← All transcripts

Watch Me Build an AI Agency in 24 Minutes Transcript, AI Summary & Key Points

Chris Koerner on The Koerner Office Podcast · Nov 13, 2025 · Education · 24:31 · EN

📄 Transcript

Searchable transcript of Watch Me Build an AI Agency in 24 Minutes — Chris Koerner on The Koerner Office Podcast (24:31). Search for a phrase, then click its timestamp to jump straight to that moment in the video.

Captions sourced from the original video on YouTube, published by Chris Koerner on The Koerner Office Podcast. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.

00:00 Okay, one of my most popular episodes ever was this one right here where within 56 minutes I went and started a business and found a customer. But it's been a few months and the voice agents have only gotten better and I want to do it again, but with a twist. I want to do it in a different industry that sells an even more expensive product and I want to see if the business owner will test the voice agent themselves because once they call their own business and see that AI is answering questions on behalf of them, I

00:26 think that they will be much more likely to pay me hundreds or thousands of dollars per month for this exact service. And I want to show you just how easy it is to find the customers and to build the thing. So, let's freaking go. All right, first thing I'm going to do is go to Outscraper and I'm going to look for roofers. Why roofer? Well, they tend to be more tech-savvy.

00:46 We want to fish where the fish are. Roofers are used to paying other agencies hundreds to thousands of dollars a month for stuff like this, especially if it can help them get more jobs. Last time we did garage door repair owners, which is great, but this is much higher ticket and much more advanced of an industry. So, we're going to do something a little tricky here.

01:05 We're going to go roofing contractor, but then Google also has subcategories that roofers mistakenly list themselves in when they first register their business without knowing better. These are categories that other dirty marketers like myself are not targeting. They're only targeting the most common ones. So, we'll probably have much less competition in this category.

01:23 Roofing service. Now, if you look how Outscraper works, it lists all of the Google business profile categories from most to least popular. So, number one, doctor, dentist, insurance, attorney, real estate. Makes sense. If you scroll way, way down and there's thousands of these, thousands of categories. You've got Norwegian restaurant, dance restaurant, recycling container location, you get the idea.

01:47 So, when I start typing in the word roof, roofing supply store is number two, waterproofing, auto sunroof shop and roofing service. So, this is even less common than auto sunroof shop, which I love. There might only be 200 roofing companies per state that mistakenly listed in this category, but it's something. Okay, now I want to text people that are close to me geographically.

02:11 I live in the Dallas-Fort Worth area. I want to be able to honestly tell them that I am near them. I'm one of them. I'm a local. If they want to meet me in person and shake my hand, that is possible. This stuff goes a long way in helping your close rate go up up up. So, I'm going to go Dallas. I'll do all the zip codes. We'll do Fort Worth, all the zip codes, and then let's go Plano and Frisco.

02:36 Those are basically four of the biggest cities in the metroplex. McKinney, Richardson, they're both up there, too. Now, I want to cap the results because I don't want this to take forever and I don't want to spend a bunch of money. So, I don't need to change anything with these lead gen packs, enrichments, etc., but I do want phone numbers. So, I'm going to go to advanced parameters and I'm going to check with phone.

02:56 It has to have a phone number. Get data. All right, it's going to cost me between a buck 50 and 10 bucks. Oh, I forgot to show you my watch. So, you're just going to have to see it upside down. It's 9:23 and I started this video about 5 minutes ago. So, let's say 9:20. I started this at 9:20. We're going to go to tasks. All right, it's running. While it's running, let's make good use of our time and let's create a voice agent in GoHighLevel.

03:25 Okay, so I log in to GoHighLevel and then I'm going to go into one of my sub accounts. I'll pick this one and then I'm going to click on AI agents. Create your first AI voice agent. I'm going to build one from scratch because it's quite easy. I'm going to name it Roofers 1 TK Anders LLC voice. Let's listen to Jessica. >> Hey worldly 11 Labs friends, let's talk.

