Adam builds an AI agent in n8n that scrapes Facebook Marketplace for service requests, uses GPT-4o to classify posts by intent, and emails daily leads. The workflow pulls posts from group URLs, filters for service requests, and sends results to email, Slack, or HubSpot. You can adapt it for any niche or data source.
Chapters
It's really just a drag-and-drop workflow automator. We've just built the workflow that is the operating procedure. The AI is looking at the text of every post and saying, what do they want? This message eval AI, this is what you would say to ChatGPT. At exactly midnight, it will reprocess this Facebook group.
Dude, look at that. So this is what your AI agent sends over. This is a collection of people who are on Facebook Marketplaces looking to buy something. I see someone who's wanting to pay for swim lessons. for their kid. I see another person who wants T-ball for their children. You scrolled over earlier before we got started and you showed me that there's someone who wants to pay to have cabinets built in their house. So imagine if I was a cabinetmaker and I needed to get customers, that kind of project, what they're looking for. Can you click over?
Yeah. So if you click through here, you see the post.
Yeah, let's see the post. This kind of project is like an $8,000 to $15,000 project that they want to pay someone to do for them. And I could imagine this going and scraping Facebook Marketplace for all kinds of service providers.
Yeah.
Imagine if you're someone who just had like a pressure washer and you were trying to clean up people's sidewalks, or we're almost at the time of year when I need a lawn mowing service. I might put it on Facebook Marketplace asking for someone who could do it. If there's someone who's watching us right now who wants to go and mow lawns, boom, all they have to do is use your AI agent to scrape Facebook Marketplace and get a list of customers who are begging, who are seeking out A service provider who does that. All right. The bigger picture though, Adam, is you build AI agents. This is not just for people who are cabinetmakers who are trying to get customers. This is to understand how you do AI agents. Because when I interviewed you about the AI agents you were building and you started talking about how quickly you were building them, people said, how? I wanna know it. I wanna see it. And so you offered today to show us how you built this one very practical AI agent.
Yes, sir.
All right. What am I looking at? What's, what's this app? I don't recognize this one.
Yeah. So this, this app is called N8N. I believe their website is n8n.io. It's, if you're familiar with make.com, Zapier, even if you're not, it's really just a drag and drop workflow automator. You can, you can step through and say, I want to do this. I want to grab this, change this, and then send me an email. So the N8N makes it very easy with no code or with very little code. To just connect the dots and grab data and process it from one place and send it to another. So here we're grabbing it from Facebook. We're scraping a bunch of Facebook groups like this, this moms and dads group in Clayton County, this one in St. Louis, this other one in St. Louis. We can grab— the scraper will read all of these posts and look for who's requesting, you know, a service that I could offer. Let's say I'm doing cabinets, right? It's going to go in and it's going to find that post that says, hey, I want someone, a contractor, to install these cabinets. And I actually want to— I want to point out there's not even a great response to this one. This was 2 days ago at this point. And this isn't a— this isn't like a thing that's tapped out where every time someone posts, they always get 50 responses and you're just one of the 50. Like, if you had a good reply to this post and you had this bot, that could have told you, hey, here's a post that you should reply to, you would have been the first person to reply and yours would be the best quality reply. You probably get the quote for this, for this job.
Yeah, it looks like this one person, it took them 24 hours at least to respond and all they said was, we can help you, please reach out. There's not like pictures to cabinets that they've done before. Yeah, because these groups tend to be dead and if you have a response to someone who's urgent, who's in urgent need, it's great. Here's the thing that I'm wondering though, Adam. What's AI about all this? What I'm looking at here is a basic flowchart. If this, then that. Do this, then do that. What's the, what's the AI?
