Skip to content
← Back to work

AI Outreach Systems

Turning two dormant contact databases into working outreach, without either firm hiring anyone to run it.

Role

Sole engineer — for two client firms

Context

Two real estate investment firms

Status

In production

Verified by

Live campaign delivery · carrier registration

Results

100K+

Contacts enrolled · In live campaigns

Outreach system completely run by my custom-built AI pipeline, handling over 100K+ contacts — producing leads and generating revenue, all while staying fully compliant with regulations.

The problem

Two real estate investment firms were sitting on large contact databases they had no capacity to work. The contacts were real and the interest was plausible; there was simply nobody with the hours to make the calls.

This is the kind of problem automation is genuinely good at — high volume, repetitive, and bounded by rules. It is also the kind where doing it carelessly is worse than not doing it at all, because the failure mode is regulatory rather than merely embarrassing.

Constraints

  • Carrier compliance and DNC rules are hard limits, not guidelines.
  • Sending too quickly burns the phone numbers, which ends the channel permanently — pacing is a functional requirement, not an optimization.
  • Neither firm could add headcount to operate whatever was built.

How I built it

The first system runs voicemail-drop campaigns across a 100K+ contact CRM, with drip throttling and delivery windows so volume is spread the way a carrier expects to see it rather than in bursts that get numbers flagged.

The second is a Python pipeline that takes a contact list, generates a personalized script per contact, synthesizes it in a synthesized voice, and delivers it through an API — with idempotency and send caps, because the expensive mistake in outreach automation is sending the same message twice.

Underneath both is the unglamorous part that makes them legal to run: full A2P 10DLC brand and campaign registration, end to end.

What I’d do differently

I would instrument reply-quality tracking from day one. Delivery was measured from the start; whether the messages actually landed well was something I could only reason about later.

Stack

Python · REST APIs · CRM automation · voice cloning · A2P 10DLC compliance