion, but recently left my operational role. I'm still a special advisor and support the board. But yeah, really what I'm doing right now is advising businesses on AI growth strategy and operations. You know, for me, I, I felt the need to go all in on AI and just make sure that I was, you know, staying up to date and really maximising this new tech wave and bringing those learnings back into Finder as well as new companies as well. Now, it's a personal note, there's me and my fam on the right. Just ran a marathon August 31st this year. It was pretty wild. To get that done, I had to lose 20 kilos as well. So yeah, if you ever want to talk about marathon training and how brutal it is, I'm happy to talk about that as well. So yeah, let's get into it. Firstly, my deck design sucks on purpose. I really want it to be a bit jarring for you. I chose literally the black and white template in Google Slides. And the goal is to really have you focus on the substance, not pretty pictures. It's not a fancy thing here. I think if you really focus in, you'll hopefully get some value out of today's session. So yeah, and another thing is I know the title is Smarter Not Harder, but I think it's smarter than harder. AI really is not going to replace hard work. I think if you think that you can really copy-paste from ChatGPT and still have a job, you know, 2 years from now, I think you really need to have another, I guess, change your position on that. Yes, there's a lot of value in the short term and kind of what it can do, but hopefully from here you can really learn a lot more and how to better maximize, you know, many other tools beyond ChatGPT as well. Upfront, I wanted to ask for a commitment from all of you. You know, for me, it's, these are fundamentals and I know we're talking about systems and AI, but these are actually universally true and things that I think you need to have in place in order to have your company in a position where you're best maximising any system in your company, you know, traditional or AI. So first and foremost, you have to have some central way of keeping all your projects together. I think in the most basic sense, every company should start with a spreadsheet. Like if you're a solopreneur, this is where it's at. Literally have a spreadsheet. You can't keep all your stuff scattered across messages and emails and all this sort of thing. You have to bring it all together into one core system. And beyond that, as I said, it's really important to have all your documents living in one place as well. It's one of the habits I think is that often email can become your source of truth. And I think that's a little bit challenging as a foundation to properly be able to scale. You need to push all of that context into a system that has, you know, all of your documents in there that can be editable. You can contribute to it. So that's really critical. Um, the other one, it's, it's a really important one and it's a bit uncomfortable at first, but every single meeting, meeting you participate in, ideally you are transcribing that. And you have to ask permission for that and just, you know, make it normal, make it like, hey guys, about to, you know, jump into this call. Do you mind if we transcribe? It just makes it easier for me to follow up after, after this. And it's really powerful in ways that I'll explain later. Um, again, this is an odd suggestion to talk about hiring a virtual assistant. We're talking about AI and systems today, but I think the key is that if you have a virtual assistant, no matter the size of your company, you have— you're basically forced to start delegating and learning how to you know, build things into processes that can be repeatable and ultimately automated in the end. And again, there's no silver bullets here. It's not about, you know, shortcuts. It's really about rolling up your sleeves and learning. And hopefully I'll learn from you guys one day as well. So yeah, I actually, yeah, I did use AI to create this diagram. So it's a bit average. But I wanted to say something, which I think is really interesting. When people talk about using AI, right now what's coming up in conversation is most people are using ChatGPT. Does Yes, it's AI-powered and it's a GenAI tool, but it's not the end of your chat— it's of your AI journey. You really must go through this evolution over time where you start off with something as simple as ChatGPT and start evolving into more complex systems, which we'll talk about today as well. And again, on the piece of foundations, I think the stack that you work on really, really matters. For me, this is my stack, and the reason why I've chosen most of the tools on this list is that they're built with AI in mind. So they either are tools that that were, you know, came post, you know, AI becoming a bigger thing or are foundational and universally able to be connected into all of these AI tools. So yeah, across the board there, you know, maybe it's an uncomfortable one to say use ChatGPT Atlas. It's a brand new release browser. But the reason why I suggest that one is that you need to learn about where technology is going. And it's a really good thing if you can start using technology early and start, you know, leaning into the friction of, you know, actually switching to these newer tech, softwares that are out there. And I think you'll find that, um, it becomes more native to you and you'll be able to adapt and be more quickly moving as a company, uh, by doing that. Yeah, so across the board, there's a whole bunch of things here. I think the main call out is that I prefer Google over Microsoft's 365, mainly because it's a better base across integrations. Um, but yeah, if you are on Office 365, you can still use AI. I just feel it's a little bit slower to get integrated into a lot of the tools that I use. So yeah, across the board here, um, whole bunch of stuff that you can cheque out. But yeah, let's get into it. So yeah, as I said, I think the most critical thing you can do is get a really well-designed spreadsheet for your company. This is actually one that I built in 2009 for Finder. It still exists. And what I did was, you know, I realised a couple of weeks into my company that it was going to be near impossible to achieve our ambition without having some level of system and way of building processes that could ultimately be delegated to people to execute on my behalf. In that first few weeks, I was writing every single article. I was, you know, making every single edit, changing all the rates and information on the website. Literally, it was insane. And I realised that in order to scale, I needed to first, you know, find someone that I could work with and very clearly delegate each task. And I put all of that into a spreadsheet, and that was my core system that I ran All The Finder out of. From that, I, my natural way of, I guess, documenting things is to demonstrate live to someone what I mean from a process. So I was using, I'm not sure if it was Loom back in the day, but it's