DEV Community

RyanCwynar
RyanCwynar

Posted on • Originally published at ryancwynar.com

How I Built an AI Prospecting Machine That Finds 250+ Leads on Autopilot

Last week I crossed 250 prospects in my outreach queue. I didn't find a single one of them manually.

I built an automated prospecting system that runs 5 times a day, searches Google Maps for local businesses, deduplicates against my existing queue, and adds qualified leads — all without me lifting a finger. Here's how it works and what I learned.

The Problem

I'm selling AI services to local businesses: medical practices, law firms, dental offices, CPAs. The kind of businesses that still answer phones with hold music and "press 1 for billing."

The bottleneck was never the product. It was finding enough businesses to talk to. Manual prospecting is soul-crushing — you search, you copy numbers, you check for duplicates, you add them to a spreadsheet. Repeat 50 times. By the time you're done, you don't even want to make the calls.

The Architecture

The system is dead simple:

  1. Cron jobs fire 5 times daily (10am, 1pm, 3pm, 5pm, 8pm UTC)
  2. Each job runs a different campaign with its own search strategy
  3. A prospect finder searches Google Maps for businesses matching target categories and geographies
  4. Results get deduplicated against the existing queue (phone number is the unique key)
  5. New prospects get added to a PostgreSQL database with category, location, and campaign tags

Each run typically finds 15-20 candidates and adds 3-6 after dedup. That's 15-30 new prospects per day, completely hands-free.

Campaign Strategy

I run four parallel campaigns, each targeting the same verticals but with different angles:

  • Receptionist — Pitching AI phone answering
  • Reviews — Pitching automated review management
  • AI — General AI automation pitch
  • Boring — Straightforward, no-hype approach

Why four campaigns for the same prospect pool? A/B testing at scale. Different messaging resonates with different businesses. The dentist who ignores "AI-powered receptionist" might respond to "never miss a patient call again."

The Dedup Problem

This is where most people's automation breaks down. Once you've been running for a week, 60-70% of search results are already in your queue. Without dedup, you're wasting time and embarrassing yourself by calling the same office twice.

My solution: phone number as the canonical ID. Before adding any prospect, the system checks if that number already exists. Simple, but it catches edge cases that name-matching misses — businesses with multiple locations, name changes, or slight spelling variations.

The hit rate tells you something useful too. When I'm seeing 80%+ duplicates for a geography, it's time to expand to new areas. I started in Miami and Fort Lauderdale, then expanded to Boca Raton, Palm Beach Gardens, Jupiter, Delray Beach, Coral Springs, and Plantation.

What I'd Do Differently

Start with more geographic diversity. I saturated South Florida faster than expected. Within a week, my Fort Lauderdale dentist searches were returning 90% duplicates.

Add scoring earlier. Right now every prospect is equal in the queue. I should weight by Google reviews count, years in business, and whether their website looks like it was built in 2008 (those are actually the best prospects — they need the most help).

Track the full funnel. I know how many prospects I'm adding, but I need better metrics on call outcomes. Which campaigns produce prospects that actually pick up? Which categories convert?

The Numbers

After one week of running:

  • 256 prospects in the queue
  • 5 campaigns running in parallel
  • ~25 new prospects/day average
  • Zero manual prospecting time

The ROI math is simple. If I spent 2 hours building this system and it saves me 30 minutes of prospecting per day, it paid for itself in 4 days. Everything after that is free leverage.

The Bigger Lesson

The best automation isn't the flashy stuff. It's the boring, repetitive work that you'd otherwise skip entirely. I wasn't going to manually prospect 250 businesses. I just wasn't. The automation didn't make me faster — it made the work actually happen.

That's the real unlock with AI tooling. Not doing things 10x faster. Doing things that wouldn't get done at all.

Top comments (0)