I have a confession: I hate manual prospecting. The hours of Googling businesses, copying phone numbers into spreadsheets, qualifying leads one by one — it always felt like the worst use of my time as a developer.
So I automated the entire thing.
Over the past few weeks, I have been building an AI-powered prospecting system that runs on autopilot. It finds businesses, qualifies them, queues them for outreach, and even makes the calls. Here is how it works and what I have learned.
The Architecture
The system runs as a set of cron jobs on my VPS, each targeting a different campaign vertical:
- Receptionist campaign — dental offices, law firms, and medical practices that need better phone handling
- Reviews campaign — businesses with weak online reviews that could use reputation management
- AI campaign — companies that could benefit from AI automation (CPAs, law firms, medical groups)
- Boring campaign — wealth managers, dermatologists, estate planners (the unsexy but profitable niches)
Each campaign runs at staggered intervals throughout the day. The AI agent searches for businesses in target metros (South Florida right now), evaluates their web presence, and adds qualified prospects to a Redis queue with their contact info.
How Prospect Discovery Works
The agent uses web search to find businesses matching specific criteria. For the receptionist campaign, it looks for small practices (5-50 employees) that likely still have a human answering phones. For the reviews campaign, it finds businesses with few or mediocre Google reviews.
Once it finds a candidate, it pulls their website, checks for red flags (already using AI tools, too large, out of area), and extracts the phone number. Qualified prospects get pushed to Redis with metadata about which campaign they belong to and why they were selected.
On a typical day, the system adds 15-25 new prospects across all campaigns. That is 100+ qualified leads a week with zero manual effort.
The Outreach Layer
Prospects in the Redis queue get picked up by an outbound dialer. This is where it gets interesting — the calls are made by an AI voice agent (built on Retell AI) that can have natural conversations, qualify interest, and book follow-ups.
The voice agent handles objections, answers questions about the product, and logs the call outcome. Interested prospects get flagged for human follow-up. Not interested? They get marked and we move on.
Last week alone, the dialer burned through 37 calls from the queue autonomously.
What I Have Learned
1. Campaign segmentation matters. Early on I had one generic campaign. Splitting into verticals improved qualification rates dramatically because the search criteria and pitch could be tailored.
2. Stagger your automation. Running all campaigns simultaneously looked suspicious and hit rate limits. Spreading them across the day (10 AM, 1 PM, 3 PM, 5 PM, 8 PM) keeps things smooth.
3. Deduplication is non-trivial. The same business shows up across campaigns constantly. Redis sets help, but you also need fuzzy matching on business names and phone numbers.
4. The queue is the product. The real value is not any single prospect — it is having a constantly replenishing pipeline. Even with low conversion rates, volume wins.
5. Voice AI is ready. Six months ago I would not have trusted an AI to make sales calls. The current generation (Retell, ElevenLabs, OpenAI Realtime) can handle 80% of conversations naturally enough that prospects do not realize they are talking to a bot.
The Numbers
In the past week:
- 100+ prospects discovered and qualified automatically
- 4 campaigns running in parallel
- 37 outbound calls completed by AI
- $0 in manual labor for lead generation
The whole system runs on a single VPS (Hostinger, about $15/month) plus API costs for search, voice, and AI inference.
What is Next
I am working on closing the loop — adding automated email follow-ups for prospects who show interest on the call, and building a dashboard to track conversion rates across campaigns. The goal is a fully autonomous sales pipeline from discovery to closed deal.
If you are a developer sitting on a SaaS product and dreading the sales side, consider this: the same AI tools you are building products with can also sell those products. The meta-game of using AI to sell AI is very real, and it is a massive leverage point for solo founders.
Building in public at byldr.co. Follow the journey on Dev.to.
Top comments (0)