certainly some Skype video recording thing. And I use a tool called Granola. It's really epic for transcribing. It basically, the main reason why I really like Granola is that firstly, it's on your mobile and on your desktop. But you actually, after you get transcriptions, can adapt those transcriptions to be whatever you need, including a process document or a product requirement document. So it's a really effective tool. And I grab that information, then I can, if I, you know, I think Granola has its limitations as well. And so I like getting the transcripts or that final output of a document, tweaking it further in ChatGPT, turning it into a Google Doc, which becomes, you know, a process document, adding to the spreadsheet where your team or your future team can start looking at. And I think that, you know, with everything I'm doing, it's really about building with the end in mind. I want to have systems that I know that the moment I can hire, you know, if I can't afford to hire someone today, that's, you know, maybe that's, that's not critical. But I think if you're building the foundations, it makes it easy for you to move fast when you need to. So yeah, um, as I said, you know, when we kicked off Finder, I remember I was in my bedroom on a call with Fran and Frank, and we were talking about all of the roles inside the business, and pretty much it was just our names across all the boxes. And, um, but we were very specific in the way we designed things. We wanted to see, you know, various functions and function leads. And I think that's really critical, even if you're a of one. And there's a really key book that I read on this called E-Myth, which goes into the, the various ways you can delegate inside your business. Yeah, I think starting off with yourself across all the boxes, but, you know, really clear in your org design. And then looking at that human team, as I mentioned, of like, which are the critical roles you'd hire next. And then the human and the AI team. So that's, in my view, where you start bringing that support to really help scale each individual and help bring in automation to make their jobs easier. So I don't know about you guys, but when I've written job descriptions in the past, I think we write them once, they get attached to a contract, they get signed, and they never get looked at again. Um, but they are really critical. And I think that, you know, companies as they scale, or individuals after they come on board, it's easy for them to kind of forget what their role's about. And, you know, it kind of adapts over time. And, you know, I think that's partly why companies become chaotic. The interesting thing is to use AI correctly, you actually need to write job descriptions for AI agents. And so getting that skill right in the beginning for a human job description, but keeping it constant and always going for an AI agent, are really critical. An AI agent doesn't forget. An AI agent sticks to the rules most of the time. Um, so yeah, for me, I think building that skill of running a clear job description of what you're wanting to achieve with that role, and as specialist and specific as possible, is critical. That's going to get you the best results. And on that point again, I think in building a standard operating procedure, I think there's critical steps which have to be, you know, the way I kind of in my mind think about it is if I were a robot and could not complete a task unless it was specified, how would I write that up? And literally step by step writing this up is key. The great thing is there's actually some pretty cool tools that are being developed to make it easier to produce these documents. You know, obviously Lumen and Granola, like I mentioned before, are really key, but there's others that can look at what's happening on your screen and actually start developing screenshots and material to support those standard operating procedures as well. So keeping that flow constantly going for everything you do in your business, putting it back into the index. And, you know, this is how you break down that job description and I guess the operating procedure there for an AI agent as well there on the right. So seeing them, the traditional thing on the left and the, I guess, where we're at now with AI on the right. So let's get into the basics. So yeah, I think, like I mentioned, ChatGPT, obviously AI-powered, is fantastic. The thing that I see most people doing is a zero-shot prompt, literally writing one sentence, typing it in, submitting it, and ultimately getting a pretty, a very generic response from ChatGPT. And I think that's a big problem because, like, firstly, it's very noticeable. Like, your customers are going to be fully aware of that and see that you put in zero effort. So you really need to get better at prompting and you know, there's a whole thing called prompt engineering, but you know, even if you just took away this one tip today, I think if you prompted better by asking ChatGPT for the prompts to put into itself, that's a really clever way to get a better result. So you effectively ask ChatGPT to be a prompt engineer and it spits out a prompt that you paste into a new chat and it takes it away from that generic output to something that's crafted, like specifically for what you're trying to solve for in your company. So really cool tip there. The second is, if you've got context around your company, things like pricing, product lists, you know, frequently asked questions, whatever it might be, that information, it doesn't necessarily have to be public on your website, but if it's information that you're constantly referring to when answering a client or developing a document or whatever it might be, putting that into a project file will ensure that ChatGPT can draw from that knowledge and isn't just giving you a generic response. You know, this is an example here. I, uh, there's about another 7 documents there, but when I ran my marathon, I had, I think, yeah, this huge amount of documents and actually prompted ChatGPT 3,049 times in the preparation and, and actually during the race as well, which is pretty crazy. But the quality of the response was really key. And I think, um, it's important to have that in place. And lastly, um, once you've got those project files in place and you're using solid prompts, the, the model you choose really matters. By default, ChatGPT is going to have it on automatic and that you know, you've got to think about OpenAI and how much cash burn they've got. Effectively, they don't always want to give you the best model, to answer your question. So if you put it on Thinking, most of the time that's going to get you a better result. Pro can be used for certain things if you've got a Pro account as well, but I think Thinking tends to be the best. And now there's some new features where you can, if you need a fast response, you can just turn that off midway and you'll get a faster response. So yeah, try model selection and seeing the different results you get with each