From Zero to $10,000/month: Breaking into the AI Industry and Landing Your First Clients

Hey everyone, and welcome back to the AI Automations Podcast. I am super excited to have today's guest with us. It's Mark Kashov. He's one of my good friends, and he's also the founder of promptadvisors.com. So today, what we're going to be talking about is how to break into AI, how to get started, find your first clients, and the most popular AI services that are being offered at the moment. So welcome, Mark. It's great to have you here.

Thank you so much, Helena. It's a pleasure, and I'm looking forward to our conversation today.

Yeah. First of all, why does it say 43rd time a charm on your shirt?

Absolutely. So I've done, since I graduated high school, I've started 42 businesses. 37 were massive failures. 4 did okay, but this is the first time that it's actually going really well. And I think it's because it's in an area that I'm like passionate about. I'm offering stuff that I'm also passionate about. So it's finally working for once. So I've had to make a lot of swag to commemorate that.

Wow, that's amazing. Shows the perseverance. So, you know, today I want to kind of want to talk to you, how do you, you know, kind of, because you're, you're, you do quite well with your AI agency. Can you kind of give us a little bit of background and how you got the idea to start an AI agency?

Sure. So when ChatGPT came out in 2022, I was in between businesses. As you could say, my last business was completely unrelated. I actually started a handyman business in the Ottawa and Toronto area, which was so out of my depth. But when it came out a few days later, I looked at it and having done a master's in AI where one of my specialties was doing natural language processing, I knew that the average user wouldn't be able to write a proper chat, like a prompt to ChatGPT. So I started thinking people will probably start paying other people to help them use it.

Mm-hmm.

So I started posting my first gig ever on a platform called Fiverr, and I got zero traction for the first month. But around January to February, that's when the average person started even hearing about it. And then I started getting some conversations on like building people prompts for very cheap. for like $10 to $20. And then that ended up rolling up into much more advanced prompts and then building chatbots, building AI apps. So that took me for the next 4 or 5 months down that route. And yeah, ever since then, we've been doing really well on Fiverr. We just became a top-rated AI seller last 2 weeks ago. So it's been good ever since.

Cool. I want to dive more into how you got started on Fiverr, But broadly in 2024, if someone's looking to start some sort of AI side hustle on the end, how do they find their first clients? Like what channels would you use? Fiverr is one of them, but on the high level, which— what are the top ones that you think people should use?

Totally. So you have your standard Fiverr, Upwork, those kinds of platforms, freelancer.com, where someone's going there with the intent of looking for something. I think what most people don't think about is that a lot of people still don't know what's possible, which is why we're having this podcast, right?

Yeah.

A lot of folks, friends of mine are doing pretty well on LinkedIn, for example, where they're just posting a 10-minute Loom of them building some solution that automates X. And over time they're getting tons of traction. You can also even run things like a Twitter ad for like $10. You could post your solution and say, hey, Here's a video of me automating this entire process, and you kind of put yourself in the ether there. So if you have no traction, no client base, I would kind of spread seeds in different areas from Upwork to Fiverr to something more off-brand from these sites where you're kind of showcasing what you can do. And that's where you can start getting those initial DMs that end up in a project.

So just kind of building it up from From there, have you tried cold calling or cold emailing or had any success with that, or is it just more about building your brand?

So it really depends on what your offer is. So we're about to start cold calling for a new service, a microservice we're offering that we can talk about later.

I'm going to talk about that in a second.

Yeah. Absolutely. Because we know exactly what we're offering and we know exactly how to position it and package it. If you're Offering something that's very vanilla or open, like I'll build you a chatbot, that's your service. Cold calling is a bit tricky because how do you differentiate your bespoke chatbot versus what they can get for $20 a month online? That's trickier. Cold calling, cold emailing. Yeah. Not say that you can't do it, but it'll take you thousands to get one lead versus maybe hundreds if you actually have a well-positioned offer, something you can reproduce over and over again and make a good margin on it.

Right, right. That's, that's a really, really good distinction there. Like, if you have a very, very specific offer to a specific niche, like cold calling, cold emailing could still work. But other than that, it's, you know, building up and showing people what's possible. Because a big part of, I think, AI is just education. It's a good analogy I heard was this. If you go into a car dealership, The salesperson doesn't have to explain to you what a car is, you know what a car is. So you're, you're going in there, knowing that you're going to buy a car. But a lot of when people approach or talk to an AI agency, they have no idea what they're buying. They're just like, oh, I heard this buzzword. And, and, and I want to see what it is.