03:48 Let's have a conversation about anything. My name's Jessica and I'm looking forward to chatting with you. >> Not bad. Not bad. How about Christopher? Christopher's probably a good one. >> Chris Maguire here from Chicago. I turn complex ideas into >> No, he sounds like he's from Chicago. >> Sharing kindness is like a >> No. No. No. >> Um Okay, so this might sound silly, but I just got this weird >> Ooh, that one sounds real.

04:12 >> Here's my super conversational Scottish voice. >> No. No Scottish voices. We're going to go Hope. Hope gives me hope. I like the sound of that. Very conversational. Time zone is central. GPT-4o. Two cents per minute with 4o. Yeah, if it's recommended, I'll go with that. All right. So, here's the tricky thing. I'm not going to send the roofer the link or the phone number to this AI voice agent until I have a roofer say yes.

04:40 Because I want it to be customized and personalized to them and their business and their business name, location, etc. That's what's going to make their minds explode. So, right now I'm just going to say "Hey, you've reached Greg's Roofing. How can I help you today?" Okay? But then, when I get someone to say yes, then I'm going to plug in their personal info.

04:58 Now, I could do this in a less personalized way. I could just create a generic roofing voice agent and send the number to people, but it's just not going to pop as much. It's not going to wow them as much. Now, you can also go into advanced settings and set some parameters so you don't accidentally get charged too much. Like, what if the person on the other end of the phone never hangs up for some reason and you have an 8-hour call that cost you 100 bucks.

05:21 So, we can do a max call time, idle time, reminder frequency, agent behavior. We want the response speed to be fast. LLM temperature, just leave that at zero. Don't even worry about it. Voice settings, voice speed, voice temperature. I'm going to leave all of that as just the default settings. Okay. Now, HighLevel still needs a knowledge base. The app, the voice agent needs something to work from.

05:43 What does it talk about? How does it answer questions? So, we're not going to spend 2 hours thinking up all the FAQs, scraping them from websites. We're just going to ask ChatGPT, okay? So, I'm going to ChatGPT and I'm going to say, here's my prompt. I'm creating a simple AI voice agent within HighLevel to sell to roofing businesses. Create a knowledge base based on everything you know about the roofing business.

06:07 Basically, a ton of FAQs you can answer. Most questions will revolve around getting a quote and setting up an appointment for an inspection. I want to be able to easily swap out the name and location of the roofing business, so make that part of the knowledge base easy to find and access. And then, I'm going to say, "If this prompt isn't good enough to create this knowledge base, then just return the perfect prompt to provide you to get this knowledge base, okay?"

06:30 So, I'm asking it how to ask it in a better way. All right, so clearly my prompt wasn't good enough cuz it's now giving me the perfect prompt. I'm not going to read all of this cuz it's a lot, but it's giving me the structure, the customization section, the format, the goal, roof leak repairs, insurance claims, emergency calls. If you paste that next, I'll generate the entire AI-ready roofing knowledge base structured for HighLevel voice agent use.

06:55 I'll all include this in the show notes so you guys can use this, save you some time. All right, so now I'm just going to copy this prompt and I'm going to try two things. I'm going to paste this prompt in this window and then I'm going to open it a new window and paste it there as well and see if the output is different. Okay, so here's the output when I asked it to give me the prompt, this one, and then here's the output when I just pasted the prompt in a brand new chat window.

07:23 This is so cool. They look about the same, honestly. So, I'll just go with this one. So, it puts these custom variables in brackets here, so I can swap these out with whatever roofing customer decides to respond to me and then the agent will swap them out here within the actual knowledge base. So, I'm going to copy that. We'll go back to the voice agent, create new, create knowledge base.

07:51 Roofing one. Create. So, this is awesome. I honestly didn't know it could do this. You could just add a web crawler, you could add a link. Like, I could just add a link to the FAQs page on the roofing company's website, which would actually be a a lot easier than what I just showed you. So, add source. File upload. So, I'm just going to do rich text.

