Relevance AI in Action: Building Custom Agents for Lead Gen – Case Study of a No-Code Solo Ecom Hustle
Key Takeaways
- AI agents are no longer science fiction. Solopreneurs can use no-code platforms like Relevance AI to build autonomous "digital employees" that handle tedious tasks like lead generation.
- Building a successful AI agent requires a hyper-specific Ideal Customer Profile (ICP). You must clearly define what a "good lead" looks like before the agent can find them effectively.
- The results can be transformative. This case study generated 85 qualified leads and saved 40 hours of manual work in the first month, leading to a 5x higher reply rate on outreach.
Did you know that 22% of sales teams have already fully replaced their human Sales Development Reps (SDRs) with AI? That’s not a typo. While everyone’s debating whether AI will take jobs, a significant chunk of the sales world has already handed over the keys.
As a solo e-commerce entrepreneur, I don't have an SDR team to replace. I am the SDR team. I'm also the marketing department, the product developer, and the customer support agent.
Frankly, the SDR part of my job—the endless, soul-crushing grind of lead generation—is the one I’d happily fire myself from. So I decided to build my own replacement: an AI employee. Here's how I did it using Relevance AI, and why it's a total game-changer for any no-code hustle.
The Solopreneur's Dilemma: Drowning in Manual Lead Gen
The Grind: Why Manual Prospecting Doesn't Scale
Let’s be real. Manual lead gen is a numbers game where the house always wins. You spend hours scraping LinkedIn, guessing at email formats, and sending out generic messages that mostly get ignored. You’re hunting for a needle in a haystack, and the haystack is the size of the internet.
For my e-commerce brand, which sells bespoke, high-end 3D printed organizers for niche hobbies, my target audience is specific. Finding them felt like a full-time job on top of my actual full-time job. I needed leverage.
The Search for a 'Digital Employee': Why I Chose Relevance AI
I started looking for something more than just a scraper or an email blaster. I wanted an autonomous system—a digital employee that could understand my goals, use tools to research potential leads, and make decisions on my behalf. I wanted a piece of that action for my own lead gen.
That's when I stumbled upon Relevance AI. It's a platform that lets you build and deploy custom AI agents with a no-code interface. Instead of just being a fancy chatbot, it lets you chain together tools, logic, and LLMs to create workflows that can execute complex, multi-step tasks. It was exactly the "digital employee" I was looking for.
The No-Code Tech Stack: An Overview
What is Relevance AI? (And Why It’s a Game-Changer for Non-Devs)
Think of Relevance AI as a set of digital LEGO blocks for building AI workers. You get access to different LLMs (like GPT-4, Claude, etc.), and you can give them "tools." These tools can be anything from a simple web search to connecting to a Google Sheet or an internal API.
The magic is that you don't need to write a single line of Python. You define the agent's goal in plain English, give it the tools it needs, and set it loose. It democratizes the creation of sophisticated AI systems.
Defining the Mission: Building a Hyper-Specific E-commerce Lead Agent
My goal was clear: create an AI agent that could autonomously find and qualify potential B2B clients for my custom 3D printed organizers.
The Mission: "Identify US-based online retailers, community forums, and influential bloggers in the [My Niche Hobby] space who do not currently sell custom organizers but have an engaged audience. Qualify them based on their site traffic, social media following, and brand aesthetic. Finally, generate a summary for each qualified lead and add it to my CRM."
Case Study: Building My Lead Gen Agent, Step-by-Step
Building this felt less like programming and more like onboarding a new (and incredibly fast) research assistant.
Step 1: Defining the Ideal Lead Profile (The 'Brain's' Instructions)
First, you have to teach the agent what "good" looks like. A garbage-in, garbage-out model applies here more than ever. I needed to be crystal clear about my ideal customer profile (ICP).
My ICP criteria included: * Business Type: E-commerce store, popular blog, or forum. * Niche: Must be directly related to [My Niche Hobby]. * Exclusion: Cannot already have a prominent category for "3D printed organizers." * Audience Size: Minimum of 10k monthly visitors (estimated) or 20k+ followers on a primary social channel. * Brand Vibe: Modern, clean aesthetic (to match my product design).
Step 2: Giving the Agent 'Tools' (e.g., Web Search, Data Analysis)
With the profile defined, I gave my agent its toolkit in Relevance AI: 1. Web Search: The ability to perform Google searches to find initial lists of potential leads. 2. Website Crawler: The power to visit a URL, read its content, and analyze it for keywords. 3. Data Analysis: A simple tool to look at the crawled text and social media numbers to see if they meet my criteria.
Step 3: Crafting the Master Prompt & Logic Flow
This is where the magic happens. I wrote a master prompt that acted as the agent's job description and standard operating procedure (SOP).