Yeah, that, that I think is actually the, the, the biggest thing we've learned here in the last couple months is that the best solution for AI is a place where the AI does like one thing very, very, very well, and it's tuned to do that job very well. Honestly, if you or I, if, if we wanted to go and scrape all of these groups ourselves, you know, without AI, without a scraper, without a bot, we could do it, right? And we could write out a process and say, okay, you load these 10 groups, you scroll down the page, you look at every message or every post that's on the page, and if it meets these criteria, drop it into a Google Sheet or a spreadsheet somewhere. We could do that, right? How long would it take you to do that for the— I don't know, there's probably 200 to 300 local groups in your area. It would take a long time. You would need several full-time employees. It wouldn't be worth it. So the magic though is it is pretty simple. You just scroll down and we could actually write out that SOP, the operating procedure to say, do this, do this, do this. Well, that's what we've done here is we've just built the workflow that is the operating procedure. But there's one thing that without AI is difficult. which is, okay, here's some random post. What does this person want? What does this person want? The AI is looking at the text of every post and saying, what do they want? And they don't always say, I'm looking for a kitchen remodeler that, you know, is in this price range. They might say, you know, I'm tired of that place I cook and I wish I could change that room. Well, the AI could say, oh, that's a kitchen and they're looking to remodel it. So the AI is going to go out and flag each one of those, process each one and say, okay, what are they actually asking for? I've read a lot of these messages now and, and a lot of them are quite vague. Oh, water leak in the basement. Help. They didn't say the word.
Right.
So if you're searching for the word plumber, then you're not going to find that message. But the AI, because it looks at every message, tries to understand the intent of the, of the post. It can then flag it and say, oh yeah, they're looking for a plumber. A plumber is the solution to whatever it is They're, they're complaining about here.
Okay. All right. I get it now. Walk me through what's going on here. Maybe give me an overview of what all these different colored boxes are. What's the text on top? And then click into some and, and tell me what they each do.
Yeah. Yeah. I can give you an overview. I'll also say like, we'll put a link to the description in the description here, a link to exactly this workflow. You could follow along on your computer. You could click through. Uh, we don't have to, you don't have to just like reproduce this from my screen. But there's only a couple steps. Look at the colored blocks. We'll walk through each one. The first one in N8N is it's a workflow. The workflow has to start. So, you know, if I was running this as a business, I would use one of these blocks that does like run daily. So every, every day at midnight or every morning you get your report. So we can talk through the setup. The next piece is, hey, there's a lot of Facebook groups. We got to get all the posts from all those groups that are relevant to you. And so we use this tool called Apify that does the scraping of those, which just literally, it just loads each one in turn and just churns out the post data. So that's this block.
So blue is scrape those groups and collect the data.
Okay. Green is the AI magic. Now that we've got all of the posts, the AI is going to look at every post in the list in turn and say, what does this person want to buy, if anything? And if they want to buy something and it matches your criteria because you've typed into the prompt here, I only want people looking for home remodel. If they meet your filter criteria, they'll drop into the report block. And the report block, in this case, I have it sending me this email. You already saw the email. It sends me this summary email of all of the things that met the filter criteria. Uh, but N8N has lots of integrations. You can click through and see them all, but I pulled out a couple here. You know, if you want a Slack message, HubSpot, ClickUp, if you use the Zoho CRM, I don't know what tool you're using. It can send WhatsApp messages, it could send text messages, whatever you want, whatever the right way is for you to process these potential leads. You, you connect it at, at this report step.
This is so easy to understand cuz I love your color blocks. And then I'm also understanding at the end, Right now you put the endpoint as send email. You could say instead of that, pull that away and then drag add to ClickUp if you want to add a task in ClickUp, or you can do both email and ClickUp.
Yeah. If I said no, I want it to go to all these places, you just drag across the lines, you connect them. Yep.
I like how simple that is. Can we take a look at the beginning again?
Yes, sir.
All right. In the beginning, what we've got is you've decided define URLs. What does that mean? Define URL as the initiation point?
Yeah. So, so one thing that, that this bot doesn't do for free for you is You actually have to go and find your local Facebook groups. It's pretty easy, you can search in Facebook for groups specifically and you could search in areas, but you've got to go and find the ones that are local to you. You're offering a local service, you need to find the local Facebook groups, and there are, depending on the size of your city, probably 50 to hundreds of these groups. I find like searching for like place name and moms is a good one, there's a lot of moms groups out there, But there are a lot, there are a lot of all sorts of these groups. So you just scroll around, you make that list. Once you've got your groups, each of those you have a URL, right? So if you copy that, the URL to that group, go over and this is where we paste in. So you see here I've got this St. Louis Moms is, is already pasted in here. So this is the thing that's going to be scraped, that's going to be processed daily.
Got it. So the way that you've got it set up right now is whenever there's an email, whenever there's an address for a web page popped in there, This whole automation kicks off. And you— could you click into the one that says delayed run daily?