08:11 Boom, there's the variables. It allows up to 25,000 characters. I have 7,600. Call it roofing one. Now, remember, the more characters you put in your knowledge base, the more tokens you will spend cuz it's more work that the LLM, that the agent has to do. This is not money that HighLevel's charging you. This is money that HighLevel, therefore they're passing it on to you.

08:39 So, it's important that the knowledge base is thorough enough to get the job done, but we don't want to put more in there than necessary or else we'll spend more money than necessary. And right now, it's not that critical because I'm just using this for example purposes. But, if I were to actually get, or should I say when I actually get, a roofing customer to buy this from me, it will matter very much because he's going to be taking calls with this voice agent on a regular basis.

09:08 And I'm going to be passing on those costs. Okay, my knowledge base is done. Okay, so I've connected the knowledge base. The agent is active 24/7. That's the default. We'll leave it there. We want to have a number that is local to DFW. Okay, so choosing your phone number is a very important part of this because we want to choose a phone number that is local to the area that we are marketing in.

09:32 So, 469 is an area code that's local to Dallas-Fort Worth. We're going to use this number here. We're going to set the default for the agent to work 24/7. And then, let's test it. Take my phone off Do Not Disturb. All right. Let's see if this works. There it is. HL for High Level. I already have this number saved. >> Hey, you have reached Greg's Roofing.

10:01 How can I help you today? >> Yes, I need a quote on my house. I think I might have some hail damage. >> Got it. We can definitely help with that. The best way to get an accurate quote is to schedule a free roof inspection. Our specialist will check for hail damage and provide a written estimate on the spot. Would you prefer a morning or afternoon appointment?

10:21 >> Let's do 3:30 in the afternoon on November 29th. >> All right. Let me confirm that for you. You're looking for an appointment at 3:30 p.m. on Wednesday, November 29th, correct? >> Yes. >> Perfect. You're all set for a free roof inspection at 3:30 p.m. on Wednesday, November 29th. We'll confirm the exact arrival time closer to the date. Anything else you'd like to go over?

10:43 >> Where are you guys located? >> We're based in Dallas, Texas, and we serve the entire Dallas Fort Worth area. Let me know if you need anything else. >> How long have you been in business? >> We've been serving Dallas Fort Worth homeowners for years with trusted, high-quality roofing work. >> What's the name of the owner? >> The company is called Kerner Roofing, but I don't have the owner's name on hand.

11:02 If you'd like, I can connect you with someone who can provide more details. >> All right, I'm good. Thank you. All right, pretty dang good. Remember, on my knowledge base, I hadn't like customized it with location and all that. It said Dallas. I'm not sure why. It might have been in the knowledge base, but works great. Then, you can even rate the call to fine-tune your AI right here.

11:23 So, I'm going to give it a thumbs up. There's the transcript, very cool. There's the recording. And just like that, it's saved. Okay, look, if you look at the variables here, Chad GPT called it Corner Roofing. I didn't notice that, but it's pretty cool. All right, let's go check on our scraped leads. There we go. It's all done. We're at 947 so far. We're 27 minutes into this.

11:45 We already have a voice agent. We already have our leads. We're going to click on the green button to open up these leads in Google Sheets, cuz who uses Excel anymore? I mean, come on. All right, now here we are. First thing we want to do, highlight all of them by clicking this button in the top left, data, data cleanup, remove duplicates. We do not want duplicate phone numbers, okay?

12:09 So, we're going to click column H, that's where the phone numbers are, remove duplicates. 23 found and removed. All right, 978 leads left. Now, the next thing we want to do is we want to filter this. We'll highlight all of them, click the filter button, and we only want mobile phones. No toll-free, no landline, no unknown, no VOIP. Only mobile phones.

12:31 All right, so now we only have cell phones, and we only have non-duplicates. So, these are all unique roofing companies in Dallas, Frisco, Plano, and Fort Worth. There's 431 left, which is awesome. Now, here's what I'm going to do. I'm going to remove I'm going to do a control find. You click the three dots. I'm going to do space LLC. I'm going to remove all of the names that have space LLC in them.