So like, you know, absolutely, it's kind of like leading the blind. So on Fiverr right now, Our profile has like 6 or 7 gigs that we offer. A gig is basically like a service. Despite having all of that, and it's well explained, people will still message me today and say, I can't tell what exact services you offer. Can you summarize? So the average user that doesn't know anything about AI doesn't even know what they're looking for.

Yeah.

Some folks are saying, hey, I'm in the car rental industry and I operate my own real estate fund or property management firm. Can you tell us what AI can do for us? It's very large questions and you have to be able to kind of scope that down and boil it down to a few key things that you know you can deliver well, but also will give the client a lot of ROI.

Right. Absolutely. Absolutely. So what kind of gigs do you, did you post on Fiverr? Can you kind of like walk us through your Fiverr strategy and how you, how you were able to find clients on Fiverr?

Yeah. So the first couple months it was literally, I will build bespoke prompts for you, or I will apply AutoGPT. At the time, a year ago, literally in April, AutoGPT got released and people thought that was AGI because it could do a couple things on its own. So I kind of hopped on that trend and said, I'll implement it for you on your laptop. That was one of my gigs. And then you had another one, prompt engineering. When I started getting real traction, that's where I started venturing into, build you a chatbot or something on Voiceflow or Botpress. And then eventually became, I'll build you a whole AI SaaS app when I started scaling and actually built a development team to help me deliver on these bigger projects. And now we offer anywhere from building end-to-end apps, SaaS or not, to still doing some prompt engineering where it makes sense. And now we're doing AI call-in, which is kind of a brand new service area that became More popular in the last couple months.

Perfect, perfect. I want to dive into that in a moment. So what are your top three most popular services on Fiverr right now?

Gotcha. So popular. So the prompting is always a good what I like to say a gateway drug. So someone will come to you saying, "Hey, we built an app already. We already have a prompt, but our users don't like XYZ." Could you help us optimize it? So I engage for a very small amount because that kind of gets me in the door. And then when you can establish authority that you are who you say you are and you know what to do, then that leads to, okay, could you actually just build this next product feature? So that's my most— still my most popular gateway drug gig overall.

Interesting. So prompts are not that hard to write though. Um, I mean, uh, um, so like I guess I'm just trying to understand the value proposition of like, can you give an example of someone who comes to you to write a prompt? Because to me, they're, they're pretty simple for the most part, unless I'm missing something.

It's simple to do simple things. So it's like, can you build me, can you write me an email? Yeah, it's a very simple prompt. But the type of companies that are coming are, for example, one company produces videos from a prompt. So text to prompt. So they have like 4 or 5 prompts that like take user input as to what they want this video to be about. And then we transform it, and then we break it up into 3-second intervals. Each 3-second interval is then sent to, like, Stable Diffusion to make an image of it. And then we animate that image, and then we clip that all together. So we have to create a prompt for the script, a prompt for what the visual prompt should be to Stable Diffusion. So when you add those more bespoke applications, that's where you need someone that really knows what they're doing, because it even comes down to cost optimization. If you have very big prompts, it's very expensive to run at scale of hundreds of thousands of users. So how do we compress a prompt to do the same thing but with less tokens? So it comes down to that kind of art form when it comes to science like that.

Amazing. Amazing. Okay, cool. And then moving on. So you've got, you've got, so you would recommend someone starting out putting a profile on Fiverr. Any other tips for Fiverr? Besides making a gig?

Look, go and search on Fiverr for gaps that don't exist. So, um, I even released a video on this, but in general, if you go on Fiverr right now, you go on AI services, you're going to see a few new categories that have the label new on it. So one of them was literally writing custom prompts, or one of them is, if I just pull it up right now, if I have it, and I don't know, this is an audio podcast, correct?

It's going on YouTube too.

Okay, I get it. So if I go here, I can actually share my screen for 2 seconds. Oh.

Sure, go ahead.

Yeah, no worries.

Go ahead, you can share your screen.

Got it. So if we go here, do you see my screen okay?

Yep.

Yeah, so if you go to AI Services, you're going to see a few of these that have new as tags. So custom GPT apps, that's a new category. AI strategy, AI lessons, custom writing prompts. And the reason why it's good to hop on these categories is I don't even rank well in these categories because the, the search and the metadata is not even optimized yet.

