100 cold emails: what I actually learned about B2B outbound

Before I automated anything, I sent 100 cold emails manually. Not because I enjoy cold email. Because I knew I couldn't build a good AI outbound system without understanding what good outreach actually looked like — what got replies, what got ignored, and what I was consistently getting wrong. Here's what those 100 emails taught me.

The setup

ICP: founders and VP Sales at B2B SaaS companies, Series A–C, 20–200 employees. I found prospects manually in Apollo and LinkedIn. I wrote each email individually. I tracked opens, replies, and what people said. I tried 8 different message angles across the 100 emails.

The numbers

Overall open rate: 41%. Overall reply rate: 6%. Positive replies (interest expressed): 3 out of 100. Demos booked: 2. One of those became a paying customer.

100 emails, 2% conversion to demo, one paying customer. Run that same ratio at 500 emails a week with a message that's improving each cycle, and you're looking at 10 demos a week from outreach alone. The bottleneck wasn't the math — it was doing it consistently without it consuming my entire week.

What I got wrong

My first 20 emails were too long. I was trying to explain everything upfront — the product, the architecture, the differentiation. No one replied. The emails that got replies were under 80 words and asked one specific question. The lesson I kept re-learning: the first email's job is to get a reply, not to explain the product.

My ICP was too broad in weeks one and two. "B2B SaaS founder" includes everyone from a solo indie hacker to a 200-person company. The replies that converted came from a very specific slice: Series A with 25–80 employees, doing founder-led sales, no dedicated SDR. When I got that specific, reply rate doubled.

I was underselling the mechanism. My early emails said "AI that runs outbound for you." The emails that got replies were the ones that named the specific mechanism: "An AI SDR running on 60-minute cycles, finding prospects, writing outreach, and handing warm replies to an AI AE automatically." Founders want to understand how it works before they trust it.

What worked

The emails that referenced a specific signal (SDR job posting, recent funding, post they wrote) had a 9% reply rate. The emails with no signal had a 3% reply rate. The research took an extra two minutes per email. The 3x reply rate made it obviously worth it.

The break-up email on the 5th touch generated 2 of my 6 positive replies. Both prospects had opened multiple prior emails. They just needed the pressure removed.

What this taught me about building Ektie

The things that separated working outreach from failing outreach were all things an AI agent can do: tight ICP filtering, account research before writing, mechanism-specific messaging, varied angles per sequence step, and consistent follow-through on the sequence. None of it required human judgment. It required process and research.

The part that requires human judgment is knowing whether the ICP definition is right in the first place, and coaching the messaging when you see what's working. That's exactly the separation Ektie is built around: agents handle the execution, humans provide the judgment and coaching that shapes it.