12:56 95 of them did. Basically a quarter of them. That way it looks a lot more professional. Look, l.l.c. That's still there. Yada yada yada. So, we'll just fix that one manually. Okay. I'm going to rename that to first name, and then I'm going to fill in all of the city names if they're not there. So, I want this to look personalized. So, if for whatever reason their business does not have a city name listed on their Google Business Profile, which is where we scrape these from, I'm going to fill it in with DFW because

13:31 that's going to be another variable. And this will make sense why I'm calling these first name and last name here in a minute. Delete all of those columns that are unnecessary. All right, first name is the name of the business. Last name is the city they're in and then phone number. We're going to duplicate this again. I don't want to text 431 of them.

13:51 That's too many to text. I'm going to only text 100 of them. We'll see if that does anything. That should keep me pretty busy. So, to text 100 of them, I always like to do this little trick here. I want to randomize the list. I don't want to accidentally text only the 100 people that are in Dallas. Like if I just removed everything after 101 here, then I would only text people in these cities.

14:16 And that's not a pure study. It's not a pure sample size, right? So, I'm going to perfectly randomize it by doing the formula equals rand open and then close parentheses, and then I'm going to sort it. Sort it any way you want. Doesn't matter. And then I'm going to delete everything after 101. Also, this was edited out, but I had a ton of issues formatting the spreadsheets.

14:36 Our time-sensitive experiment here has been somewhat delayed by stupid Microsoft Excel. Okay, we have 100 leads here. They're perfectly randomized. We know the name of the business. If we wanted to make that proper and not all caps, we could, but we're not going to worry about it. Then we're going to download that as a CSV file. And then we're going to go to this texting software.

14:58 By the way, if you guys are loving this so far, which I hope you are, you need to check out playmakersai.com. That's my business. That's my community. I'm in there every week, and it's a little different from TK owners because it's only for people that are starting an AI implementation or automation business from scratch. Or you could just call it AI consulting.

15:16 I genuinely think that is the biggest opportunity right now is to help small businesses implement AI and to get paid thousands of dollars a month for it. So, check out playmakers.ai.com and we'd love to have you join our community. All right, well, that sucks. It's been a few days since I filmed what you just watched and my computer stopped recording my screen right then, right when I was ready to start using the texting software.

15:37 So, I'm going to recreate that as best as I can so you can still see how to do this. Okay, I'm going to open straight text. That is my go-to texting app here. It's a little interesting. It works with Androids as well, but it probably works best with iPhones and MacBooks. It will sync your iPhone to your MacBook. It will send them from both iMessage on your MacBook and on your phone and you'll see in real time the messages getting sent out.

16:03 Okay, so I'm going to click new broadcast and I'm going to go to upload CSV file, choose file, roofing contractor tasks, 100 valid recipients found and now I'm going to type my message to them. I'm going to say good morning. Do you still own and then insert first name here in insert last name cuz remember their quote and quote first name is their business name and their quote and quote last name is their city name.

16:34 And then I'm going to put dash Chris so they can see this is coming from a real person. We're going to click save and then we're going to click start sending. And look right here. It looks like my custom variable import worked. That's the name of the business right there and that's the name of the city. And then I'm just going to click send all. And then it's going to take about 7 to 8 minutes for all of these to send and I can see on my phone that they are sending successfully.

17:00 It pulled in the business name and it's going to send like one every 5 seconds. So, I'm going to see here in a second if it's pulling in the business name and the city name wow where and how I need it to. Okay, looks like it works. So, if you look at my phone here, you can see that these texts are being sent from my phone. No responses just yet. It's 10:04.

17:24 We've been at this for 46 minutes. Personalizing the text messages is done for two specific reasons. Number one, our response rate will be higher because it will look more personalized. Number two, our phone won't flag us for sending the same message to too many people all at once. Every message will be unique. All right, first response, who's this?