Yeah. Yeah. So it's the same, it's the same deal here. You would just tell it, I want it to run, you know, I think it makes sense to run it at midnight. And you can tell the Facebook scraper, only give me posts from the last day. So, you know, every day you'll get a digest of whatever posts were made that day.
You could do— would you connect that to the other, to the URL that you showed me earlier?
Um, yeah, so yeah, you, you define it. I, I actually don't have it set right now, but I could set this in the, in the template to make it easy. Uh, set mock data. Boom.
But if I, if I'm understanding you right, you're saying you could set the delay to kick off and then from the delay it goes into this box, which says, here's the email, here's the web address. And then it goes and grabs that.
Now that I've got it connected, and if I activate my workflow every day at midnight, At exactly midnight, it will reprocess this Facebook group.
And how did you get all those different Facebook groups then if this is showing just one group?
Yeah, so you can set in a list. I think to make the demo, especially if you're new to this tool, to make the demo make sense, it'll be conceptually easier if we think about it with one Facebook group at a time. I can set up a thing in here in the copy and paste when you're ready to process 100 at a time, that will work. It also takes time, by the way. So if you pull— if you're pulling 100 messages, 100 posts out of the Facebook group, it takes 2 to 3 minutes for that 100 messages. So if you have 50 groups, there's 100 messages each, like, this tool might take an hour to run. That's okay. It's just going to run from midnight to 1 AM.
Right.
And then you'll get your email at 1:06 AM every day. What's that bottom one? You know, we don't want to sit here and stare at the tool running for an hour.
What's the workflow input trigger?
Yeah, workflow input trigger. This, this I think is, is the, the pro version here where instead of just saying run this thing every day or run it when I click this button at the bottom, which is what the middle one does, this is you have another workflow inside N8N that triggers this workflow to run. So another thing you could do, you could build another scraper that's automatically searching Facebook for Google Groups that spits out all of the local Google Groups.
Sure.
And then every day that runs and then it triggers this one. So we do a lot of like tool building in n8n where you could think of this conceptually as one workflow, but it's part of a bigger business, right? So as you try to automate more bits of the business, the workflows start to talk to each other and you end up automating bigger and bigger sections of the business. the business that way. Okay.
All right. So now that takes us into the configuration. What's the configuration mean?
Yeah, this is pretty simple. The only thing I'm setting here is how many total posts to grab. Our scraper is going to use this and it's going to use the current time to say, I only want the ones from the last day. So all we're, all we're doing is setting this up. You know, generally you wouldn't probably edit this. It's, it's, it's good to go. And it grabs— the way these workflows run is All of the data kind of builds up over time. So by the time you're through these first 2 blocks, now once you're here, you've got the data from define URLs and the data from configuration, and then you get to do something with it.
All right, I want to keep walking through this. Now the line from configuration to run Apify Scraper, that just means the next step is to run this scraper. That's a different tool, apart from your data?
This is a different tool. So Apify Apify is really powerful. You know, your customers may or may not be on Facebook. A lot of people are on Facebook, of course, but Apify has thousands and thousands of scrapers in here for Facebook, Instagram, TikTok. They have a generic one that can scrape any website. It takes a little more customization. Twitter, Reddit, I don't know, man, everything that you could imagine, they've built out a scraper for that.
So, okay.
What we do is instead of building our own scraper or trying to figure that out, we use the Apify scrapers, which are really very cheap. The scraper we're using here is $40 a month to run. It's not very expensive if you get, you know, one $10,000 deal, like it's paid for itself many, many times over. So this scraper is going to kick it off. It's going to run. And here we've— you can see a couple past runs. So here we ran it against some Facebook group. And then you don't have to log into this tool anymore. You know, the, it's all, the automation is handled from N8N, but you could see here, it's, it's grabbing from this group. Here's the post, this group, here's the post, here's the post, here's the post. Uh, it's, it's just grabbing all that context for you. So this whole block is just set up to run the scraper, wait for it to run, cuz it takes a couple minutes, and then get the data out at the end.
So, and the scraper, you, you, Set it up using Apify, and then you connect Apify to N8n.
Yeah.
N8n.
Yeah, it's hard to say.
And so then it's scraping, and then it's waiting until— so that's the next step.