Got it.

So if you go to custom GPT apps, I happen to be there, number one, but in general, sometimes it'll mishmash, but you'll start as you go down. Start seeing that not all of these are necessarily custom GPTs, or they're not— they don't have much competition. I don't see any reviews here. So you only have the first row or 2 with folks that actually have reviews. So there's definitely an area that you can play in if you're offering more bespoke solutions.

Got it. That's really good advice. That's really good advice. So Fiverr, start posting some content on LinkedIn, LinkedIn is pretty easy to get in. And basically just getting the word out there, and a huge part of it is education. Got it. Okay, now that we kind of got that part, let's talk about the services that are really popular within AI right now. Because, you know, it's so broad, you can generate— I think a big part of it is people think you could just play around with AI, but what are the practical use cases of it, right? Like, you can use ChatGPT to write images. I think a lady asked me the other day, that's cool, so Midjourney can generate these images, but what do I ever use it for? And how do I make money? More importantly, how do I make money with it? Right?

Absolutely. So typically when you are telling a client how they can make money with it, one useful skill, just being able to share how much time they're saving with it, because time is money. So if you save them time, then that's ROI as well. So when it comes to most popular services, we'll, we'll pass off prompt engineering to go to the some more bespoke ones.

Yeah.

It's more so how can I use things like ChatGPT Vision to do my job better? So as an example, one client just came up in our pipeline who has blueprints that he offer— like, he usually goes over with his team. He's a contractor in construction. They try to look at a blueprint to determine exactly what materials they have. So they'll go on AutoCAD and create a whole diagram for what a stadium should look like or a house should look like. And it has all these hints as to what the materials represent. They typically comb through and come up with a materials list themselves. We're trying to use something like GPT Vision to be able to scan those blueprints, get trained on what materials are involved, and then do, do that for them, which is saving them 5 to 6 hours. We're using, again, very small components of technology that exists to try to save tons of time. If that makes sense. When it comes to agentic AI, which is a fancy word for saying multiple agents.

Yeah.

When it comes to a workflow, like, let's say, making social media content, you can have one AI agent that goes and web scrapes a couple websites in terms of what's the most current news. And then you have another agent that takes that news as an input, and then it has a prompt, and it writes a draft of a LinkedIn post. Then you have another agent that takes that LinkedIn post and transforms it to an Instagram post. So you can get multiple outputs from one input. That's also increasingly becoming more popular, especially when you start hearing things like agent swarms, which is basically like having tens of these agents that are all doing things concurrently that all converge into one main output.

Right. Right. And what platform are you using to make that no-code, or are you using code platforms to make that?

Yeah. So 2 ones is like Relevance AI is helpful for no-code, especially if you wanna build like fairly simple agents you run in tandem. You can also use VoiceFlow to an extent to actually do these agents. For example, if you combine the Assistant API, which is OpenAI's agentic API, with VoiceFlow, you can start routing different tasks to different assistants. And the last one is more custom development where we're actually doing it all Pythonically. There's new agent libraries where you can spin up a few things Pythonically, then you serve it to a front end in some way, and then you can leverage it that way. So you have the ability for both.

Yeah, yeah. I just don't feel like any of the no-code solutions are really there yet. I used to recommend Relevance, except they just 10x their price.

Yeah. So if you're new to Relevance, then yeah, we got grandfathered in. I assume you did.

Yeah, I did too. I'm like, I'm never canceling my account now.

Never, never. They got us locked in. But When it comes to things that are more bespoke, I think when it comes to agent things, even function calling can be considered multi-agent. Because if you have GPT-3.5 Turbo, or now Claude can use tools as well, you can make a function call to different other Pythonic functions that do different things. So agents is not just one, one fits all solution. There's the low-code, like, relevance, which is like, it's pretty on the eyes how it works. It's very visual. But I think in the next 1 to 2 years, it's gonna be like text to agent where you can go in GPT. And right now, if you do@ in GPT, you can call one of your custom GPTs to actually do a task specifically. I think in 3, 4 months, you'll be able to do@ multiple custom GPTs at once where you can have multiple GPTs work together to decide which one should do the task you're asking for.