17:43 I told them who I am and where I live. Second response said yes. Now, keep in mind, it goes both ways. There is a trade-off when reaching out to roofers because everyone wants to reach out to roofers. So, it will be more competitive. The question is, is the additional competition worth still going after this industry? And the answer to that is it depends.

18:04 How much are roofers in your area getting beat up by marketers? You won't know until you try. You don't have to start with roofing. This is just for the purposes of this video. There are over a thousand different home service verticals or local business verticals. Roofing is just one of them. If I were you, I would take exactly what I'm doing and do it across three to five different industries.

18:22 Keep track of every single response, what your response rates are, what the sentiment is like. Are people mad? Are they saying mean things? Are they generally nice? Do they not have any money? And then whatever industry seems to show the most promise, just go there. Okay, and we are done. Four of them failed for some reason. They were probably bad phone numbers.

18:40 Now, let's see how many responses I got. 1 2 3 4 5 6 7 8 9. Okay, nine out of 100, which is not great. And I think I know why. I think that my approach should have been less personalized in this case. I should have said something like, "Hey, do you inspect roofs? - Chris." That's it. Cuz that would have sounded more like a potential customer. Whereas I definitely sound like a marketer here.

19:10 If I had their first name, which you can get from Outscraper, but it takes a little more time and I didn't take the time to do it. If I had their first name, I would have had a much better response rate. Oh, there's 10 10% response rate. Now, for email, this is incredible. For text, this is pretty low. But, we're going to work with what we've got. All right, here's what I'm saying.

19:26 Awesome. I know this is random, but I created an AI voice agent for your business. It'll help you quote more jobs and never miss a call. Can I just have it call you real fast so you can see how well it works? You don't need to pay me for it. It's literally custom-built for your biz. I'm hearing Dave W as well. So, if I can get him to use this for free, then I can get him to pay me for something else.

19:44 Right now, before I have zero yeses, zero customers, I need to do whatever I can do just to get them to pay attention to me. And so, yes, I'm telling him it's free because this can be free. It will be free. It could be free for 100 calls a month, free for 10 calls a month, free for this, but paid for something even more custom. I just want to get them to respond and to try it.

20:05 Okay, my response rate is up to 13% by now. All right, here we go. Even though I'm just texting, it still makes me a little nervous, I'm not going to lie. There's three, four, five. One of these guys is a little angry, so I'm not going to give him the pitch. All right, I just got an idea. I just got an idea. This is a little risky, but I kind of just want to have the AI voice agent call all of these guys proactively and then I will follow up with them whether they text me back or not and say, "Hey, that call you got, I

20:38 built that." Oh, man. Is it devious? Is it genius? Yes, to both, maybe. But, I think it might just work. And guys, you got to believe in what you're doing here. Okay, you got to actually believe that what you're offering them will help them. It'll help them make more money or save more money. It'll help them in their lives. And I do. So, I'm going into the knowledge base for the agent and I'm changing the name to example roofing company.

21:00 And then I'll be able to see in the transcript the conversations that these guys had with it. >> Okay. Okay. Thanks for reaching out. Before we schedule a call, could you please share your company name, website, and a bit more information about your AI voice agent? I'd like to review your business first. >> Hey. Okay. I'll take it. By the way, guys, I actually do do this as a business.

21:21 So, I am actually looking for customers right now. And also making a video at the same time. All right. I said, "No problem at all. My business is Repeat Leads, repeatleads.com. My name is Chris Kern. You can Google my name." Now, I'm going to take it a step further. 0843. And I'm going to go here. I'm going to find this phone number. 0843. There it is.

21:42 Creation Roofing & Foundation. And I'm going to send him a little call. I know you're busy, but the AI voice agent will call you from a 469 number if you'd like to see how it works. It's 100% AI, not me. Also, I'm not a woman. All right. He didn't answer, but I'm going to do it one more time, and I'm going to swap it out with his business. We're going to try it one more time.