Yeah. So let me—
Until it says we're done.
I could show you the configuration. This is another good example. Like, we've already— you don't have to edit this in N8n because it's just pulling. Anytime you see these green blocks here, that's a variable that we pulled from one of the last steps. So you're just telling N8N, or you're telling Apify rather, like, go ahead, do this, grab all this stuff. So grab the URL that I gave you, grab the number of posts that I told you to grab. So all that setup is in these like pretty simple blocks over here. And then this one, even though there's lots of blocks going on, like this is every time I use Apify, I copy and paste this exact setup. If you go through my, all of my N8N workflows, you'll see exactly this thing in probably 100 different places.
Okay. What's the switch one? I could see it's running, it's waiting, it's checking the status, and then switch means—
Yeah. If you, if you follow the, the workflow, you just follow the line. So you run it, you, you wait because it takes a couple minutes. So I think we, we have this set up to wait 5 seconds. Um, and then after 5 seconds we check, hey, is it, is it done? If it's done, then you'll see we'll hit the succeeded part of the switch, which spits out the data. If it's still running, which it will be for the first couple minutes, you follow the arrow all the way back around, and you drop into wait again. So you wait 5 more seconds, you check the status. So if we run this, and I'll actually just hit the run button now so you can see it. And then one of the cool things, let's see. Oh, come on. I'm going to refresh. I don't know why. Okay. So in N8N, you can run the workflow and you can actually follow through. You see, this is green. This is green. So you could see the steps that it's run in the order that it's running them.
Mm-hmm.
And so you can literally follow this process around. It's waiting 5 seconds, it'll check the status, which it's done 3 times now. It hits this, it falls back into running, it goes to wait again. So you can follow the process as it moves all the way through live, which makes it very easy to debug. It makes it easy to build, easy to edit these things, cuz you can, you can see the logic. You can see if you have branches where you're, where you're deciding something. You could see exactly what it did as it's doing it. And for each of those, you could also click in and see, oh, here's the data that it returned in that moment. So the scraper, here's what the scraper was spitting out. And, you know, so in real time, in real time, it's working. I can see it as it's running. So it makes it, it makes it very easy to develop in the platform.
So you're not connecting it. To me, it feels like after it's done, it would go to Get Data and then from Get Data it would go to AI Magic. Right now you've got What does that pretend to say?
Yeah, yeah, yeah. So, so this was another like demo magic behind the scenes thing, which is, which is because this takes multiple minutes to run, you know, an easy way to demo it is I just saved, whoops, I just saved the result from one of the exports. So see, this is exactly what you're seeing from Apify, Facebook URL, here's the username, the post, the link to it, all that stuff. So I just saved it here because it makes it easy to run over and over, which is another like N8N like pro tip. If you're running this thing and here's part of the automation that takes 10 minutes to run, well, if you're trying to develop around that, you have to wait 10 minutes every time you change something. Well, instead of that, you, you just drop in some data. You can, this little, this little pin says—
Mm-hmm.
I've pasted in the data and I've just told it return this data every time. I literally just copied and pasted it out of that. And it makes it easy as you're especially editing your, your AI prompts and all that. You don't want to re-scrape every time you want to test a new prompt. So I save the data in order to skip, you know, when I'm, when I'm doing my development, I actually have it skip entirely the scraper because I know the scraper works. The scrapers on Apify is super reliable. It's going to work every time, but it takes time. So in this case, we'll just run straight through, hit the pretend to scrape block, right? And then you can drop the AI map.
And now the pretend to scrape is going to 2 different spots, or even the get data if people were going to run this for real would go to 2 different spots. What's the first one? Message to eval AI. And what's the second one, which is merge?
Yeah.
Why not go into one?
This is the meat basically of what makes this thing special. This is the prompt. This message eval AI is the prompt that's saying— here, I'll pull it up. I'll make it bigger here. This is saying like, this is what you would say to ChatGPT, right? You could imagine You know, you could go to ChatGPT and say, hey, ChatGPT, look at this Facebook group post and tell me, is this post looking for a plumber or is this post looking for a cabinet installer? You could imagine that, right? And you just paste—
Yeah.
