AI Marketing for the Smart Entrepreneur: Future-Proof Your Business Now with Farlen Mischel

[00:00:00]

Farlen: I think generally the biggest issue I've seen with businesses are their education around the subject. So what is AI? What is automation? What are its limits?

What can I do with it? What can't I do with it? So, you know, a lot of my time is spent kind of, like, showing people the possibilities and how you can leverage this technology​

Allan: Welcome back to the Lean Marketing Podcast. My name is Allan Dib. I'm your host. I'm joined with a very special guest. He's a little bit of a different guest than we normally have. It's Farlen Mischel. He's an AI expert and specifically he'll help small businesses integrate AI into their day to day workflows so that we can become more efficient, do more with less, which is what lean marketing is really all about.

Farlen, welcome to the show.

Farlen: Thanks, Allan. Happy to be here.

Allan: Great to have you. [00:01:00] So, maybe give us a little bit of your background. What do you do? Who do you do it for?

Farlen: Yeah, great. So I started I've been running businesses my whole life ever since I graduated college. I was always afraid of. Going through an interview process, so I never actually made it that far. And uh, just kind of out of that necessity I've always started running companies. My 1st company was a mobile application company, offline map application, and then transitioned out of the tech industry into Video production advertising.

I ran a video production company for a while in San Francisco, and then into the cannabis industry when that was pretty hot, we ran a consultancy heavily science based where we would consult with new operations and growing operations. And then once AI came about, I got really interested in the technological aspect of it.

So I learned [00:02:00] how to code in JavaScript, Python, and React, and then got really interested in AI and the ability of using large language models. and then I just kind of, like, started to thread those 2 things together. And I just kind of fell in love with it. And I help businesses now with that same thing, bringing the technology and the process to different use cases and helping with,automating processes s and building minimum viable products

Allan: That's really, really cool. You and I have, I guess, a pretty similar background, I think, because we've integrated worlds of tech. You know, it's hard to lose your tech geek. I think it runs in the blood somewhat. And also the marketing world, which I think actually have a lot of crossover because people kind of think of marketing as like this super creative pursuit, you know, people on Madison Avenue on skateboards and all of that sort of stuff.

Whereas I think lot of marketing and a lot of the [00:03:00] success that. people have had with marketing are really just doing geek stuff, which is analyzing numbers, trying to improve on controls little tests and things like that. And I from what I've observed, people from a kind of geeky background tend to do a lot better in marketing than the super creative people.

I'm not saying creativity isn't important. Creativity is a very important part, but a lot of what we do in marketing is kind of the repetitive, boring iteration. And, tech geeks tend to do that second nature. So, a couple of things I want to hit on with AI and marketing, where are you seeing Some of the biggest gains that people are having from integrating AI into their marketing processes or really into their general business processes, because we've seen tons of tools, like literally every single day, there's a new tool that does this, that, or the other.

But when I look at, behind the curtain of most businesses, they're still using [00:04:00] spreadsheets, Google sheets, Notion, all the old same tools. Rarely am I looking behind the curtain saying, wow, they've really integrated AI and automation and workflows into their business. So what's kind of that gap right now?

Are we too early with the tools? Are people not understanding them? What's happening right now? And what are some of the biggest wins that you can see that, someone can get right now by integrating some of these tools?

Farlen: Yeah, that's actually a really great question. It really brings it yeah to what businesses really want, I think generally the biggest issue I've seen with businesses are their education around the subject. So what is AI? What is automation? What are its limits?

What can I do with it? What can't I do with it? So, you know, a lot of my time is spent kind of, like, showing people the possibilities and how you can leverage this technology.

With that said, though kind of a more concrete example [00:05:00] in the marketing space, if you look at a marketing, you have content creation, you have operational posting, you have operational execution content creation, the AI can really take 80 of that load.

And then you do the next 20 because you don't want to be creating content with AI and then posting it immediately. you need to look it over and put your own voice and your own things to it. And then, with the operational execution aspect of it, that's where the automation really comes in.

