Launch an AI Voice Agent Business for Local Services in Under an Hour

Introduction to AI Voice Agents for Local Businesses

One of my most popular episodes chronicled how I launched an online business and secured a customer within 56 minutes. Building on that success, I wanted to explore the rapidly evolving landscape of AI voice agents, which have become significantly more sophisticated. My goal was to replicate the previous feat, but with a twist: target a different, more expensive industry, and observe if a business owner would be willing to test the AI voice agent themselves. The hypothesis is that once business owners experience AI handling inquiries on their behalf, they'll be far more inclined to invest in such a service, potentially paying hundreds or thousands of dollars per month. This video aims to demonstrate the simplicity of identifying potential clients and deploying an AI solution.

Finding Roofing Leads with Google Maps Scraper

My first step was to utilize Outscraper, a powerful tool for extracting business data from Google Maps. I chose to focus on roofing contractors, as they typically understand the value of investing in services that can generate more jobs, often already allocating budgets for lead generation. Unlike my previous venture into garage door repair, roofing projects tend to involve higher ticket values, making the service more attractive for clients. A key strategy was to target less obvious Google Business Profile categories where roofers might inadvertently list themselves, reducing competition from other marketers. I specifically searched for "Roofing Contractor" and the less common "Roofing Service" categories. For geographical targeting, I focused on the Dallas-Fort Worth metropolitan area, including Dallas, Fort Worth, Plano, and Frisco, to ensure I could genuinely offer local, in-person support if needed. I limited the results to 1000 and ensured that each entry included a phone number, which was crucial for outreach.

Setting Up Your AI Voice Agent in GoHighLevel

While Outscraper was busy gathering leads, I logged into GoHighLevel, a comprehensive marketing platform, to set up the AI voice agent. Navigating to the AI Agents section, I opted to create a custom agent from scratch. I named the agent "Roofers 1" and assigned my business name, "TK Owners LLC." For the voice, I sampled several options and selected "Hope," a friendly and conversational American English female voice, which seemed ideal for a business context. I set the time zone to Central (Chicago) and chose GPT-4 as the underlying language model, noting its recommended status and cost of approximately 2 cents per minute. The initial total message for the agent was kept generic: "Hey, you have reached Greg's Roofing, how can I help you today?" The plan was to customize this message with the specific business name and location once a potential client showed interest, making the interaction highly personalized.

Crafting a Knowledge Base with ChatGPT

For the AI voice agent to effectively answer questions, it needed a robust knowledge base. Instead of manually compiling FAQs, I turned to ChatGPT. My prompt instructed ChatGPT to create a comprehensive knowledge base for an AI voice agent designed to handle inbound and outbound calls for a roofing business. The main goals were to answer common questions about roofing services, provide basic company info (hours, location), gather lead info (name, address, roof type), and schedule free roof inspections or quotes. I requested 50-100 FAQs and answers in a natural, conversational tone, focusing 70% on booking questions (hail damage, insurance claims, roof replacement) and 30% on general info (service areas, materials). Crucially, I asked ChatGPT to embed customizable variables like `BUSINESS_NAME` and `SERVICE_AREA` to allow easy swapping of business-specific details. This generated a highly structured and detailed knowledge base, ready for import into GoHighLevel.

Testing the AI Voice Agent and Refining Settings

With the knowledge base created, I proceeded to configure the agent's phone and availability. I selected a local 469 area code number (Dallas-Fort Worth) for the agent, ensuring it would appear as a local call. The agent was set to be active 24/7. I then performed a test call to my own phone. The AI voice agent successfully answered, confirmed an appointment for a roof inspection, and provided details about the company's location. Although it initially stated "Dallas, Texas" instead of the specific business name I had set, the overall interaction was impressive. I could access the call transcript and recording, allowing for detailed review and fine-tuning. Advanced settings provided options to control maximum call time, idle time, and response speed, which I set to "fast" for optimal conversational flow. This initial test confirmed the agent's ability to handle basic inquiries and schedule appointments effectively.

Cleaning and Preparing Leads for Outreach

After the Outscraper task completed, I had a Google Sheet filled with hundreds of roofing contractor leads. The next crucial step was to clean and prepare this data for cold outreach. First, I highlighted all data and used the "Remove duplicates" feature, focusing on the phone number column to ensure each contact was unique. This reduced the list significantly. Next, I filtered the phone numbers to include only mobile phones, excluding landlines, toll-free, and unknown numbers, as these are generally less effective for cold texting. I also standardized the company names by removing suffixes like "LLC" to ensure a cleaner, more professional appearance. For entries where the city name was missing from the Google Business Profile data, I manually filled them in with a relevant DFW city. Finally, I organized the data into a simplified format of "First Name" (business name), "Last Name" (city), and "Phone" to facilitate easy import into the texting software. To avoid texting the entire list of 431 unique mobile contacts, I randomized the order and selected the first 100 for my initial outreach.

Executing Personalized Cold Text Outreach

I used a texting application (ImageText) that syncs with my iPhone and MacBook to send personalized cold text messages. This software allows for bulk messaging while ensuring each message is unique, which is critical for avoiding spam filters and maintaining a personal touch. I uploaded the CSV file with the 100 randomized leads. The message I crafted was: "Good morning, Do you still own [First Name] here in [Last Name]? - Chris." The `[First Name]` variable would insert the business name, and `[Last Name]` would insert the city. This personalization is key to higher response rates and avoids mass-message flagging. I initiated the broadcast, and the software began sending messages at a rate of approximately one every five seconds. I observed in real-time as the messages were delivered from my phone.

Analyzing Results and Scaling Your AI Business

Out of 96 delivered text messages (4 failed), I received 15 responses, representing a 15% response rate. Of those 15, three were warm leads, showing genuine interest in the AI voice agent's capabilities. This translates to a 20% conversion rate from initial responses to warm leads. This is a significant result, especially considering the relatively low personalization in the initial outreach. For example, one lead expressed interest in seeing how the AI worked and requested a call. Another lead said "not today, next week," indicating future interest. This experiment wasn't fully optimized; with more detailed personalization (e.g., individual owner names) and proactive AI voice agent calls, the conversion rates could be even higher. The takeaway is clear: this business model offers immense potential. Even with a 4% lead conversion rate from initial texts, one could generate substantial revenue by scaling operations. This demonstrates a massive opportunity for AI consultants to help local businesses implement AI solutions, saving them time and money while increasing their efficiency. By iterating on outreach strategies and AI agent capabilities, the potential for growth is exponential.