Mhmm. Yeah, so you're seeing a lot of agents pop up. You're seeing still a lot of custom prompts. And of course, we have to talk about voice. In my opinion, that's probably the biggest one. I don't know about your thoughts on that. Do you think it's—

It's one of the biggest, but it's also one of our favorite because it's also very fun to build. So in the last 2 months, we've gotten inquiries from anywhere from dental offices to real estate firms to Now we're working with a couple medical practices to try to like reschedule appointments for patients or listen, help them quickly diagnose very minor issues. So when it comes to voice, 6 months ago, took like 2 good seconds to get a response and it sounded hyper robotic.

Yeah. Now it's almost instant.

It's almost instant. Literally, as of 2 weeks ago, it's only like 0.2 milliseconds depending on the platform you're using, whether it's Bland, VAPI, Synflo. And there's a brand new one called Hume, which is more emotion detection, where it's kind of like it's hearing that you're angry, so it responds way calmer, that kind of thing. That's one area that's really interesting for us, where that develops in the next 3 to 4 months.

Yeah. So what— are you using one platform to develop all of your voice agents, or one specific one? And I'm assuming you use all of it, so maybe we can talk about the pros and cons of these different platforms as well.

Totally. I could give the audience a TL;DR of what we use for what. So, Blends are a favorite for using for 70% of our AI solutions. The reason why is they have a nice conversation builder now they just implemented, which gives you like a mini chatbot experience where you can plan for different scenarios that pop up in the conversation. Their cost per minute has gone down a lot, especially even the last couple weeks.

I didn't check. It was at 12 cents a minute when I first checked. Like—

Yeah, now it's 9 cents. 9 cents. Okay.

So they went down a lot.

And where we think it's going is we think by August, all of these voice agents will be almost free. And then you'll be paying a subscription for the platform to maybe integrate with other pieces of your backend or like automations. So we think that'll trend towards zero where that's helpful, especially if you're selling this as a micro SaaS. We sell it by a minute allocation. Kind of like calling cards that our parents used to have. Like, you buy 500 minutes to call Egypt back home. We're, we're kind of tiering it that way. We're making a bit of an arbitrage on how we price those agents.

Mhmm.

Yeah. So we use Bland for 70%. We use VAPI when we wanna switch the large language model that we're using. So if you want to use something like Groq, which is a super fast LLM that's very quick at generating information, You can swap that into VAPI to be able to swap the brain of the voice caller, whereas Bland, it's a bit more black box. When it comes to— what was that last one? Oh, Synflow. Synflow is the easiest platform across the board to use if you have no coding knowledge whatsoever. It's literally a series of clicks and API keys. The drawback is just the very high price per month and the very high price per minute.

Mm-hmm.

So they know what they're doing in terms of how they're pricing this because they've made it approachable for anyone to be able to build a voice agent.

Got it. You had to be on their, I think, agency plan for $1,400, and then it goes down to $0.08 per minute, which is comparable to what Blend is offering. But you had to have the volume to do that. You had to have the clients and the volume to do that.

Yeah, you have to have some solid lead flow. And they have to look like very similar clients. So one of Synflow's main integrations for CRM, for example, is GoHighLevel. So if you have 90% of your clients using GoHighLevel and you're kind of reproducing everything fairly similarly, might be a good idea for you if you're making enough margin per client. And ideally you have some form of retainer to help you continuously pay off that $1,400 a month.

For sure, for sure. So what are the use cases that you're currently seeing in voice? I know, I know you kind of mentioned it, but just just wanted to be thorough and see if we didn't forget any. You said appointment setting for doctors, any, any like other kind of use cases that you're seeing?

Yeah, sure. I could rapid-fire them. So we said doctors, dentists, in terms of rescheduling, you know, those appointment reminders, typically a receptionist will take 4 hours of her day almost every day to do appointment reminders or appointment rescheduling. So they just call, hey, Helena, we have a a dentist appointment tomorrow. Are you still good to confirm?

Yeah.

You can take all of that, right?

Yeah.

In terms of real estate, we've seen outbound and inbound. So outbound to cold call other agents that are listing properties to see if they can get showings, stuff like that. Or inbound where they're being asked like, what's available in this area? Is it a good time to buy? And then it can even query live information on the internet saying, oh yeah, our interest rates are here. It seems like things are slowing down in the Colorado area. And then you can have a back and forth that way. Some more bespoke applications are, let's say, a therapist. So someone calls in, they want to be able to have a conversation back and forth, right? That's something that we're seeing pop up with a couple inquiries here and there. I think that will be more popular in 3 to 4 months when the emotion recognition comes in and now it's kind of more empathetic instead of just like a robotic kind of conversation.