model. And I think you'll get much more out of ChatGPT and Claude as well. So beyond that, this is the next level. So that what we just showed was basic ChatGPT and Claude usage. What's happening now, like with that tool stack that I mentioned, there's a whole bunch that are pre-integrated with these tools. They're called connectors. And in those connectors, essentially you add that application in and it's, it's, it's not a technical process. You literally authenticate it. You're logged in. And then you can interact with that application through that interface. And I think that's really cool. You can do it to create documents. You can do it to, you know, create a Canva design or whatever it might be. Each app is very unique and different. Um, but yeah, from my standpoint, it really starts introducing that scale and using AI into your company. And, you know, there is going to be a whole marketplace emerging here around ChatGPT apps, and it's going to get pretty crazy. Um, they just got announced a few weeks ago. But from here, I think if you can start by, you know, Assessing what tools are available in the tools that you're using every day, ChatGPT and Claude, you'd be surprised at what you can do just by interacting with it instead of through the standard interface, but through the chat interface that you're using every day. There's something as well called Model Context Protocol, MCP. So if a connector doesn't exist officially, these are effectively ways to interact with that software. And it's the way I describe it is like it will, it's like an agreed protocol for a system and an AI agent to interact with it. So if you have some information inside, say, a CRM, and there's an MCP for that particular CRM, you can actually query it and it'll go and get the information on your behalf agentically. It's not perfect all the time, but it's pretty powerful. There's a screenshot here actually showing an MCP for WhatsApp. So you can, through Claude, actually write WhatsApp messages and reply to people. It's pretty dangerous as well. You gotta be careful. But yeah, it's, it's a, incredible technology, and I think as things evolve, it's going to get more and more powerful. Yeah, and again, the model selection, I think, is really, really critical. So once you've gotten through and you've mastered those first two steps, I think from there you're going to start noticing things that you're repeatedly doing in your company, things that I think you shouldn't start from scratch from every single time. And so the first thing is to try and like My view is like, instead of trying to build something from scratch, you should try and borrow somebody else's time that's been invested to build something really cool that solves your business problem. There's often really cool free things and sometimes some paid solutions. So yeah, cheque out these marketplaces connected to these apps. ChatGPT has a GPT marketplace. And effectively what that is, is, you know, when you, when you prompt ChatGPT, you're training it each time on, I guess, you know, in the context of the conversation, what you're looking for and what you need it to do for you. These are pre-prompted, pre-tuned for very specific tasks. So that's the useful thing here. If you can, you know, plug in a GPT, you'll be able to use it to complete something like, you know, it's tuned for writing, it's tuned for, you know, generating images or whatever it might be. And so instead of kind of having this blank canvas every single time you start ChatGPT, it's pre-solving some of those repeated tasks for your company. The other cool thing is you can actually build your own GPTs. So taking an approach like you've done in a normal ChatGPT conversation and actually, you know, going back and forth saying, here's how I want it to look, here's how I want it to work, and so on. You can build these GPTs that solve repeated tasks. There's one I built called a warm intro writer. So I introduced people daily, and I found that it was getting time-consuming. And so what I was, you know, wanting to do was ensure that I could mention two names very quickly with a voice prompt, say, hey, introduce these two people, um, here's the context. And that's literally it. And then it spits out a perfectly crafted email or WhatsApp intro. I actually just used it before I came into this session. I asked the guys, what do you think? And they said it was 10 out of 10. So if you want a link, uh, you know, copy this of this, you know, you can DM me on LinkedIn or send me an email But I'm looking for beta testers, but it's a pretty cool tool. And then you've got Artifacts, which is Claude's version of something that's like a bridge between, I guess, your conversations and creating apps automatically for you. There's tools like Lovable and some of these other things which are, you know, designed for building apps from scratch, whereas Claude's identifying these things from your conversation that it can immediately turn into an app for you. So cheque it out. Really cool. And something you need to play with for sure. Okay, so firstly, I just want to differentiate a couple of things. So people are often describing everything they do in AI as an AI agent, and there's kind of two types of ways of leveraging some of these things. So there's AI workflow builders, and that's most commonly what people are actually referring to when they say AI agent. However, there are steps in these that are agentic. So what that means is effectively completing a task without direct specific instruction and it's got like a broad description of what it needs to do. And that's more an agentic flow. And then you've got the workflow builders, which are actually deterministic. So they're step by step connecting, you know, each of the steps on your, and you define it like you might with, you know, a tool like Zapier, which is a pretty old tool nowadays, probably a decade old. Um, or it's more like an if, sorry, an if that, then this, um, solution for getting something done. So yeah, AI workflow builders is where I'd begin versus going straight to an AI agent where you totally hand the keys off and it gets done for you automatically. I think you're going to get some pretty low quality output from most AI agents. I think they're still emerging and I'll go through a few that are actually working. But yeah, once you're inside these platforms, whether it be Relevance AI, Relay App, or N8N Make, they actually have these templates that are pre-designed by the company itself. And the reason why they do that is they want to ensure that your first experience is high quality. So I think trying the workflows built by the actual software provider is key. Um, there's an example here of, you know, Relevance AI with a multi-platform workforce. So they're pre-integrated with things like Gmail, GSheets, Notion, and so on. And when you're chatting to it, it's done all the connectors already and it's, it's more tuned to a better result than, you know, building one from scratch. So you can see there's 