Post at the bottom and you could then imagine doing that 10 times or 50 times. You could pay an assistant to do that. So what this is doing is it's just setting that prompt For every single one of the messages coming through. So in order to develop the prompt, you actually don't have to iterate on the prompt inside the tool. You know, when I'm working on one of these new prompts, I literally go to ChatGPT, I paste it all in, and I go down and I say, this is the message, or this is the Facebook post. I want swim lessons for my kid. And then it would run and evaluate. And so you can test out, you can even see here I've got like some little tricks that I'm trying to give it. I've tried this a bunch of times on these groups. Read between the lines, right? They don't always say exactly what they're looking for. I've got a water leak, what do I do? Well, we know, the AI should know, that's a plumber. And then buyers, not sellers. There's a lot of people that are looking to buy, but there's just as many people posting looking to sell something, looking to offer some local service.
And we don't want— it's literally a ChatGPT chat window that you're opening up there. Are you connecting it to your ChatGPT account somehow?
Yes. So this AI is, is talking to OpenAI chat models. This is talking to GPT-4o. Um, so it's the—
And that's part of the 8N8 subscription, or you're not connecting your own personal API or anything from ChatGPT?
No, no, no, no, no. You just, you click through, you, you, there's like a little login process to connect it over to to ChatGPT. So if you have an account there, you can, you can connect the two together.
Got it. So you are using your personal account or your company account, connecting it in here. Got it. And then it's using that.
Yeah.
And then there's the structured output. What's that?
Yeah. So one thing you'll notice in the prompt here is obviously the prompt is very— I think of it as like very soft and squishy. This, this sense that like it could detect anything you're trying to use, very natural language. At the same time, N8N, this workflow tool that can move data from one place to another, is not soft and squishy. There are very hard lines connecting these blocks. And if you want to pull out the data, the data has to come out in a very predictable format, not in a, you know, human-written format. So when we—
Right.
When we write prompts like this, we're very explicit. Return in this format here. And then this is a thing you'll— if you don't understand JSON, that's fine. This is called JSON. You would, you would literally just copy and paste this and say, look, I want some— yes, looking for service, true or false. So are the— is this post looking for a service? If so, what is the service? And then the prompt will, will spit out in the result, yes, like a big list of messages that are all labeled true or false. And for the ones that are labeled true, it'll have the name of the service there.
Okay.
All right.
So you're now structuring the data in a way that the next steps could understand. Understand it. Yep. Then you're taking all the little bits, if I'm understanding that merge right, All the little responses that you've scraped off of Facebook that meet your criteria? No, all of them, it seems like. And you're merging them all together, right?
Yep. So I'm going to run it now just so you can, so you can actually see the live data in there. And so you could see for each one of the posts, I think there were 10 posts in my copy data. It just ran the AI model 10 times. And for each of those now, we want to merge together the data that we had, which was what's the post? And what's the result? So here's the result in this first post. All the post data is over here. You've seen that a couple times now. But this new column of output, yes, this one, it is looking for something. They're looking for a realtor. This one, not looking for anything. Next one, yes, looking for something, mommy and me classes. So for every post, it's going to make that decision and output in this predictable way so that the very next step is we want to filter. Are they looking for a service? If so, they get to move into the report, which is that final big block. So we're only gonna filter out, yes, they're looking for a service.
I do want, and we're gonna give everyone who's listening to this or watching this, the, this template. I do notice that you have 2 different lines coming into merge. Why?
Yeah, it makes it easy to combine the POST data and the output from the AI, because in the output from the AI, I'm just reporting This long list of looking for service, service type, looking for service, service type. But notice the post isn't actually here anymore. So we just wanna say, oh, that, that block that had the post and the block, this new block that has the, the decision, the evaluation, just combine those for me.
Got it. I see. Okay. So now you've merged those 2 and you're filtering for just the service providers and then you've got cleanup, which I'm assuming is—
Yeah. Yeah. I mean, you'll, if you, if you look at the, the current data that we've got here, there's a lot of fields. It's a lot to look at and I don't know what the legacy ID is. I don't care. I don't care about this ID. There's all this like garbage that's there that makes it hard to look at. So I'll often have these kind of cleanup steps where I'll scroll through and I'm just looking through like, oh, well, of course I want my output. So I dragged that over. That's the decision. Like, oh yeah, I want the Facebook group URL. So I dragged that over, scroll over the post URL. Yep. I want that. The time, maybe I care, maybe I don't. So in any event, when you want fields, you can just say, oh yeah, like I actually, I really want the user's name to pop up. So you just grab the name, you drag it over, you drop it into this list, and now that name will appear in the output of this block.