That's so cool.

Yeah, absolutely. The other ones, rapid-fire restaurants, easy ones. So like ordering from a pizza spot, you can only have so many combinations that you can walk it through.

Yeah, yeah.

Anything delivery-oriented, like right now we're working with a partner in Paris where they're digitizing people's menus, taking a picture of it, Loading it in, and that gets passed into our AI calling agent. So when you call and say like, hey, what's on the menu? Or can I make a reservation? There's a very quick way to take old-style restaurants and digitize them that way.

Wow, that's so cool. So basically, like, the main use case is as a receptionist and booking appointments that you're seeing with voice. Like, that's like, would you say 70, 80% of the use case?

Completely. Where, where we're heading in terms of, we're building something a bit more Standalone as more of a moat is we're building an agent that has consistent memory, meaning when Helena calls from her number, when it picks up the phone, it already has looked at all the conversation history with Helena to say, hey, nice to hear from you. Do you want to rebook this appointment or did you want to do this thing specifically? So that's when you get a real receptionist. Right now, the thing that a human receptionist has that AI doesn't Is you can remember Dr. James telling you XYZ last time. So the conversation is very different. You don't have to repeat yourself over and over again in the same flow. We're not there yet exact with AI.

Right. Right. So it's, it's being able to look up that person's phone number and then look at all of the transcripts that that person called from before.

Exactly. Exactly.

That's something custom that you're building right now?

Yeah, because we expect that more platforms like Bland, Vapi, etc., are going to pop up as this space gets a bit more fatty and you start seeing it more on YouTube with random people that don't know AI saying make $10,000 a month. As soon as we get to that point, that's when we have tons of people selling the same thing. So right now we're planning on how do we differentiate ourselves 5 months from now kind of thing.

Sure. So it's not just calling, but you have your own custom Sass attached to it. That will make it— yeah, perfect. Well, and then I lost my train of thought here. Not really related to what we were just talking about, but would you say all vector databases are created equal?

No.

Okay, and why? Let's say you load all of your summaries into Voiceflow Or Pinecone, would you get the same results? Like, is a vector database a vector database, or is there a difference between how they're actually made and formed?

So that's a loaded question because it's actually multi-layered. So it's not necessarily just vector DB to vector DB comparison, it's RAG to RAG. So VoiceFlow is doing its own version of RAG to chunk your data and embed it. Um, Pinecone is just a vector DB. It doesn't really do the chunking for you. You have to do that Kind of Pythonically.

Yeah.

And you can do that a million types of ways. So if we're looking just at vector DBs, it's more so from a usability standpoint. Pinecone is like industry standard. It's quick and dirty. You can get data in, data out. It has like 4 or 5 things you could do. In terms of visibility on like the metric side, usage, when a vector was added, being able to easily configure and delete a specific vector amongst thousands of them, It's trickier if you use something like Qdrant or Weaviate. We like Weaviate quite a bit. It's much more transparent in terms of every vector, when it was added, how you can do vector management. Let's say if you accidentally vectorize the wrong thing and now you're looking for a needle in the haystack, look for that one vector that's screwing up your answer. That's a bit annoying with things like Pinecone. So I'd say there's usability. And then there's RAG, which is how do you chunk it? How do you embed it? Are you using blind chunking, which meaning every 400 characters we're going to cut it off and chunk it and embed it? Or what we're doing right now for most clients is semantic chunking, where because 3.5 Turbo is almost free, we actually let the LLM go through 2 pages of text and then summarize that as an FAQ itself based on its best judgment. And then keep doing it that way instead of relying on something programmatic where there's a lot of cutoff and overlap.

Okay, so you're summarizing it, uploading it with GPT-3.5.

Yeah, it's doing all the work. It has like a sliding window that looks like every part of the document. And then based on that slide or that window, it asks itself, if I was an X persona, what would my main questions be about this paragraph? So that way, You don't worry about random cutoffs because it knows that this is the end of a sentence because it can read English, unlike Python.

Right. And then what are you using for retrieval mainly? If something like VoiceFlow as the knowledge base, or would you use something like Pinecone so that you have more control over the write portion?

