I woke up one morning with 325 new sales leads in my CRM. Not a dream, but the result of an AI agent called OpenClaw that runs my three brands on autopilot while I’m out riding or coding. I'm Billy, a BMX rider and AI engineer who’s built tools to make my life easier. Here's how I turned night scripts into a lead-gen engine.
The Problem
The Problem
My first brand—a custom BMX frame shop—showed me cold outreach was a massive time drain. I spent 10+ hours weekly chasing contacts, email addresses, and follow-ups that rarely converted. My side projects faced the same issue: a high-performance brake line kit and a maintenance subscription service needed scaling without scaling my workload.
Building OpenClaw
OpenClaw is a lightweight AI agent orchestrated with n8n, an open-source workflow automation tool. It wakes up every night at 2 AM, pulls data, talks to language models like Claude and Gemini, and drops leads into my CRM. I wired it in three stages:
Client Acquisition
- Data Capture – A Python scraper inside an n8n node pulled 1,200 raw leads from LinkedIn, directories, and event attendee lists. It logged company name, domain, and a relevance score.
- Enrichment & Qualification – I fired off two calls: one to Claude for concise company summaries, another to Gemini for sentiment analysis on recent news. Both fed into a decision node that kept leads with a score >80 and positive or neutral sentiment.
- Outreach Automation – The final step used n8n’s HTTP Request node to hit a personalized email template written by Claude. Emails included tailored hooks and Calendly links for 15-minute demos. After sending, the workflow listened for reply events and handed hot leads off to my sales team.
All runs inside OpenClaw, triggered manually, with cron jobs, or on Git pushes.
The Night-Time Run
Last week I set it to run from 2 AM to 4 AM on a cheap EC2 spot instance. Here’s the breakdown:
- Raw scraped contacts: 1,247
- Leads filtered by relevance: 562
- Leads passing sentiment check: 325
- Emails sent: 325 (one per qualified lead)
- Replies received within 24 hours: 48
- Meetings booked: 22
- Closed deals (first month): 7
That’s 325 leads while I slept, rode the park, or watched BMX videos. Best part: cost per lead was under $2.
Leveraging Claude and Gemini
Tools and Setup
I didn’t use generic templates. For each prospect, I wrote two distinct prompts—Claude for a sharp hook and Gemini for nuanced language cues:
- Claude Prompt: “Write a 120-character subject line and a 3-sentence outreach email referencing the prospect’s recent product launch and offering a free brake inspection.” Claude’s output was precise.
- Gemini Prompt: “Analyze the tone of the prospect’s last three LinkedIn posts and suggest a personalized hook.” Gemini provided nuanced hooks like “I see you’re into sustainability—our kits reduce waste by 30%.”
By alternating between models, I saw a 15% lift in open rates.
n8n Workflow Schematic
- Trigger – A simple "Schedule" node set to run daily at 2 AM UTC.
- Scrape – A custom node hitting public APIs and writing to a CSV.
- Enrich – Parallel Claude and Gemini nodes for summaries and sentiment analysis.
- Outreach – HTTP Request node sending personalized emails.
Implementation Details
Call To Action
If you want to see this in action, check out my real AI tools at axon.nepa-ai.com.