So you have automation for content creation, sure, but then the automation for. Connecting all these platforms, integrate, between your draft documents, and then automatically post them to LinkedIn, to Twitter, to Instagram, to Facebook. And of course, you can delineate between the content that you want to post to certain platforms.

Of course, there's different voices for each of those platforms. And [00:06:00] all of that can be, definitely executed.

But of course, then you have, you know, Internal kind of business operations integrations between, like, Notion and HubSpot, CRMs and Slack and all of your staff and kind of those mundane notifications and updates

Allan: I'll give you a bit of my take. So from what I've seen is most businesses have very poor business systems, meaning a lot of things are done in a haphazard way. The knowledge is in someone's mind. Someone's just been doing the job forever and they've got one way of doing it. And. One of the things I wrote in my book is that really what business systems are is just a series of checklists of, you know, here's how we do it.

Here's how we invoice a client. Here's how we collect money from a client. Here's how we push out a piece of content. And then there's a series of steps that says who does what and when. And so, If [00:07:00] you don't have that in place, it's very hard to start automating that stuff. Because if you do have those systems in place, if you do have those checklists, then we can say, hang on, this step that we're doing manually like right now, we can get a machine to do that.

And then this other third step, we can get a machine to do that. And before you know it, The vast majority of the process can be done by machines, but you can never get to that stage unless you've got the systems in place already where, you know, even if it's 100 done by humans.

That, you know, here's step one, here's step two and really breaking down each transaction in the workflow. So really understanding what does it take to go from idea to YouTube video or from client onboarding to billing or whatever it is. So for everything that you do in your business, there's a whole series of steps.

And most of, people do it either from their head, from their mind, and kind of wing it as they go along. Sometimes they do it this way, sometimes they do it that way, sometimes they do it a different way, [00:08:00] and hence it's very hard to automate that. It's very hard to plug in an AI or an automation tool or whatever to help you with that.

So really a lot of times there's a First step of systemizing what you do, putting it into a series of checklists, and then understanding, okay, great, these are the steps that we use right now to do this task or do this workflow. It's mostly done by humans now, what steps can be automated or done by AI?

And feel that's often the gap, usually well systemized businesses. it's pretty easy to plug in AI tools and it's easy to plug in automation workflows. But for a lot of businesses, they've got a step to do ahead of that and really figure that stuff out.

Farlen: Yeah, I mean, you touched on it perfectly. Those like standard operating procedures for different aspects of your business. That's what you want to automate. So, if you don't have those, you've got to create them, or you got to get [00:09:00] them from somewhere because it's hard to translate what you're winging in your mind and then telling an op, you know, an automation consultant or someone, hey, I want to do this like, oh, well, how do you do that?

don't have it in words or in writing, the AI isn't going to know what to do.

Allan: Yeah. What are some unusual, or maybe not necessarily unusual, but underutilized features of large language models? And when we talk about large language models, we're talking about stuff like ChatGPT or Claude, or, you know, there's a ton out there now.

They're probably the two best known, but, Literally every single day, I'm finding new uses, new ways to use them. They're like a paintbrush and a canvas, right? So the possibilities are endless. And so, I'm using it. personally in my life for diet, for exercise, for medical things in my business, I'm using it as a writing assistant, as an editor, I'm using it for mathematical formulas and things like that for [00:10:00] coding, for all sorts of things.

What are some of the out of the box applications that you've used, I've seen used large language models for?

Farlen: Yeah, this touches earlier you calling it about

just like changing so quickly this environment. There's always new products and there's those new tools. It is hard to stay on top of it. 1 of the tools I really like to use that helped me kind of with my curiosity and my education is Notebook LM, which is by Gemini. I'm sure you've used it, but for, for your listeners it's this amazing Powerful.

They use LLM to index some article or transcript or anything that you want. That's text based. And then you can create a podcast about it and listen to it. and you can also the podcast can be created based on certain aspects of the document that you want to understand better as well. So that kind of changed the game for me in terms of.