22:01 Fortune favors the bold, I guess. I actually put his real business name in here this time. Guys, let me just recap. I texted 100 people, and I got two warm leads. Well, news flash, guys. Update. I got another lead, aka my warm lead rate went up by 50% from two to three. Let me show you what he said. There it is. He said, "Not today, next week." Okay?

22:21 Boom. So, in case you're not counting, 96 text messages were delivered, four failed. Of those 96, 15 have responded to me so far. Of those 15, three of them are leads, warm leads. They already know what I do, and they've shown interest in it. That's huge. That means 20% three out of 15 20% of the people that actually know why I'm texting them, cuz the other 82 people or whatever don't know why I'm texting them yet.

22:46 20% are interested in what I'm doing. That is insane, guys. That's massive. Could I close one of those three? Absolutely. Two? Maybe. Three? Probably not. We are in business here. One of them said, "Hey, would love to hop on a call, but just share your name first. I need more information about you. The next one said, "Not today, next week." And guess what?

23:06 It's 10:29. I've been doing this for an hour and 9 minutes. My response rate is already 15%. That's probably going to double because usually my response rate in the first hour is half of the response rate I get once it's all said and done. So, call it a 30% response rate, not bad at all. If we can extrapolate that out, I'm going to have 4% of these become leads.

23:25 And this isn't even optimized. I could have done this a lot better. Just imagine how many of those remaining 96 phone numbers would become leads if I proactively sent them a call from the AI voice agent, whether they respond to me or not. So, if you're watching this and you're discouraged thinking, "Only 2% are leads and of those maybe you'll only close one."

23:45 Yeah. Cuz if this is a thousand dollars a month, then I just know I need to text a thousand people to make ten thousand dollars a month, or ten thousand people to make a hundred thousand dollars a month. This is awesome. Now we know how the math breaks down. If I didn't have all these tech issues, I could have gotten these leads in about 45 minutes flat.

24:04 So, what are you waiting for? Please use my Go High Level affiliate link. They give me a little kickback if you use them. I use High Level every day. I have multiple of my businesses built on it. I can't say enough good things about it. You can build this entire platform on top of Go High Level. You can even send out these text messages through Go High Level.

24:18 And again, if you guys become AI consultants, if you want to ride this massive tidal wave called implementing AI into small businesses, then check out playmakersai.com. We'll see you next time on the Career Office.

💡 Answer

An AI agency can be built in about 45 minutes when the tools and process are ready, although the demonstrated experiment took longer because of technical issues.

🧠 AI Summary

A local AI agency can be built quickly by targeting roofing contractors with customized voice agents. The process uses Outscraper to find local leads, GoHighLevel to build an AI voice agent, ChatGPT to generate its knowledge base, Google Sheets to clean and randomize contacts, and Straight Text to send personalized outreach. Texting 100 roofing businesses produced 96 delivered messages, 15 responses, and 3 warm leads. The approach can be tested across three to five industries, with response rates, sentiment, and purchasing power tracked to identify the most promising market.

🔑 Key Points

  • Roofing contractors are a higher-ticket niche accustomed to paying agencies hundreds to thousands of dollars per month.
  • Less-common Google Business Profile categories may have less competition than common categories.
  • A personalized, local AI voice agent is intended to make prospects more likely to test and purchase the service.
  • The voice agent can use a knowledge base covering quotes, inspections, roof leak repairs, insurance claims, and emergency calls.
  • The knowledge base should be thorough enough to work but limited because more characters increase token costs.
  • The outreach experiment delivered 96 of 100 text messages, generated 15 responses, and produced 3 warm leads.
  • The outreach method was not optimized, and the initial message may have sounded too much like a marketer.
  • The same process should be tested across three to five industries while tracking responses, sentiment, and purchasing power.

