How AelonSync built its own AI outreach engine

We didn't want to sell AI automation to clients without using it on ourselves first. Here's what we built, how it actually works, and where a person still has to say yes before anything gets sent.

The problem we had

Cold outreach for a new agency is a numbers game, but generic outreach is a waste of everyone's time — the prospect's and ours. Writing a genuinely relevant first message to a business means reading their website, understanding what they actually do, and finding something specific worth mentioning. Doing that by hand for every prospect doesn't scale past a handful a day.

So before offering AI automation as a service, we built one for our own use: a system that does that research and drafting work, and leaves the judgment call to a person.

What it actually does

The pipeline has five steps:

• Find a prospect that matches the profile we're targeting for a given campaign.
• Research the company — reading their site and public information to understand what they do and how they present themselves.
• Find one relevant detail worth mentioning — not a generic compliment, something that shows we actually looked.
• Draft a message built around that detail, not a fill-in-the-blank template.
• Human review — every single message is read and approved by a person before it's ever sent.

Why the last step isn't optional

An AI system can misread context, get a detail slightly wrong, or draft something that's technically accurate but tone-deaf. For outreach — where a first impression is the whole point — that's not a risk worth taking to save a few minutes. So the system's job is to do the research and the first draft; a person's job is to catch what the system gets wrong before a prospect ever sees it. That split is deliberate, and it's the same principle we build into automation for clients: AI handles the repetitive research and drafting, a person keeps the judgment calls.

What it's built with

Nothing exotic: a React front end, a Node.js backend, Tailwind for styling, and LLM APIs doing the research-and-draft work, tied together with a straightforward automation pipeline. The same stack we'd recommend to a client building something similar — we don't keep a different, fancier toolkit for our own projects.

What this means if you're considering AI automation

The lesson that carried over into how we scope client work: automation earns its place by removing the repetitive 80% of a task — research, drafting, formatting, chasing follow-ups — not by trying to remove the judgment call at the end. Systems that try to automate the whole thing end up either too cautious to be useful or too confident to be safe. The ones that work leave a clear, fast checkpoint for a person.