749 users of that particular agent and you can clone it very easily and give it a go for your company. Inside these marketplaces as well, there's actually third-party built workflows. And why I like that is If you've paid for something, genuine, what it means is that person's incentivized to maintain it because you'll stop paying for it suddenly. So, you know, in my view, try these out. They might be actually a lot better than some of the default ones you're seeing. You know, I think one of the things that's happening right now is everybody's commenting on LinkedIn. It's like, comment this for some crazy workflow that looks magical. Most of those don't work, or if they do, they work, but it's low quality. So in my view, be, be very careful in what you're implementing into your company. Um, it's going to lead to potentially terrible results if you go too fast too soon. And then lastly, I think it's not as complex as you think. A lot of these tools since being released have launched a copilot, which effectively allows you to build your own workflows by describing it like, you know, by prompting it in a conversation saying, hey, I need to build a workflow that does this thing. I want it to connect with Gmail, Google Sheets, so on and so on. And it'll actually by default connect it all up for you, or at least prepare the integrations and then you start authenticating one by one and you're able to get these workflows up and running for your company. So definitely have a go. I think it's really important to start learning and rolling up your sleeves. So yeah, as I mentioned, there's some AI agents that are actually working. You know, I've spoken to a whole bunch of these companies directly, been in, you know, sales calls, you know, spoken to staff, you know, customers and so on, and really triangulating around what actually works. And the first that I've seen work is, and the reason why I know it works is a lot of these companies using it for their own staff themselves, is sales research is extremely powerful and an agent that I think everybody could leverage. What does it do? So you basically, you have your calendar and you have exactly who's connected and it can automatically see who you're meeting with and tell you a little bit about them. And you can actually, you know, define exactly what you want to know, but effectively it does research, tells you about the company, tells you about the individual and gives you these talking points, which if your calendar's full, you may not have time to do this sort of work. So it might take you a couple of hours normally, and this agent will literally run, prepare it for you, send you a message on maybe WhatsApp or Slack or wherever your core operating system is. So you go into the meeting, you quickly have a brief prior to that and it's really effective. The second is for customer service. So this has been in place for, I guess, quite some time in the sense that customer service automation has existed for a while. There's like knowledge bases and FAQs and things like that, which made it easy for some of these automations to, to answer questions. But the difference now with generative AI is that you can be more effective in kind of, I guess, answering a question going beyond what's specifically provided. So again, you know, it's all about the knowledge base. You've got to add all the information that you can that you're willing to to, I guess, make publicly available. You can also build rules around things that the agent can refer to but not publicly reference. But yeah, it's pretty powerful. The one on the screen here is from a company called Intercom. They've got a subsidiary brand called Fin— Fin, I think it's called. Fin.ai is the website. And effectively, it's become a very accurate customer service agent, and it's one that's worth checking out. It's a little bit expensive relative to others, but if you don't want to have the risk of getting it wrong, like, for your customers, in my view, definitely install this, give it a go, and you'll find things like your conversion rate on your website improves and your response time rapidly improves as well, which I hope leads to happier customers for you guys. The last one is around data enrichment scraping. So this is, I guess, a broad thing. You know, the first one there, the Sales Research Agent, does a little bit of this, but my, my thing here is that ultimately you've got systems that need to be maintained. So whether it be your spreadsheet, your CRM, or whatever you're using as a core from what I shared before. These agents will go and get information and go and update your system on your behalf. And more than just using a pre-built API, they're actually pretty clever and they can actually scrape things. So scraping would be, say, as an example, like Google doesn't have an official API for its search and all the tools that leverage Google search information are actually scraping Google. So what this will do is go to Google, you know, run a search on your behalf if you want, grab that information, find competitors similar to the company, all this sort of stuff, and actually include it into your CRM. And that becomes context for you to build additional agents in the future around a particular customer. The one here is actually a leads hunter. So what is happening with this one is like when you see a competitor that has had people comment on their post or engage with their post, you can actually scrape those people from a competitor's post, enrich those profiles and know a little bit about those companies and then add the information to your CRM. So, you know, I guess what it does is you know that these people are interested. So you've got this high intent, I guess, um, towards what your product does and your services that you sell. So you're not coming in cold. You're like, hey, I saw you engage with some sort of tool that solves a particular thing recently, or another company that does this. We actually offer something similar. Did you want to jump on a call? So it's really effective, and an AI agent that actually works. So yeah, um, the other thing I mentioned before is around custom apps. So vibe coding became a term, I think, coined back in March this year. I think it was like a head of AI or someone like that at Tesla, you know, coined the term. But essentially what it is, is using ChatGPT-like prompts into a tool, um, into an interface, and it'll actually develop a piece of software that's fully working, uh, for you on your behalf. It's pretty wild. Um, the tools that I find easiest and quickest to get off the ground are Replit and Lovable. Why I like those is that they can host it automatically for you at a URL. So once you've built it, you jump in and you can actually start using it. So here's an example one of what I called a PAA extractor. So I don't know if you guys have seen in Google, it says People Also Ask. So what it does is it basically goes into Google and that section where it says the common, I think it's 5 questions listed normally. It'll