Upgrading my skill set and upgrading my knowledge and understanding [00:11:00] of these tools as well. And then just, I think 1 of the biggest differences I've been using recently as these new models are coming out is using the different types of models for different types of executions.

So, 01, for example, I'll use for like big picture kind of design questions and stuff like that. Then maybe Sonnet 3. 5 for executing code on those big things, or 4. 0 for just, you know, And this is OpenAI's 4. 0 for different kind of daily routine tasks or writing blog posts and stuff like that.

Allan: So, um, I, I found that as well. There are some models that are tuned to specific tasks, so, I find Claude really good. Claude son at 3.5. For writing. So it's writing is way better than anything from GPT. As you mentioned, 01 is currently, I mean, this will be outdated probably by the time we [00:12:00] publish, it's their thinking model.

So it's the model that will kind of do a little bit of reasoning. So if you've got a complex problem to solve, 01 is a really good model. So there are so many models out there now. There are. I mean, most people know ChatGPT, some people know Claude and, but there are really tons of different models that are tuned to different tasks.

So sometimes it can be a little bit of work just to keep up with what's what, but one of the things that they all have in common. is, and really this is something that's common to all things computing, is you know, for the longest time there's been this acronym GIGO, which is garbage in, garbage out, right?

And a lot of people ask really terrible questions, right? So The quality of the answers that you get, not just from a large language model, but from a mentor, from anyone, is often going to be a function of the quality of your question. So, and [00:13:00] you know, as someone who gets questions sent my way every single day, I can see that there's a vast difference in the quality of questions.

Like sometimes I'll get a two line email from someone saying, I want to start a marketing agency. What's your advice? That's the question. I'm like I don't know what to say to that. Whereas sometimes I'll have someone really dump a lot of context. They'll be like, here's my goal for the business.

Here are some of the resources I've got. Here are some of my talents. Here's what I've tried. And here are the results that have happened. Now, here's my specific question. How do I blah, blah, blah. So now. You know, I've got context, I've got their goals, I've got their resources, I really under, understand at a deep level, and then I can give them a pretty thoughtful, well thought out response.

And in a similar way, the results that you're going to get from a large language model, like a GPT or Claude or whatever, is going to be the result of How good your prompting is, you know, and I mean, we've all seen those videos and YouTubes and all that about writing prompts, [00:14:00] but really you want to think of it as like a super smart mentor.

Now, if you were in the same room as Warren Buffett and you asked a question like you know, how do I get rich? I mean, that's a pretty stupid question to, to waste when you're with that guy. But if you're like, look here You know, here's my portfolio, here's some of my mix, here's my thinking around what I'm thinking the S& P 500 is going to do, or whatever, and here are some of my goals, and so on and so forth, then he can give you a much better answer in a similar way when you're prompting one of these large language models.

You want to start, okay, here's my goal. here's what I want to accomplish or here's the kind of answer that I want. Here's the format that I want the answer in. I want it in a table or I want it in a sentence or I want it in whatever. Here's some of the context around my thinking. So I've been doing this, that, and the other.

Now, Can you give me an answer based on that? You're going to get a much better response than if you're just sending some one liner or whatever, and it just generates [00:15:00] some random output, which is not going to be really helpful at all.

Farlen: Yeah, and the only thing I would add to that is 2 things. 1, 13, 24, All those things you mentioned on the beginning of the prompt is essential. And then also what helps as well is providing examples of outputs as well that also helps and possibly some important key reminders at the bottom, all of the large language model, the frontier companies, OpenAI, Sonnet, if you go to their developer docs, they do have, like, these prompting outlines that actually Are specific to each model, which is also very helpful which helps a lot.

Allan: Which is kind of the same way that you would work with an employee showing them, Hey, Here's what I want you to accomplish. Here's what good looks like. And I want you to, you know, create an output that's similar to this. Now, here's how we've done it in the past. We've done step one, step two, step three.