Yeah, so anything that's a chat agent will typically go with VoiceFlow. We build actual web-based apps for RAG applications where you have a web-based app. If it's local, especially from patient or anything doctor-related, they have to have it serviced on a private network. That's where we're doing RAG ourselves in Pinecone and then serving that to a UI that we also build. So that's more of an A to Z build.

So you would do everything from scratch, essentially?

Everything from scratch, from actually ingesting the information, Making it easy for them to upload brand information, doing automatic chunking and embedding, and then it gets added to their knowledge base for that agent, which is a bit more complicated obviously to build, but we've had a couple asks come up in the past year.

Okay. Are those— I guess those are more rare. You're finding like when you package the services, that's like when you kind of— it's like hot potatoes, like you sell way more of those.

Exactly. Custom builds slow you down unless they're high ticket. So obviously we've only done these high ticket for industries that are obsessed with security like HIPAA, Or SOC 2, where they need something built in a very specific way where they have full autonomy on where that data is being transmitted and how.

Right, right, right, right. Going back to AI voice, because I lost my train of thought earlier. So how did you— do you want to tell us a little bit about your new service and how you packaged that up?

Totally, totally. Sure. So we have like 3 tiers. One tier that we offer, this is more so for client-facing. We also have a separate deal we do for white labeling because we're working with other agencies now in different countries where they take our service, we negotiate a bit on the pricing, and they mark it up and sell to their clients. So they don't have to know how to build it. We're just the fulfillment partner. So we're doing kind of bifurcated delivery. For customer-facing, we have tier 1, which has like X amount of minutes, and then you get X amount of automations, let's say with Make or Zapier. Like sending a text or sending an email after the call. And then for tier 2 is where we integrate with your CRM. Could be whatever CRM you want— Zoho. Um, we're working with med spas right now, so there's this new platform in Canada called Zenoti we didn't know about, but now we're integrating with Zenoti. Uh, so that's more bespoke builds, and then we give you like 10 to 20 automations you could do using Make, or we set it up and manage it for you. Then tier 3 is our premium tier where we do voice cloning of your voice We add that in, we add everything else I just mentioned, but you get unlimited automations. So if you have a very bespoke workflow, the voice agent can trigger tons of other micro workflows as well.

Wow. Wow. That sounds really cool. And you're selling this to med spas, gym, probably less so gyms, med spas, doctors, real estate agents, et cetera.

Dentists, mechanics, mechanics, when they want to call you, say, hey, your car is ready. Can you come in? When can you come in and pick it up? That kind of thing. Or book an oil change, that That kind of thing. Anything service-oriented has been super popular with us. And yeah, we're aiming to get more into enterprise when we build that end solution that has more memory. Yeah.

Right, right. And how are you getting these small service businesses to know and hear about you?

Yeah, so again, because we did well on Fiverr, SEO-wise, I'm allowed to run ads. So as soon as AI calling came out more popular, beginning of February, I made my first ad, I started pumping ad spend behind it. So now every day we're getting 2 to 3 leads just for voice, which is awesome. And yeah, so we're— go ahead.

Google Ads, I was saying.

Google Ads. For me, we're making a big push on YouTube. So we're releasing 3 or 4 videos of us just trialing 14 different types of voice agents because this is the type of thing that someone has to see it to buy into it or more so listen to it. to buy into it. With a chatbot, it's a different type of selling method because people know what a chatbot is, especially with ChatGPT. They get it. But with voice calling, until you hear it, you don't know exactly how far we've come.

Right, right. Absolutely. Absolutely. That's, that's, that's really, that's really interesting. So it's, yeah, I think the hardest part is just education and getting people to know about you and To hear about, you know, your services, you know.

It's tricky, but when, if you have a couple testimonials or you have a couple examples you could show, you could even run Facebook ads and you just show that little snippet of you calling someone or it talking to you, something like that. One of our first projects was literally building a Santa AI caller that we posted on YouTube and we started getting some leads from that because it sounds like Santa. It collects all of the gift information that the child asks for, and then it sends a text to the parent. And even the links to the products are like affiliate links so that the actual vendor gets a kickback from anything recommended. So when we built that out and we've got some initial like laughs, those laughs turn into leads. So anything you can do to put yourself out there, especially on platforms like YouTube, LinkedIn, to show that you can actually do the thing. will put you out there much more easily.