That looks so easy. I like the way that it does the drag and drop into this, into the cleanup section.
Yeah, it looks a lot more complicated than it is cuz you're like, oh, user.name and what's all this crap? But you really do just drag and drop everything and then they kind of do all of the math in the backend to say, oh, Well, exactly the name of that field is this. And every time you run it, you always see, oh, here's an example of what that field would have. So here's an example text from your actual data. Here's example version of text. So this is the Facebook post. Here's the group URL. So you could see it all there. So you just go through slowly and say, oh, I want this, I want this. And as you drag them, they appear in the output.
Okay, so let's see what the next step is then.
Yep. So now we've got that. We have this and the way N8N works, you actually see between each of these blocks, here's this like items count. So we started over here with 10 posts and we merged that, but then we filtered. There were only 4 in this set that were looking for a service. And so now we want to combine that into one list because I don't want 4 emails, I want one email that has the whole list. So we aggregate this. This step is very simple. There's almost no configuration. We've said, I want all the item data, all 4 items in this case, drop 'em into a single list that's called data. I think these are even the defaults of this block. You drag the block in and now you can see this block is outputting 1 item. And then we've got this like hairy thing right here we'll talk about. Um, and then you send the email. So do you wanna talk about the hairy thing?
Yeah, let's look at the hairy thing. That's the scary one because—
So, so there's code, there's code. Uh, I will tell you though, I did not write this code. And, and luckily ChatGPT, I think I've got the chat open. ChatGPT understands N8N very well. It understands the code blocks very well. And so I've got this prompt here that I've used in a lot of different ways. I tell it I'm building an N8N workflow that, you know, here's the goal of my thing. In my workflow, I've got the filtered and processed data, but I want to turn it into a nice looking email, right? I don't want just like garbage data to drop into my email. So, hey, Mr. Codeblock, I want a codeblock. Turn it into simple email, simple HTML to be injected into an email. Oh, and I specifically tell it I don't want to have to edit the code. I don't want to touch this. I just want something that just works. I don't want to talk about it.
You write it.
Yeah, yeah, you write it for me. And then I just take the data that, that we've got, you know, we have from our aggregate here. You just go and you can copy all of this data. You drop that in and scroll down. And then it gave me some HTML. I did not look at this. I did not read it. I just copied it. I dropped that into this code block and then dragged a line from the code block to send email. And at this point, the email, so it's taking it all, whatever the heck the code was, and it's gonna send me an email. And then that's exactly what we were seeing at the, at the beginning here. It, it generated out this email. So also, by the way, if you said, oh, you know, like I want this, I want. Whatever, some different piece of data. I want to actually see all the comments as well. Well, that's okay. You would just go back into your chat and say, okay, I actually want the comments as well, or I want this, or I want, you know, nice dividers between each one. So you would chat with, with this in order to generate new code that, that generates a slightly different email. But you still don't have to really understand how to code in order to use this. You don't even have to.
This is much clearer than I expected. Can you go back to that flowchart in 8? I'm keep talking slowly. And then the last item underneath is Markdown. Markdown is a nice way of displaying content. And if you want to display it in something like Notion or, um, I don't know if HubSpot does it.
That's— so I, yeah, I take this, this, I use Markdown a lot again to do like the, the development side. So this is the, the Markdown field. Um, uh, it just makes it easier without actually having to send myself 1,000 emails. I can just paste the Markdown into some simple tool. First of all, you could just read this. But oh, here's what the email would look like. So that's that's again, it's just that was just there. You could see it's not actually connected to anything. It's literally just there.
It's just so you could see what it would look like.
Yep. Got it. Yep.
All right. I think this makes a lot of sense. What other agents have you built using this type of process?
So so I think but before we even talk about other agents, like can I talk about how you would customize this for your specific use case? Because I don't think there's anyone in the world. that cares about people looking for realtors and mommy and me classes and swim schools.
Yeah.