✅ Actionable items

  • Search Outscraper for roofing contractors in selected local areas and require phone numbers.
  • Build a GoHighLevel voice agent from scratch and select its voice, time zone, model, and call limits.
  • Use ChatGPT to create a roofing FAQ knowledge base with easily replaceable business-name and location variables.
  • Use a local phone number, such as a 469 area code for Dallas-Fort Worth, and keep the agent active 24/7.
  • Test the voice agent by asking for a quote, scheduling an inspection, and asking about location and business history.
  • Export scraped leads to Google Sheets, remove duplicate phone numbers, and filter for mobile phones.
  • Remove unnecessary columns, fill missing city names with DFW, and randomize the leads before selecting 100 contacts.
  • Send personalized outreach using a CSV broadcast and custom business-name and city variables.
  • Offer prospects a free trial or limited free usage to encourage them to test the voice agent before purchasing a more customized service.
  • Test the outreach process across three to five industries and track response rates, sentiment, and signs of purchasing ability.

💡 Business ideas

Sell customized AI voice agents to roofing contractors and other local businesses.00:17

Create an agent that answers questions, helps quote jobs, and schedules inspection appointments.

For
Roofing contractors and other small local businesses.
Solves
Missed calls and the need to handle customer questions, quotes, and appointments.
Validate by
Build a demonstration agent, send outreach to 100 prospects, and measure responses and warm leads.
  • A roofing agent that schedules free roof inspections
  • A voice agent customized with a prospect's business name and location

🏗️ Business models

AI implementation and automation consulting15:00

Build and implement AI systems for small businesses in exchange for recurring monthly payments.

  1. Choose a local-business industry.
  2. Build a voice agent customized to the business.
  3. Demonstrate the agent to a business owner.
  4. Charge for ongoing use and additional customization.
  • AI voice agents for roofing businesses
  • AI consulting for small businesses

💰 Monetization

Recurring monthly AI voice-agent service Hundreds or thousands of dollars per month 00:27

Charge businesses for a customized voice agent and pass through ongoing usage costs.

  • Free for 100 calls a month or 10 calls a month, with payment for more custom functionality

📣 Marketing

Sales

  • Ask prospects to test the agent before requesting payment.
  • Position the agent as a way to quote more jobs and never miss a call.
  • Follow up with interested prospects and offer to let them hear the agent.

Branding

  • Use a local phone number and emphasize being nearby.
  • Sign messages with a real person's name.
  • Customize the voice agent with each prospect's business name and location.

Distribution

  • Google Sheets was used to organize the lead list.
  • Straight Text was used to send the broadcast from an iPhone and MacBook.

Customer acquisition

  • Scrape local roofing businesses from Google Business Profile categories.
  • Target businesses geographically close to the agency.
  • Send personalized text messages to a randomized sample of 100 mobile numbers.
  • Offer a free demonstration call from the AI voice agent.
  • Test multiple local-business verticals and compare results.

🧭 Frameworks

Local AI agency validation workflow18:00
  1. Select a higher-ticket local-business vertical.
  2. Find geographically nearby businesses.
  3. Build a generic voice-agent prototype.
  4. Create a customizable knowledge base.
  5. Test the agent by phone.
  6. Clean and randomize a sample of leads.
  7. Send personalized outreach.
  8. Track responses, sentiment, and warm leads.
  9. Repeat across three to five industries and choose the most promising one.

🧰 Tools & AI usage

  • Outscraper — Scrape roofing-business leads and phone numbers.00:42
  • GoHighLevel — Create, configure, test, and operate the AI voice agent.05:25
  • ChatGPT — Generate the voice agent's roofing knowledge base.05:55
  • Google Sheets — Open, clean, filter, randomize, and export the lead list.11:31
  • Straight Text — Send personalized text broadcasts from an iPhone and MacBook.15:43
  • GPT-4o — Power the AI voice agent.04:24
  • 11 Labs — Provide voice options evaluated for the agent.03:45

AI is used for

  • Generate a roofing-business knowledge base — Create FAQs and answer content for quotes, inspections, roof repairs, insurance claims, and emergency calls.05:55
  • Improve the knowledge-base prompt — Ask ChatGPT to return a better prompt when the initial prompt was insufficient.06:24
  • Answer inbound roofing questions and schedule appointments — Handle customer calls for quotes and free roof inspections.09:55

📊 Numbers mentioned

Costs

  • The Outscraper data request was estimated to cost between a buck 50 and 10 bucks.
  • Larger knowledge bases increase token costs passed through by HighLevel.