So that way. So that this is an employee, someone who's smart, can follow [00:16:00] instructions, but you need to give them specific steps. So, and I think they're going to get better at thinking and reasoning. They're already very, very good. Like a lot of times I will maybe mistype something or type something the wrong way and it's guessed what I meant rather than what I wrote.

So I think that's been a pleasant surprise and it's going to be something that's going to continue to improve over time.

Agents are kind of being hot on and off and, you know, there's been a lot of talk about agents, but I'm yet to see them really working super well for anyone just yet. And I don't know, maybe that's just me, but what are your thoughts on agents and how they're being utilized right now? Are there any agent technologies that are working well, or is it sort of still beta experimental form?

Farlen: Yeah. I think as the models get better agents are going to be maybe becoming more ubiquitous within these solutions. I believe, [00:17:00] right now, we have things like MCP, which is essentially like model context protocol, which you can give tools to your AI, to your LLM, and then they can go execute.

They could check the weather. They could, you know, make reservations. They can do all these things. The biggest differentiation between, I think, that and next step, which would be like autonomous agent is its ability to figure out which tool to use best when and then go and actually execute it and then not run into a million errors due to context because a lot of the reasons why the agents don't work right now is because the context limits.

So that is the amount of information the AI in one. Message can understand about what's going on. They hit these limits. So you have frameworks like open zero or swarm or you know, crew AI that you can create little agents that do specific tools. And then it's like a hierarchy, and then you can have, like, a, manager overseer [00:18:00] on all these tools, but you still run into these context limits because there's so much going on.

So once we increase context lens, I think or there's open AI has been playing with different tokenization methods. And that's a little more complex, but we're going there.

Allan: just to clarify, so a context window is basically how much you can put into kind of a single message or a chat window into either ChatGPT or Claude or one of these other large language models to give it context on what you're doing. And currently it's fairly limited.

so context is very helpful for the large language model to understand what you want to do. And it becomes a problem when you're working with a lot of data.

Like, for example let's say you are wanting to query a book or a huge set of code or whatever. So a workaround for this right now is. to chunk down the tasks. So instead of saying, look analyze this book and give me the, whatever, [00:19:00] you might do it chapter by chapter, or you might do it section by section, same with code that you're working on, which again comes back down to how good are the, Instructions that you've got.

How good are the standard operating procedures and systems that you've got? Can you chunk down a task into 10 different subtasks and then the subtasks can be done really well and understood by the model and then you can string those together. So, that's something that's absolutely critical right now.

If you haven't systemized your marketing, if you haven't systemized your business, you're not going to be able to leverage a lot of these tools, whether it be agents, large language models, or even just normal automation tools, because you haven't chunked down the tasks and you haven't got a repeatable systemized process for doing that.

One other thing I wanted to talk to you about is MVPs, so Minimum Viable Products. So, this is kind of a concept that came from the Lean Startup. It's probably, I don't know, [00:20:00] maybe 15 years old, something like that. It's something that, I've heard some people say, you know, that's no longer relevant in today's environment because, you know, people want polished apps, polished tools, all of those sorts of things.

Whereas other people are like, no let's just get a janky version up of whatever we want. Let's see if it's working and then let's course correct. What are your thoughts around minimum viable products? If I've got an idea for something in my business, an app or some way of doing it is a minimum viable product still the way to do it?

How do I go about it in a cheap, easy way where I don't have to spend half a million dollars and tons of developer time and all of this sort of stuff, getting my vision out and then find out actually this is not really what people want or whatever else.

Farlen: Yeah, I love how you quoted lean startup. That's came out like right when I started everything. So that was definitely a big driver for me. And that essentially that driver is all about. You know, the MVP [00:21:00] concept is really just about, getting it out to the customers to see what they want first, because building anything in an echo chamber or building anything in the dark is, you don't really know what you're doing until you interact with people.

