Isn't this just another AI SDR?

No. Those tools send whatever they generate and charge you either way. Here anything unclear goes to a person before it sends, and you only pay when a meeting happens.

Will this hurt my domain?

Your main domain never sends. We build separate domains and warmed inboxes at no charge, and unsubscribes are honored within 24 hours.

Can you guarantee a number of meetings?

No. Volume depends on your offer and your market, which is why you pay per meeting. A slow month costs you less.

What if my offer doesn't work cold?

We'll tell you at the fit check, before you pay. If we find out later, you'll hear it from us first, and you can leave with 30 days' notice.

Agents built on 10,000+ campaigns of playbooks, plus me, Erik Hernal. I review anything unclear and I'm accountable for every invoice.

What do you need from me?

A short form, sign-off on the buyer profile, your calendar link and honest feedback on the first five meetings.

Experiment Loop

Status: production on my own outbound, weekly cadence, fully automatic.

Problem

Everyone says "test your campaigns" and almost nobody defines when a test is over. So teams kill arms after 40 sends on vibes, or scale a fluke because one meeting landed on a Tuesday. I wanted the kill/scale decision taken away from mood entirely.

How it runs

Once a week, the loop scores every running campaign arm by positive reply rate (an 11-label reply taxonomy, because "unsubscribe" and "not right now" and "tell me more" are three different facts). Then it applies a two-tier statistical rule:

Offer-hunting arms need on the order of 1,000 sends per arm before a chi-square comparison is allowed to conclude anything. Below that, the loop keeps its mouth shut and keeps sending.
Optimization tests (subject lines, framing, list cuts) only run on arms that already cleared a 1 percent positive floor. You don't A/B the paint on a car that doesn't start.

Concluded experiments write what they proved to a knowledge base, then the loop proposes the next single-variable round, deploys it inside a fixed send budget, and launches. One variable per round, enforced. Tripwires (bounce rate over 2 percent, deliverability drift) auto-pause an arm without waiting for the weekly pass.

It runs full-auto and notifies me in detail afterward. The guardrails are the approval.

Stack

Node scripts, cron on an always-on VM, the sending platform's API for arm stats, Postgres for experiment state and conclusions.

Worth stealing

The two-tier rule. Most outbound "tests" are noise laundering. Deciding your minimum sample and your entry floor before the experiment starts is boring, and it's the entire difference between a finding and an anecdote.