And what's something cool that you've learned about, like any cool tools, cool things you've learned about AI lately? If, you know, it could be multiple as well.

Cool things about AI lately?

Like what's the latest things you've been playing with, experimenting with?

And yeah, so actually, yeah, back to our conversation on the whole chunking method. One thing we've been figuring out is anyone who we built a chat agent for, or like a Voiceflow chatbot, or even this AI calling stuff, but more so for the chatbot stuff, is what happens after you give them the solution. So a lot of them have needs where like, what if I just want to continually drag and drop something onto a platform and that just gets added to my Voiceflow knowledge base? So it's like it's more live and current. So one thing we've been trying to solve for is that. So it's not more so Tools, we've been using something called VoiceGlow, if you've heard of it.

No, I haven't.

So it's actually an add-on to VoiceFlow where it allows you to have front-end pieces, and it gives you the whole back-end piece you plug into your chatbot to offer things like drag and dropping files directly into a chat, which automatically gets added to the knowledge base, that kind of thing.

Cool. The thing I like about VoiceFlow is that it's deployment. You can only deploy it over a web, which is just very limiting.

Not anymore, as of a few weeks ago. So I was actually going to get to that. They added a new way. They even posted a YouTube video on it on how to do it where you could install it as an iframe. So it takes your whole page up. So you can go to a webpage, you set it up for your client, and then the whole page is kind of like a ChatGPT interface, but it's Voiceflow. Oh, so it's not just a widget, you can get a whole page chat.

Oh, so it's— yeah, you're getting it into— okay. Yeah. So essentially it's like, um, like it looks like ChatGPT.

Yeah, exactly. That's number 2. And method number 3, we just signed a client today for, is using the Voiceflow API. So the Dialog API, once you've built your chat flow, You can invoke that chat flow from any third-party app. So they're building a frontend where it has like 4 or 5 cute little agents, but behind the scenes it's just sending a request to a Dialog API and we're routing it to one specific prompt in that Voiceflow bot where it gives the impression that it's multi-agent. So there's a lot of things you could do with Voiceflow now that you couldn't do a year ago.

Yeah, this technology moves really quickly. Well, I guess the thing I don't like about it is, um, you can't really deploy easily on things like Facebook or You know, or, or Instagram. Yeah, that's what I was thinking.

There we go. I'm glad we're having this conversation so I can tell you about FlowBridge came out 3 weeks ago that allows easier integration of Voiceflow into Messenger and WhatsApp.

I've seen it. I've seen it, and I played around with it. The thing I don't like about it, well, I don't like about it, their security looks really weak.

Yeah, it is.

Which makes me very sketchy about using it. I don't want to just give out my API and they didn't even hide it. And I was like, that's security 101. That, you know, it felt like really someone hacked in their platform, then they got all of our Voiceflow API keys. Second thing I didn't like about it was you still had to create your own Facebook app or Instagram app, which is The biggest pain ever, and it'll take weeks for Facebook to approve because Facebook got the most notoriously strict, anal, bad, unresponsive, um, app review team ever on this planet. So I— and the third thing I don't like about FlowBridge— sorry, I'm going on a rant here after using it— is it doesn't ever allow you to reset the chat. So it's like natural. Yeah. It's just like, it's like if you want to go through the bot again, it doesn't let you. So someone asks you a question once, it gives you the right answer, but then it's like, how do I— and then it just like kind of ends that chat. It's like, well, how do I restart this? And it doesn't have an easy way for you to like restart it. There's not like an end button and then restart. Obviously you can't have buttons on there. I get it. But it doesn't like Natively refresh. So yeah, yeah, it's a great solution, to be honest.

It's just better than what was available there. But as of, since you asked me latest, 12 hours ago, Meta just released something that's going to be a bit of a risk to this whole ecosystem is they just released Meta AI to WhatsApp in areas in India now just to trial the product where You can go to WhatsApp and set up an agent from Meta using their infra, their AI, where you can talk to in WhatsApp itself and in Messenger yourself.

Oh, I can't wait for that. Yeah, definitely play around with that. What I actually found was easier than using FlowBridge was just to recreate the whole chatbot in ManyChat. It's more of a pain in the ass. The UI is not as great. I mean, I wish But it's just easier than doing the whole FlowBridge route that you said.