scrape out those questions, go and attempt to answer them so you can automatically build a frequently asked questions block for your website. Why that's cool is that information then potentially gets surfaced again in Google or even ChatGPT in future. So I think this whole thing took me about 30 minutes to build and it was hosted and ready to go. It did have its limitations in terms of scraping, so I had to kind of build a proxy and things like that into it to make it, I guess, not get blocked. But yeah, really, really effective. You can build these little mini apps using Replit and Lovable. And, you know, once I got a little bit more confident, I started looking at my OpEx and going, you know what, we're paying for this software, which I think doesn't make sense. Some of these things that cost me $800 a year, kind of frustrates me that we pay this money for the software that's so basic. An example of that was a farewell board company. I won't mention them, but essentially I was like frustrated. I was actually in a meeting, I was like, I'm pretty confident we can kill this thing. Um, and then I think it was about, yeah, 29 minutes in, I was like, hey, here we go, here's a solution. And it literally, it was a totally customised finder farewell for our employees. Um, a board where you can upload photos and say, hey, great to have worked together, wish you all the best. Um, and then it's downloadable for the employee as well. So it's pretty epic. Um, and you'll be surprised what you can achieve. So have a go. Um, you know, but in my view, like on this last point, I think these are meant to solve very specific things. If you're looking to build your entire business on it, I think you've got to be careful. I've really heard some horror storeys where people might have spent, you know, days and days working on something and then did something wrong and they chose the wrong tool that they built it in. Or, you know, so they use Claude Code instead of, you know, because everyone said it was great instead of, you know, Replit or something where you can roll back. And they broke their entire app and they had to start again. So they lost days of time. And I think that's really, really risky. Also, um, there's some fundamentals which need to be in place if you're vibecoding and building an app that's customer-facing, especially if it's going to have, um, sensitive data inside it. So be careful. Um, if you need help, like, you should really research it and pair up with someone that can really take it to the next level for you. But I think it's really great for prototyping what you can do in an app and, um, looking at ways of saving money and making yourself more efficient. So where I, where I got to in the end, it was like, if I was to take everything I've just spoken about, I think that, um, ultimately it's a system of capturing information and centrally storing it. And so the first thing is building that habit continuously to be gathering that information and, and storing it in a way that's usable across your various AI platforms. I use Loom for screen recording. And it also produces transcripts as well. And I use Granola for transcription. There is another tool called Otter, and the reason why I would use that one for transcription is if you wanted to do speaker tagging. So how that works is you grab a transcript and you upload it, and that transcript could be from anything like a YouTube video. You're probably not allowed to, but there are tools to get that, other videos and so on. You pull that down and you can pop it into the tool and it'll basically start tagging these speaker names. So the next time you're in a meeting, there's like 7 people in there. If you've pre-tagged that and everybody knew there was a transcription that context will be stored with that person's name against what they said. And that becomes content that can be used in all sorts of things down the line. You know, I actually have used some of this stuff in kind of analysing and giving feedback on myself and how I turned up in a meeting and like going, hey, in this meeting, how could I have done better? How could I have, you know, reframed something here? In this, you know, about 30 minutes in, I had this issue where, you know, I did something I wasn't happy with. How would you better approach it? And so, like, when you work with person X, I think this is what I found. You know, here's how to better approach your next meeting. So it's pretty wild what you can do. And again, I used ChatGPT in that scenario to build a prompt template. In the prompt, I actually said, use this system prompt, and I gave the documentation from the official OpenAI website on how to approach that. And it became incredible what it could achieve in terms of insights for myself and from a personal development perspective. And also, you know, I spoke to others and said, hey, we're in this meeting together. If you're interested, I did this analysis for myself. Health, I can provide you this as well. So really powerful. I think capturing that information centrally is the first step to getting to that scale inside your business. Secondly, I think it's about documenting all of this. So once you've captured it, you need to store it somewhere that can be referenced from all of your AI. So I like to start with basic systems that, you know, kind of, the way I think about it is almost like a skill tree or a technology tree. You know, where you start off with these fundamentals and eventually you keep adding and improving. But if you, you have Google Workspace and I highly recommend that, um, you've got Google Sheets out of the box, Google Docs already And I think that these are going to be key tools to be able to start sharing your information and collaborating better with other, with humans and also with AI. So from my standpoint, bringing it all together is critical. There are, I guess, fancier versions of each of these. So Notion is, I guess, it's like turning a document to have database elements inside it and similar Airtable as well. So a cell in Airtable, for example, can have logic and it can have pre-integrations into a particular cell that runs logic and does the next step for you. So it's a pretty powerful tool. It's a bit of a learning curve. So I wouldn't suggest to start with Airtable, but if you use Airtable, it is pre-integrated with a whole bunch of these AI tools as well. So AI workflow, workflow builders often use Airtable as a database pseudo spreadsheet solution. The cool thing about it is if you want a human-in-the-loop solution where you're actually going to that Airtable sheet and able to effectively approve steps before they do the next step. I think it's really powerful because it's kind of a traditional UI that you're used to, you know, spreadsheet columns and rows, and you can literally use a dropdown and, and change it from, you know, one status to, you know, say approved or whatever it is, and it'll do the next step. So extremely powerful. Similarly here, I've spoken about on