The problem with most lead lists
Everyone else scrapes one database, hands you a name and maybe an email that bounces, and calls it a lead. Half are wrong, most are dead ends, and you lose hours chasing them. We built this completely differently: instead of a list, we put a team of AI agents to work on every single lead.
A team of agents per lead
One director model does not do the work itself; it commands a team of specialist agents, and each specialist has its own sub-agents. When a lead comes in, the director assigns it out:
- One agent hunts the email, one digs for the phone, others map the company, the role, and the social profiles
- Under each, sub-agents fan out across the internet and run deep research
- The agents talk to each other: one finds a company, passes it to the next to find the right person, then to another that verifies the contact is real
- They cross-check each other's work in real time, so the output is a profile you can trust, not a guess
Grounded, not guessed
The research is grounded in official sources, not scraped from a single stale database. Stage one runs Gemini 2.5 with live Google Search, Google Maps, and URL browsing; stage two extracts a strict, validated contact record. A one-shot niche classifier decides whether a target is local (maps, messaging) or digital (professional networks, email), and a MongoDB dedup layer means you never pay to research the same contact twice across runs.
What it actually costs
Because it runs on official APIs and a lean two-model split, it is cheap to operate: measured usage is around five hundred tokens per lead, which works out to roughly one to twenty cents per verified lead depending on how deep the enrichment goes. It is built to run unattended on a wall-clock budget and pick up where it left off after an interruption.
One architecture, delivered twice
We shipped this for two clients on the same director-and-agents core: an enterprise-focused build on Vertex AI with full waterfall enrichment for a B2B software company, and a leaner, checkpoint-resilient version for an independent operator who needed the lowest possible cost per lead. Same idea, tuned to the budget and the buyer.