The Growth Hypothesis

I'm Ariel Cohen, AI researcher turned founder turned growth engineer.

---

On paper, I was the wrong hire.

M.Sc. in deep learning. Published researcher. Never ran an ad campaign. Never carried a number.

Today I'm Swan's Founding Growth Engineer.

Here's how that happened: a research problem I stumbled into, and a mistake I almost didn't notice I was making.

A year ago I started my own AI agency, building voice agents for clients. Like every founder, step one was leads. So I studied how the best operators generated demand, and landed on LinkedIn.

Eight months later: 7 million people reached. 40K followers. $100K closed, solo.

Every week a post would pop, and I'd feel like a genius.

Then the next week would come, and I couldn't answer one simple question. Who were those people?

Someone at a dream account liked my post Tuesday and visited my site Wednesday. By Thursday, they'd viewed my profile. By Friday that thread was gone. I had reach, not infrastructure.

Content without a system to catch it is inefficient.

That's the realization that actually made me the right hire.

No campaigns. No quota. Swan bet on what actually produced the 7M: form a hypothesis, build the instrument, read the signal, kill what doesn't work, automate what does, run it again. Research, pointed at humans instead of datasets.

Swan is trying to build an autonomous business. $10M ARR per employee, not headcount. Growth here means building engines: systems that watch a signal, do the research, and hand you revenue while you sleep.

Here's what that's looked like since I joined. A couple examples, out of a list that's still growing:

I turned on the one signal almost nobody watches, the "who viewed your profile" tab. 1,087 views came in. 435 were ICP-qualified. 254 landed in our Gold and Silver tiers. 14 turned into MQLs. $230K of pipeline in 14 days, out of a tab most people never open.

The second: $178K of expansion pipeline showed up in the first 10 minutes, pulled from hidden revenue that was already sitting inside our own customer base: past conversations, usage data, mapped against every buying committee we'd touched. Nobody cross-references that by hand. It's recurring now, running quietly in the background all year.

Combined, those two took about a day to build. $408K of pipeline since. Neither needed new data. It was sitting in our stack already; I just hadn't built anything to read it.

That's what this newsletter will be. The GTM engines we're building at Swan, with the real numbers and the steps to copy them: MQL scoring. Signal stacking. Content-to-pipeline systems. Where AI actually fits into GTM.

If you want a front-row seat to what growth looks like when it's engineered instead of staffed, you're in the right place.

Talk soon, Ariel