Naavim Outbound Engine: The Cold-Email Machine We Built for Ourselves

Why We Built It

Every agency that sells "lead generation" quietly rents the same stack: Apollo for data, a verifier, an Instantly-style sender. It works, but it has three problems - it costs $200-500 a month forever, the data caps are yours to live with, and your lead database lives in someone else's product.

We sell AI automation. If we couldn't automate our own client acquisition end-to-end, on our own infrastructure, that would say something. So the outbound engine became both our growth channel and our proof of work.

Full disclosure: this is our own internal system, not a paid client engagement. The numbers below are real production numbers from running it ourselves.

How It Works

The engine is five stages, running as scheduled jobs against a Supabase backend:

1. Acquire

Nine scrapers pull businesses from Google Maps and Indian business directories - the professional-body registries, JustDial, Grotal, and government company records among them. Eight sources are live at any time. Each scraper normalises into one `leads` table with source attribution, so every downstream stage is source-agnostic. 4,300+ leads acquired so far.

2. Enrich

A built-in enrichment pass - plain HTTP fetch and HTML parsing, no paid APIs - discovers each lead's website, extracts emails and social profiles, and hands the context to Gemini, which scores every lead 0-100 against our ideal customer profile and tags the vertical. The email-discovery yield across enriched leads is 79.7% - competitive with paid data tools, at zero marginal cost.

3. Verify

Early on we learned the expensive way that guessed emails burn sender reputation. Now an MX-verification gate sits in front of enrollment: only leads whose email survives real mailbox checks get enrolled into a sequence. Guessed patterns are off by default.

4. Send

Enrolled leads enter a 4-step sequence chosen by segment (business type × whether they have a working website). Sends go through a dedicated cold subdomain - SPF, DKIM, and DMARC configured - so the main domain's reputation is never at risk. A blacklist check runs before every single send. Brevo is the primary provider with Resend as fallback. 15,000+ emails sent.

5. Track

Provider webhooks write opens, clicks, bounces, and spam reports back to the database. Hard bounces are blacklisted automatically. A reply-checker watches the inbox and alerts us the moment a lead answers, so no reply waits more than a few hours.

The Deliverability Rebuild

The most instructive part of the build wasn't the happy path. Midway through, open rates told us our original setup - sending from the main domain, enrolling guessed emails - was hurting us. We rebuilt: moved all cold sends to an isolated subdomain, put the MX gate in front of enrollment, and paused volume until the fundamentals were right. That discipline - stop sending when the data says stop - is now baked into the system as config flags rather than good intentions.

What This Proves

• We can stand up a scrape → enrich → verify → send pipeline that replaces a $200-500/month SaaS stack with infrastructure you own outright.
• AI enrichment (Gemini scoring and tagging every lead) works at production scale on real, messy directory data.
• Deliverability is an engineering problem: verification gates, subdomain isolation, blacklists, and webhook feedback loops - not a growth hack.

If you want this machine for your own business - your verticals, your data, your infrastructure - this exact build is what we sell. We'll show you the live system on the kickoff call.