the automation side, I think what it's about ultimately is configuring once and, and being able to execute forever. You know, it's interesting. I'll tell a storey about the virtual assistant. So when I had built those systems in 2009, Naomi Jara was her name. She was running those processes for years afterwards and actually had forgotten about a bunch of them running. And so I went into the spreadsheet, you know, about 9 months later. I was like, whoa, still happening. And this is pretty wild. I think that's when you've done some of the steps I've talked about previously in here, like, you know, in a more traditional sense. The same kind of principles can apply to an AI system. So from my standpoint, I think, yeah, so I saw recently as well, some Zaps I built in 2015 are still running. It's like, wow, I just forgot that I even Zapier was actually powering all of this. Yeah, so Zapier is a tool that's been around for, must be a decade, but what it does is if that, then this style automation. So I think starting there is key so you can understand how the blocks come together and then taking a leap to exploring how AI agents can do that task is the next step. You know, with Relevance AI, they've got a really key system. They're an Aussie-owned company over in Surry Hills, um, and I think that, you know, the team are really cool over there. But if you can get in and, you know, leverage some of those pre-built templates, use their Copilot, you'll be surprised what you could achieve in probably 30 minutes to really get a process up and running, something automated up and going. N8N is a bit of a lift. I don't know if you've seen those kind of viral posts on LinkedIn. It's normally an N8N diagram of the spider web, all these things. And I that, yeah, the learning curve there is a bit more intense and I have a feeling that a lot of this stuff is going to disappear at some point. I think it's complicated and, and I think you should learn it because you need to understand the logic, but already Claude is allowing, allowing you to build N8N workflows by prompting it. So it gives you a JSON file that you can export out and import into N8N. And that's the shortcut instead of starting literally with one node and configure it one by one. Claude effectively becomes like a copilot to N8N and allows you to really, I guess, get to a result much more quickly. And again, on the AI app front, I think Repluss, you know, in my view, one of the best. If you're interested in e-com, Lovable's just launched an integration with Shopify, which makes you, I guess, helps you generate a Shopify website very quickly. So it's something you can play around with it and then automatically does the backend and literally spins up a Shopify instance for you. It's extremely powerful, something that's worth checking out. It's only about a week or two old. Um, but yeah, I think on the last point, I want to talk about it a little bit more than I guess some of the others. Um, inside Finder, once we hit this scale, um, you have to zoom out and you really have to have a look at how each of these things are running and what results they're achieving. So we built this process which is inspired by Amazon. They have a weekly business review and a monthly business review, and literally at the end of every month— I know it's old school— you get the physical workload finder of, you know, if you're in the executive team of how the company's performing. It literally has dozens of charts in there. You can dig into anything you want, but it also has an executive overview. And I'd encourage you, no matter the size of your company, to spend the time to build a dashboard on how your company's performing. In my perspective, I think you should sit down with somebody once a month, ideally somebody outside the company as well, and share how your business is going. Talk about the key metrics that drive your company, you know, the inputs and as well as the outputs. I think that's really, really important. Often by just spending the time to talk about your business to somebody else, you realise what you need to change yourself, but you might hear some really cool advice from somebody else as well, where they really are going to provide you that, that context and feedback around how your company is performing. For this, it's obviously at a company level, but then there's also the inputs which are hopefully driving each of those parts of your company now. And, you know, I guess each line item in your company, when you're looking at sales or, you know, costs or whatever it is, I think there's an opportunity to apply AI into all those things. And I think there's a really exciting opportunity to start testing and seeing what's working. I think some of those things I shared today are a good first step. But, you know, I think it's important to not just, you know, set and forget. It's going to need constant tuning. And I think if you can start assessing how an AI solution is performing versus your traditional human-led solution, I think that's really powerful. A lot of the companies that are having success, say, in customer service talk about response time and some of these metrics in terms of the human team and the AI team. Team. And I think that is ultimately the best solutions, human and AI, and getting really clear on where each step can be supported with human moderation and, you know, identifying things which still should be a human-only task. I think that, you know, you can augment a human task very easily with, with AI tools. And if you can train up your crew or train up yourself to really get better at it, it's going to help you so much in your company. Um, yeah, benchmarking that performance, I think, and adding it as a key measurement of your success going forward is critical. And the reason why I say that is, I don't know if anybody's been reading on this lately, but in the last few months, a lot of investors are pouring a lot of money into AI-native and AI-first businesses. So like the AI version of a thing. So AI version of a law firm, AI version of, you know, a bookkeeping company, wherever it might be. So if you think about that, by default, these companies are building the foundations from scratch, from scratch to be AI-first. And once they get it up and going and get to the quality that can replicate, you know, as close to as possible a human output, I think that's going to be pretty dangerous for your company if you haven't also made the investment yourself to try and remain competitive. The thing is this, those might be great solutions, but if you've built a company on relationships and trust, it's going to take time for them to achieve the same outcome. So from my standpoint, you still have plenty of time. It's not going to be an immediate disruption. I think that some of these technologies, when you look under the hood, they're actually not as good as they, you know, seem to be on the outside. But that doesn't mean we can be lazy, right? Like, I think it's, you've got to spend, in my view, this