Growth

  • 96 of 100 text messages were delivered.
  • 15 recipients responded.
  • 3 respondents became warm leads.
  • The initial response rate reached 15%.
  • The warm-lead rate among respondents was 20%, or 3 out of 15.
  • The estimated final response rate was 30%.
  • The estimated lead conversion from the full outreach was 4%.

Pricing

  • GPT-4o voice usage was shown at two cents per minute.
  • A call lasting 8 hours could cost 100 bucks.
  • The agent could be offered free for 100 calls a month or 10 calls a month, with more custom work paid.

Revenue

  • The service target is hundreds or thousands of dollars per month.
  • At a thousand dollars per month, texting a thousand people could produce ten thousand dollars per month, while texting ten thousand people could produce a hundred thousand dollars per month.

Traffic

  • 947 leads were initially available.
  • 978 leads remained after removing 23 duplicates.
  • 431 unique roofing companies with mobile phones remained after filtering.
  • 100 leads were selected for outreach.

⚖️ Advantages, risks & lessons

Advantages

  • Local positioning can improve close rates.
  • Personalized business and location variables make outreach and demonstrations more relevant.
  • The voice agent can operate 24/7.
  • A live phone demonstration lets prospects experience the service directly.
  • The process can be replicated across many local-business verticals.

Risks

  • Roofing is competitive because many marketers target roofers.
  • Overly large knowledge bases increase token costs.
  • Long calls can create unexpectedly high usage costs.
  • Text outreach can produce low response rates when the message sounds like marketing.
  • Some phone numbers may be invalid.
  • Prospects may respond negatively to unsolicited outreach.

Lessons

  • Targeting less-common business categories may reduce competition.
  • Local geographic relevance can help increase close rates.
  • The outreach message should sound like a potential customer's inquiry rather than a marketer's pitch.
  • Testing multiple industries is preferable to assuming one niche will work.
  • Warm leads can be generated before the service is fully customized.

💬 Quotes

I genuinely think that is the biggest opportunity right now is to help small businesses implement AI and to get paid thousands of dollars a month for it.

States the central business opportunity.15:16

I texted 100 people, and I got two warm leads.

Summarizes the initial outreach result before an additional lead arrived.22:10

Of those 15, three of them are leads, warm leads.

States the final reported warm-lead result from the delivered outreach.22:30

👤 People & companies

Jessica

Voice option evaluated for the AI agent.

03:43
Christopher

Voice option evaluated for the AI agent.

03:56
Chris Maguire

Name used in a sample voice introduction.

03:58
Hope

Conversational voice selected for the AI agent.

04:09
Chris Kern

Business owner of Repeat Leads and sender of the outreach messages.

21:30
Dave W

Prospect identified during the outreach.

19:39
Outscraper

Lead-generation tool used to find roofing businesses and phone numbers.

00:42
GoHighLevel

Platform used to create and operate the AI voice agent and potentially send text messages.

05:25
HighLevel

Short name used for GoHighLevel in the voice-agent workflow.

08:41
Repeat Leads

Chris Kern's business.

21:30
Creation Roofing & Foundation

Prospect business contacted during the outreach.

21:42
Kerner Roofing

Company name returned by the test voice agent when asked for the owner's name.

10:58
Corner Roofing

Business name appearing in the generated knowledge-base variables.

11:31
Microsoft Excel

Spreadsheet software blamed for delaying the experiment.

14:34

🔗 Links mentioned