So I actually read recently on Twitter, someone was posting that we don't build MVPs, we build. MLPs, most lovable products, because it's all about, you know, how much the community interacts with it. And the community is what iterates and drives the development of the product. So we are in a time right now where no code, low code platforms and solutions are just, they're everywhere.

You don't need to hire developers anymore to kind of flesh out ideas that you need. If you have some technical wherewithal and a little bit of time, well, maybe depending on the platform, but you can pretty much flesh out an idea, wait really quickly [00:22:00] days instead of months.

And, you know, platforms such as make. com, even Zapier or N8n, or FlutterFlow, Bubble, these no code platforms allow for a lot of Even enterprise grade web and app solutions that you don't even need to hire developers for and then once you. Your MLP or MVP gets out and gets, some interest and you kind of understand how the response is with your community, then you can scale it.

You can hire developers, but you can go from 0 to something in a week now, which maybe used to take. Months, and honestly, a lot of people, it would just, it wouldn't be feasible because it just to think about hiring a team to just figure out a solution is not, viable.

Allan: Yeah, I agree. You can, absolutely go from zero to something in a week or so. One of the things, though, [00:23:00] I have found is you still do need some domain knowledge, right? You do need to guide it. You need to tell it you need to understand what it is that you want to do, how you want to do it.

Maybe even, you know, you know, kind of have the knowledge where you know enough to be dangerous, but you don't really have a deep understanding. So where I've found some of the best gains is where you're maybe like, you know, a little bit about the domain or whatever, you understand the key concepts, but you're not a super duper expert in that domain.

And that can, really be a big leverage point for you. If you know absolutely nothing about it. You can still do something, but it's probably not going to be what you want. So a lot of times you still do need to know a little bit about a particular domain, whether it's coding, whether it's legal, whether it's financial or whatever, because again, it comes down to the quality of questions and input that you're going to give it.

and that's going to determine the quality of output and [00:24:00] what you get out. So, I need to understand a little bit about contract law or whatever to update a clause in a contract and make it right, because I need to tell the large language model, you know what, change this clause to mean whatever, pro rata, not monthly, or whatever's relevant in your case.

And if you don't understand those basic Concepts, you're not really going to be able to give it that instruction. So, I would highly recommend anyone who wants to go down that track to either get themselves up to speed in whatever domain that they want to kind of use that model with, or hire someone who's maybe junior in that area, and then they can, you know, put out results that maybe someone senior could have done only a few years ago.

So I kind of think about it like the Iron Man suit, right? So, you know, Tony Stark, geeky guy um, normal guy, right? But he puts on that Iron Man suit and he can fly, he can shoot lasers, he can do all sorts of crazy stuff. So I think of [00:25:00] the AI models as that Iron Man suit. but still he has to pilot it.

He has to tell it where to go. He has to guide it and all of that sort of thing. So in a similar way, I think of it that way. So you still need a little bit of knowledge around, and maybe that will change down the track. Maybe there'll be more with agents that really guide you step by step. But for the moment, I think of it as a leverage tool.

You still have to guide it. You still have to be the driver in the driver's seat and putting in good input for it to get good output.

Farlen: Yeah that's a great point. I mean, like, coding, for example, you can tell it to do something. It'll give you code. But if you don't, know what's, like, executing the code is 1 thing, building the code is different um, course. And then I think. Having that domain knowledge, the specific understanding of, Hey, here's my problem.

this is my pain point in my industry, in my domain. And then, maybe, like you said, hiring a junior person to build out some solution, [00:26:00] you already have an advantage significantly because. The pain point in that domain is what is driving the need for that, you know, solution. So I think the entrance the buy-in level or entrance level for building something that is helpful is just like diminishing right now.

Allan: I mean, we're talking about large language models a lot, but outside of that, there are other AI and automation tools. So, currently one that we're using very well is there are a couple of media tools that help us be much more efficient in video editing. Literally we will use that to edit this episode.