summer is critical. I think, you know, this December, Jan, normally people pull back, you know, that's kind of where I like to double down, especially on, you know, learning and developing and getting better at some of these things. I think there's some epic courses out there. Each of these platforms has normally an academy where they'll walk you through for free how to do some of these things and how to best leverage it. Heaps of independent people on YouTube are kind of walking you through step by step and having massive impact on individuals learning these things. So I think, yeah, spend the time on that and also have conversations with people that have solved it before. For me, this has been my playbook forever. If I want to learn something, I go and, like I said, I've met with all these companies. I've spoken with their teams. I've spoken with their customers. I've spoken with agencies that implement the things. I've spoken with independent consultants that specialise in implementing it as well. I've then spoken with alternative brands and understanding why people shifted from one tool to another and where the limitations were. I think that context and knowledge is hard to come by publicly and only by going through the human effort of really stepping all these, you know, through all these things, are you going to get to a quality solution for your company. And yeah, that's just going to take time. And, you know, obviously there are people that have done the work before. You can leverage those people. But yeah, please, please, please, um, research before you go and pass it all over to AI. I think that the temptation is to just literally hand it all over and be done with it and then move on to another part of your company. Um, when it comes to things like that have permanent damage, like, you know, whether it be a social post or automations and some of these things like applied to content platforms, you can actually have irreversible damage for your website or your social profile where it starts getting filtered or penalised and so on. So there are specific methods to put in place when you're using AI in a large-scale way. I just had a guy that, um, he messaged me 2 days ago. He had a massive Instagram account, 1.5 million followers, and he was using a WhatsApp automation with his personal phone number and actually lost access to his account in the last 24 hours. So yes, you can do it, but there's still some rules around how to do these things right. And if you rush too quickly into it, you can cause a lot of damage. So that's my, you know, really important advice is, you know, walk before you run. And that kind of evolution of AI kind of diagram hopes to try and, I guess, communicate that. I did want to talk about some personal productivity hacks as well, because I find that these get lost in company systems. So the first I actually have gotten more used to in the last couple of months is, you know, and I learned about this in an interesting way, is voice typing. So I was observing these AI developers and these are people that are building with AI that, you know, gun devs, but they've moved to an AI approach to building apps. And all of them are using voice typing and they're 4 times faster than typing normally. So I recommend voice typing. I also think that, you know, I was getting to nearly 234 words a minute, by the way. It's pretty wild. Soundboarding is key as well. So sometimes I was talking about documenting some of these things. If you're in a, I don't know, a groove where you can't figure out how to get through it, I think soundboarding is really powerful. Like if you want to sit there, talk to somebody else, have a conversation, but you know, ask if you can transcribe it. That's my hack to making sure I build a document that I need. Um, even this presentation, I, I soundboarded with somebody. I was like, hey, here's how I'm thinking about it, and it gave me the loose flow, and then I was able to go and manually go and develop it. And, um, yeah, and there's a lot— last two there, we've got Cursor and Claude Code. If you want to manipulate files, um, I think it's a really interesting thing on your, on your desktop. And then ChatGPT Agent Mode, I used it to, on the day it was released, buy a box of donut— doughnuts and send it to the Finder office. Um, yeah, really a whole bunch of fun. Um, give it a shot and you'll be surprised on what you can But yeah, I went through a lot. I spoke very quickly. I'm just trying to get so much value out there. But feel free to hit me up with any questions. I post about this stuff on LinkedIn and you can send me an email. But otherwise, over to the FAQ.

Boa host41:26
Thank you, Jeremy. I'll get you to stop sharing your slides so you can see everyone's faces. That was amazing. I've got a few notes myself. Taryn, I wasn't sure if yours was a question, but if it is, would you like to ask Jeremy first? Thanks. Hi, um, hi Jeremy, thank you so much for just going through, um, everything that you shared today. I am launching a direct-to-consumer fashion brand early next year, and coming from a legal background, it's obviously the total opposite to, you know, what I'm going into. Um, so systems and operations I'm pretty good at, I've got all that, you know, nailed, I think, to an extent, but I think the marketing side of things, I really want to use AI to optimise my marketing from a, you know, ads perspective and just getting my workflows better. So my question to you, and sorry that it wasn't that direct, it's just how can I use AI to optimise my ads and my marketing online? Because I won't be in store, so I'm really relying on, you know, that online saturated market space.

Jeremy42:33
Yeah. Yeah, sure. Um, so one of the use cases I've seen applied to marketing on Meta is using it to generate UGC ads. Um, and that's basically normally a very expensive thing, uh, where you have to pay somebody to kind of be on video and so on. And, um, because these ads are so quick and they're short, I think a UGC ad is actually worth exploring. Um, I think there's a tool called Arcads, A-R-C-A-D-S, which is worth looking at. Um, they specialise in this sort of thing. Um, so yeah, from a Meta side, I think that's, that's important for, um, say SEO, which is going to have a level of, I guess, investment if you're going to e-com. Um, I think there's some interesting tools out there that I guess automate repeated page formats. There's one called Psyke. It's S-P-S-Y-K-E. Um, Aussie business as well. Um, and yeah, they'll basically help develop a whole bunch of pages for you guys. But yeah, I think it's, in my view, the key thing is using AI progressively and slowly. It's not in replacement 100% of human effort. So, you know, my view, like, you know, there's some big brands in the US, for example, that are kind of looking at how they're doing things. About 5% of their time is spent using AI to develop the outcome, and then they sort of ratchet it up slowly. So I'd start humanly doing it and then use AI along the way.