My logic flow looked something like this: 1. Search: "Use the web search tool to find 'top [My Niche Hobby] blogs 2026' and 'best online stores for [My Niche Hobby] gear'." 2. Iterate & Visit: "For each result in the top 20, visit the website using the crawler tool." 3. Analyze & Qualify: "Analyze the website's content. Does it meet the ICP criteria? Specifically, check for the exclusion keywords. Estimate audience size based on social links." 4. Decide: "If the site is a match, classify it as a 'Qualified Lead.' If not, classify as 'Disqualified' and state the reason." 5. Report: "For all 'Qualified Leads,' create a summary including the URL, a brief description, and the reason they are a good fit. Add this summary to my master Google Sheet."
Step 4: Testing and Refining with Real-World Scenarios
My first few runs were… okay. The agent found some good leads but also brought back a lot of junk. I had to refine the prompt, making the exclusion criteria more strict and teaching it to better differentiate between a casual blog mention and a full-blown product category.
The Agent in Action: From Autonomous Search to Qualified Lead
A Real Example: Watch the Agent Find and Qualify a Lead
I kicked off a run before making coffee one morning. By the time I sat down, the agent had already worked through its list. Here's a real entry it produced in my spreadsheet:
- Lead: [NicheHobbyistGazette].com
- Type: Blog / Content Site
- Summary: "High-traffic blog focusing on tutorials and gear reviews for [My Niche Hobby]. Averages 50k monthly visitors. Strong Instagram presence (35k followers). They review accessories but do not sell any custom organizers, representing a potential partnership or wholesale opportunity."
- Status: Qualified.
The agent had done in 15 minutes what would have taken me 2-3 hours of manual browsing and analysis.
Integrating the Output into My Sales Workflow
This qualified list is gold. It’s not just a bunch of random URLs; it's a pre-vetted, high-intent list. For each qualified lead, I can now use the agent's summary to craft a hyper-personalized message.
This data also becomes the perfect fuel for other AI tools. I can feed this lead information directly into a system to generate bespoke landing pages for each prospect. This creates a seamless, automated flow from prospecting to personalized pitch.
The Results: Did It Actually Work?
By the Numbers: Leads Generated, Time Saved, and Initial ROI
In the first month of running the agent for about an hour a day, the results were staggering.
- Leads Generated: 85 highly qualified B2B leads. (My manual process would have yielded maybe 20 in the same time).
- Time Saved: I estimate I've reclaimed 40 hours of manual prospecting work.
- Initial ROI: Of those 85 leads, I’ve already closed deals with 7 of them for wholesale orders. Signal-based outreach like this achieves 5x higher reply rates, and I absolutely felt that.
Lessons Learned: What I’d Do Differently Next Time
My biggest lesson? Be even more specific in the initial prompt. I wasted a few early cycles because my definition of "brand aesthetic" was too vague. I eventually refined the prompt to include instructions like "Analyze the site's CSS for primary colors and font choices to determine if the aesthetic is 'modern' or 'traditional'."
Conclusion: Your Turn to Build an AI Employee
The era of the AI agent is here, and it’s not just for venture-backed tech giants. For solopreneurs, these tools are the ultimate force multipliers. They are the tireless, 24/7 employees we could never afford to hire.
The automation of tedious admin and research tasks is a freedom machine for creators and builders. Don't wait. The barrier to entry has never been lower.
Key Takeaways for Your Own Ecom Hustle
- Define Your ICP Religiously: Your agent is only as good as the instructions you give it. Be ruthlessly specific.
- Start Simple: Begin with a basic search-and-qualify agent. You can add more complex tools and logic later.
- Iterate, Don't Procrastinate: Your first version won't be perfect. Run it, see what breaks, and refine the prompt.
- Connect Your Tools: The real power comes from connecting your lead gen agent to the next step in your sales process.
A Simple Prompt to Get You Started in Relevance AI
Want to try it yourself? Here’s a super simple prompt structure you can adapt for your own agent.
Role: You are a Lead Generation Assistant for my company, [Your Company Name], which sells [Your Product].
Goal: Your objective is to find 10 potential customers who fit our Ideal Customer Profile.
ICP: * Industry: [e.g., Home Goods E-commerce] * Size: [e.g., Small business with 5-20 employees] * Signal: [e.g., Recently posted on their blog about 'home organization']
Steps: 1. Use the web search tool to find companies matching the industry and size. 2. For each company, visit their blog and search for the signal. 3. If a company matches all criteria, add their name, URL, and the blog post link to the final report. Otherwise, discard them.
Output: A list of 10 qualified leads.
Go build your first employee. You can thank me later.
Recommended Watch
💬 Thoughts? Share in the comments below!
Comments
Post a Comment