Most founders hate prospecting. The research, the data entry, the endless Googling — it is soul-crushing work that eats hours and produces inconsistent results.
So I automated the whole thing.
Over the past few weeks, I have been running an AI-powered prospecting system that finds, qualifies, and queues leads 24/7. No human intervention required. Here is how it works and what I have learned.
The Problem
I am building AI voice and automation products for small businesses — medical practices, law firms, dental offices, CPAs. The kind of businesses that answer phones all day and could benefit massively from AI.
But finding these prospects manually is brutal. You search Google Maps, check reviews, verify phone numbers, cross-reference against your existing list, and add them to a queue. For each good lead, you waste time on five duplicates or dead ends.
I needed a system that could do this while I sleep.
The Architecture
The system runs on scheduled cron jobs throughout the day. Each job targets a specific campaign (receptionist, reviews, AI assistant) and geography (South Florida to start).
Here is what happens each run:
Search — The AI agent queries for specific business types in targeted cities. Dentists in Boca Raton. Law firms in Fort Lauderdale. Medical practices in Palm Beach.
Qualify — Each result gets checked against basic criteria. Do they have a phone number? Are they the right category? Do they look like a real business?
Deduplicate — This is critical. The system checks every prospect against the existing queue by phone number. No duplicates get through.
Queue — Clean prospects get added with campaign tags, contact info, and category metadata.
Log — Everything gets written to daily memory files so I can review what happened.
The Numbers
In one week, the system grew my prospect queue from about 175 to over 256 qualified leads. That is roughly 80 new prospects found, verified, and queued — completely hands-off.
Each run adds 3-6 new prospects and skips 4-10 duplicates. The duplicate rate increasing over time actually tells me something useful: I am saturating my current geography. Time to expand.
What I Have Learned
Saturation signals are valuable. When your duplicate rate climbs above 60%, it is time to expand to new areas or categories. The system naturally tells you when a market is tapped out.
Campaign segmentation matters. Different campaigns target different pain points. A receptionist campaign focuses on high-call-volume practices. A reviews campaign targets businesses with poor online presence. Each campaign has its own search strategy and qualifying criteria.
Scheduling beats batching. Running 4-5 smaller prospecting sessions throughout the day produces better results than one massive session. Search results vary by time, and spreading the work prevents rate limiting.
Memory is everything. The system writes daily logs that I can review. When something goes wrong — bad search queries, too many duplicates, wrong categories — I can trace it back and adjust. Without this audit trail, automation is just a black box.
The Stack
Nothing exotic here:
- Cron jobs for scheduling (running on my own gateway)
- AI agent (Claude) for search, qualification, and data entry
- PostgreSQL for the prospect queue
- Daily markdown files for logging and memory
The whole thing runs on a single VPS. No fancy infrastructure, no expensive SaaS tools. Just an AI agent with clear instructions and a schedule.
What Is Next
The prospecting pipeline feeds into an AI voice calling system that actually calls these businesses. That is a whole other post. But the key insight is this: the boring work is the perfect work to automate first.
Nobody dreams of spending their day Googling dentists in Fort Lauderdale. But somebody has to do it if you want to grow. Might as well be a machine.
If you are building a B2B product and still doing manual prospecting, you are leaving time on the table. Set up a simple pipeline, let it run for a week, and see what happens. You might be surprised how much ground an AI agent can cover while you focus on what actually matters — building the product.
Top comments (0)