Boa host43:56
Okay, great. Thank you. You're welcome. Thank you. Alina. Hi there. Sorry, I'm not going to use camera today. Yeah, I have a business in beauty. It's a nail salon and I've been using like ChatGPT for market research and I'm trying to build like a marketing strategy as well. But what I struggle with is that I find it like ChatGPT is giving me a pretty basic research even though I try to write a good prompt and do deep research feature as well. But yeah, it's either like I struggle with finding, like I mean ChatGPT struggles with finding correct information because it might be just not enough information out there or I'm struggling with writing prompts, so I was wondering if there is any like courses that you might recommend, if it's worth taking courses, and what platforms you would use for market research. Yep.

Jeremy45:02
And in terms of research, are you looking to find out, you know, about that particular industry or like what are you looking to solve for?

Boa host45:10
Yeah, like preferably about the beauty industry, but also overall like customer behaviour in Australia. Yeah.

Jeremy45:18
Um, honestly, uh, you know, I'm not sure how much time you spent with customers, but I think speaking to potential customers is actually, you know, it's not an AI-led solution, but is critical. There's a platform, um, Queensland business, um, I can shoot through the details after, but essentially what they're doing, um, is creating synthetic customer profiles. And so the wild thing is, is some research a couple of weeks ago showed that, um, they're 98% accurate. And that's wild. So if you're actually plugging these in, you can speak to effectively an AI customer to see like, you know, what level of feedback they might have on your particular product or service. So I think that's one use case that's interesting. In terms of research and so on, I think some of this information's hard to come by, possibly through deep research, but you have to think about how it's all getting its information. It has to be publicly available information for it to be a part of the large language model. So if it's not public, it's not part of its knowledge. So Sometimes you might have to use a more traditional solution that has, you know, data scraping and sort of the stuff inside it that really helps give information about the industry you're looking to go into, but that's normally behind closed walls. They're charging for that. So I think some really good tools that have like a 7-day free trial might be a tool like Semrush or Ahrefs. They will give you an idea of like how much people are searching for a particular category in a a region, how much you'd have to pay for that sort of traffic. And if you get an idea about how much people are paying for a click, you can get an idea of maybe on a simple conversion rate of 5% to a website, you can start estimating what, you know, your web traffic might look like. And then you can kind of think about foot traffic, combine all of that, and you get a loose business model. So again, it's a 7-day free trial. You can kill it, make sure you kill it before the 7 days because they're expensive. But these tools aren't really AI tools, but they've, they've used AI, I guess, to gather the information that they use to give you the insight. So highly recommend that. And I've kind of done that work before for companies where, yeah, you can pull from, from various things. You can see their traffic sources, where they get referenced on the internet, like which, you know, what sort of PR they've had for their company. So really, really powerful stuff. Check it out.

Boa host47:39
Okay. Thank you very much. Thank you. Thanks, Helena. Will.

Jeremy47:47
Hi, Jeremy. Thanks very much for that, mate. That was very interesting and great timing for me personally. The question I've asked is how long it should take to train an AI agent. So for some context, my business, I'm a property valuer. So I'd like to be able to have an automated version of what I do. So to be able to say to the agent, here's the property in the database, here's it's all its details, can you go and find other similar properties in the database and come up with the value based on how I would think to compare those properties to one another? Yeah. It depends what tool you're using for this. I think the way to approach this is it's a combination of, you know, I've thought about this space recently, actually. You're going to need APIs, I think, for high quality. Ideally, or if you, if you don't have access to those, you might be able to use another system that has at least some loose estimations to work from. But I think, you know, starting with some base APIs is important and connecting those into a system, whether it be, you know, Replit or whatever it might be, like building a base app and saying, hey, here's the SDK, the software development kit for this particular API. My goal is to do property valuation. This is going to be one of the inputs that I use, and you provide all the APIs. And it'll actually construct an app for you that's based on those software development kits. And so the base design is there and you can play with it. I think that would get you to a prototype that you can probably use for yourself. And yeah, I think that's, that's the way to approach it. It's going to be, it's not really like a ChatGPT thing for what you're talking about. I think goes well beyond what the scope of ChatGPT is there for. But yeah, if you can use something like Replit or Lovable to kind of play with it and get to a point where you plug in this information, you'd be surprised of how far down the line you can get to with with it. You're welcome.

Boa host49:49
Guys, was there any other questions this morning? Now's your chance, be as selfish as you want to be. Awesome. Now look, Jeremy, if you're comfortable, I will share your slides after the chat. I know you've got some amazing systems on there, obviously your contact details as well. So please, if you want to reach out to Jeremy, he has honestly been incredible for us in the BOA team. And then for your time this morning, and we'll see you guys soon. Thank you.

Jeremy50:18
Thanks for having me, and thanks for joining, guys.