Yeah. If you want to build a pretty naive app for WhatsApp that you could do in under 10 minutes, I found this by accident completely through an engagement with someone in Germany. There's a site called timelines.ai where you just put a prompt and then you give it the number, the WhatsApp number, and then it will integrate it and add it for you in under like 5 minutes. You just authenticate. WhatsApp, and you can talk to it nonstop using 3.5 Turbo.

Oh, wait, what's the platform you used again?

Timelines.ai.

Timelines.ai.

Yeah, so let me see if I can pull it up real quick. Now watch it be magically gone as soon as I show it. Uh, let me see here.

It's super cool because Facebook just is very anal about security.

Um, So you can see here, timelines.ai, you can have multiple numbers that you do like routing to. And then all you do is you connect to your WhatsApp and it sends you a code, and that's how it does all the authentication. And once you're in, you just set up a, a system prompt on like what the agent should be able to do.

Yeah.

And I think they added a knowledge base, at least it was buggy when I was using it, but ever since I think it's been fine. You can see here you have Tracking of all of the conversations with the actual agents. It tells you the response rate, how, how fast the other user on the other end of WhatsApp is actually responding to that bot. And, yeah, you can set up some automations as well. So this is an interesting random platform that I found that actually works pretty well for WhatsApp specifically.

Interesting. I'll have to check it out after. I've never heard of that one. I've never heard that.

There you go. There you go.

Yeah. Um, and, um, anything else? Like, if you were to give someone advice, um, just as we're wrapping this up, if you were to give someone advice, um, 3 pieces of advice, uh, as they're getting started, um, what would they be to get their first client and get up and running as soon as possible?

Soon as possible. Step 1, a good client's not necessarily a paying client when you're starting out. Meaning when you are under the gun and someone's paid you to do X and you've kind of never done X before and you're figuring it out as you go, it's very stressful. So one thing that I started recommending to people and one thing that we've been doing is we've been reaching out to one nonprofit per month to build them a small application. And we typically offer something that we've never built before to try and trial it with them to see how they like it. And to nonprofits that will never have the budget for this for now, it's amazing for them. You get good karma, you're building it for free, and you're learning and you're adding to your portfolio. So when you go to an actual paid customer, you can say, I've actually done this before. Here's exactly how it works. Those discovery calls go 10 times better when you actually have something in your hand to say like, I already did this before. this is how I could switch it to make it do what you're looking for. So that confidence is if you can start building for not-for-profits or for free for individuals without, you know, killing yourself is a really good way to build that portfolio. That's number one. Number 2 is patience. I have a master's. Again, I'm only saying that not as a flex, it's more so to preface my next part. I've worked at companies like Amazon, I have a master's, and I was still selling a gig on Fiverr for $10 or $20 a pop. I did that because I wanted to build my reputation and my reviews so that I could become level 1, then level 2, and then finally top rated. So it's really patience, and you're not going to become good at this overnight, and you're also not going to become reputable overnight. It's not a hypercompetitive space, but it's gotten way more competitive in the past year as most people are getting— are trying to get the get rich quick thing. So when you offer a chatbot, 1,000 other people are offering that same chatbot. So how do you differentiate yourself? It's more so like on the patience front. What are you gonna focus on? What niche are you gonna go after? That kind of thing. And the last one is know when you need to scale. So if you're trying to just close your first client just for the sake of closing your first client, amazing. What happens when you finally get client number 10? What do you do then, right? Because then you have to worry about project management, delivery timelines. Are you going to systemize what you offer? Like, are you just so desperate to close a client for the sake of it that you're building custom things that are unrelated to each other? Meaning every single time you build something, you can't take from your past work to use for this new work. So just think of what is the path of least resistance to do something that you're a interested in building. 2, wouldn't be hard to replicate for different businesses in that niche. And 3, is easy to explain to a customer on what tweaks you'd have to make to make it perfect for them. So I know a bit long-winded, but those are the 3 things that come to mind.

Perfect. Well, thank you so much, Mark. It's always a pleasure to speak with you, and thank you for coming onto the podcast again.

Of course, my pleasure. And hopefully we can have another one a few months from now where everything changes once again.

Yeah, this industry changes so quickly that, you know, you really have to stay on top of everything.

Absolutely. Absolutely. I hope we can talk soon. Thank you, Helena.

Thank you.