So we use a, an app called Descript that really helps us with It transcribes, it helps us visually edit videos and other media, audio as well, that sort of thing. There are tools like Opus Pro that help us pull little bits of clips out as well. So there's a whole host of AI tools outside of just large language models that can be really [00:27:00] helpful.

There's knowledge based tools, you mentioned one LM Studio, which I love LM Studio to use for deep research of different modalities. So for example, if I've got maybe five different YouTube videos by the same person, maybe they've gone on different podcast interviews, then I've got maybe a copy of their book or maybe something they've written or blog posts or whatever.

And then I can query those. I can ask for summaries. I can ask questions. So for example, I had three episodes of The Physicist, Eric Weinstein. And they were like each like three hour episodes. So like there's 10 hours worth of podcasts. Then there's some of the stuff he's written, but he talks a lot about it, the same themes.

And so I was able to ask specific questions, like what does he mean about, you know, this, that, and the other. And it would give me a really intelligent output. Stuff that would literally have taken me probably a week of research to listen to all the stuff, take notes, and then really summarize. I could do literally [00:28:00] in minutes, which was really powerful.

And like you mentioned, it does have that feature where it can kind of create like a fake podcast where it summarizes those learnings. But actually I found that, that podcast thing. While it's a cool feature to demo, it was actually the least useful part of it to me. Just being able to aggregate different types of media and ask questions based on it, I thought was super, super powerful.

Farlen: Yeah, definitely. You can build your own context chatbots for your companies that are just essentially trained on all of your company's documentation internally, which is super helpful for people who are you know, I don't know, sales reps or, kind of anyone like that.

Also the chatbots are getting better and. AI voice agents that execute these chatbot interactions are getting really good. So, like, 11 labs and like, audio transcription, but voice flow is when I love using to build out these, like, branches of communication between people and online [00:29:00] and also, On the phone and stuff like that.

So.

Allan: 11Labs was a tool that I used extensively in the, my second book and the way I used it, 11Labs is basically will do text to speech and it does it in a very realistic way. So I used it to listen to my book back and then it was basically 95 percent as good as a professional narrator, but I didn't use it to record the audio book.

I used it to basically listen back to my writing to see where there were clumsy phrases, where there was something I wrote that didn't quite make sense or whatever, because I could literally be in the car listening to my book as an audio, what my audio book would sound like. And I'm like, Hang on, that, didn't really make sense, or that, that was a weird way to phrase it or whatever, and you know, I had to pull over a lot and take notes, because when you're listening back to your own writing, it's a very different experience than when you're in front of the keyboard, because when you're in front of the keyboard, you're like you think you're writing what you mean, [00:30:00] but You're writing maybe sometimes something different, you're writing in a clumsy way or whatever, but when you're sort of passively listening back and you're like, well, that didn't sound right or sometimes quite the opposite.

Sometimes, well, wow that, that was a great sentence, nailed it. So, in both scenarios, I found that extremely useful. And because it's so realistic, it sounds almost exactly the same as a narrator would read it out.

Farlen: Yeah, and that's wonderful. I actually love 11 labs because first of all, they have different accents, which is really interesting. You can play with and kind of hear your words and different accents and the language. They have different languages as well. So, very powerful stuff.

Allan: Totally. Well, Farlen I appreciate you coming on the show. I love talking tech with you always. And if you're listening to this show before the 25th of February or so, we're going to have a two day AI marketing workshop on the 25th and 26th of February in San Diego. So if you happen to [00:31:00] be in town or want to be in town reach out to me.

to my team and we'll have a chat.

Farlen, where do people find out more about you and what you do?

Farlen: Sure. Yeah, my website is enny.ai so E N N Y. ai. And feel free to drop me a message there. I'm building a small educational community where you can sign up there where you'll have access to kind of up to date resources that I'm putting together for, different business use cases that I've encountered and um, connect with me on LinkedIn, if you'd like.

Allan: Awesome. Awesome. Well, thanks Fallon. Thanks for being on and we'll talk to you real soon.

Farlen: All right. Thank you, Allan. Appreciate it.

​

‍