So, the cool part about this is, you know, at this first step, you go and find the groups. So you've got to find the Facebook groups that are relevant to you. You know, I think these like mommy groups in a bunch of places are good, but there's a bunch more local groups, but you would want to go and find and spend, I don't know, an hour or 2 just looking through Facebook finding your groups. Once you've found some that seem like they're people looking for services like yours, the scraper— this is the same for every time. So no customization here, straight copy and paste. The piece that you actually would want to customize is this prompt. My prompt, in order to demo this to as many people as possible, I'm telling it, look for anyone looking for any of these kinds of services. But that's not you, right? So if you really are this like, you're— I'm a cabinetry person or I'm a home remodeler, I wouldn't have it this. I'd say, I am a home remodeler. Remodeler. Wow, it's really hard to remember. I build kitchens. I build bathrooms. I do this. This is the right type of project for me. I like big projects. I like small projects. Whatever you want, you tell the chat model here exactly what you're looking for and the type of people, the type of requests that you want to drop to you. And if you tell it that, now the only things that are going to be flagged as looking for service coming out of this block are the ones that are very specific to you. So if I change this, I'm a local swim school, we do swim lessons for K-12, and someone is an adult that's saying, I'm looking for adult swim lessons or scuba lessons, they wouldn't show up in your list, right? Because—
No.
The chat knows, like, I'm looking for swim lessons K-12. So you tell it this is the place to customize exactly what you want. You could add more notes over time as you find that it is or isn't recognizing the right post for you. But my guess is if you just write 3 or 4 extra sentences here custom to you, it will spit out exactly the right results for you every time. And that's the whole thing customized. If you just change that one block and set your groups at the beginning, you'll get your emails. Well, replace your email in this block with You know, with your email address. But other than that, if you do that, you'll get an email every day at 1 AM that has those leads.
All right, so then what other bots have you created, and what other bots do you think could be created?
Well, there's a lot. So, you know, we've got a bunch of workflows here. We've got a couple different projects all with their own workflows. We do a bunch of, of like LinkedIn searching and scraping, looking for customers. Similar idea, right? You know, you could go to LinkedIn and you'd say, oh, I want to find people that have a certain thing in their description or that are posting about a certain topic. So we use, we use more Apify scrapers that have LinkedIn posts. We drop those into, into the Similar tool and scrape that out. Here's a, here's a tool. This is a Slack bot that similar agents. So it's all of these, you know, somewhere right in the middle of every workflow, you'll find a prompt in, in nearly every single one of our, one of our workflows here. So it's all, it's all about the prompt, right? Like there's some workflow which is, okay, I want to go to LinkedIn every day. I want to search for these posts. I want to only filter out the posts that meet these criteria. And then in this case, I want to send a Slack message. So we have a thing and some of our virtual assistants, they use this almost every day to send out this Slack message looking for exactly the right person to reach out to on LinkedIn. You know, we use it for recruiting, we use it for looking for expert advice, we use it for lead gen for some of our portfolio companies. There's, there's another one here that this, this particular tool, again, you'll get bored seeing the AI agent sitting right in the middle of every one of the workflows. In this case, we have a tool that's, that's trying to process in order for business development. We want to scrape a customer's website or a potential client's website. We're going to go through and categorize every page on their site and look, okay, is this a product page? Is this a collection page, an about page? So this AI will look through and will say, okay, for every page on the site, I'm going to categorize it. And the next step down in the workflow is We then process. We have a particular way to process. Oh, are they getting the right traffic to the right types of pages? Are they generating their their traffic to the to the best performing pages, or are they sending traffic to the least performing pages? So again, it's a little tool. This tool is like it does one little piece of the job. It's not doing some magic. It's not like what I see posted on LinkedIn a lot of like we have an AI agent that replaced our team of nine people. No, no, these tools for us are making our team way more effective and letting them do— like this BD team, it's the same size it's always been, but now they run 3 to 5 times more of their process to go and do more lead gen because the tool has automated some of the most annoying parts, which is go to the website and figure out where the traffic's going, classify every page. I don't know why a human was ever doing that bit of the process. Now the AI does that.
All right. I think this is fantastic. We'll keep doing more of these. We did this based on feedback that we got from people who were curious to see how this all worked. And if you're curious about more, ask us in the comments here or reach out to me. My email address is andrew@bootstrapgiants.com. Bye everyone.
Bye. Thanks.
