← Community

Sales · AI

From 3% to 30% Open Rates: What I Learned Building Our First Sales Agent

Kiran Panjwani · October 23, 2026

From 3% to 30% Open Rates: What I Learned Building Our First Sales Agent

Everyone feels left behind by agents right now. Even my husband keeps telling me we need more agents in the business. This moment feels different because the tools are finally good enough to actually build something that works, not just something that sends emails faster.

We spent the last few months building an outreach agent for our BD person using Claude, ChatGPT, Outlook, HubSpot, and Google Analytics. I built it myself because that's how it works when you're one person does everything.

What we built and why it's different from MailChimp or HubSpot

MailChimp and HubSpot weren't delivering for us. I mean that literally. Our emails weren't hitting inboxes. They have all the automation features in the world, but what's the point if you're landing in spam?

The agent we built is different because it's designed around testing and learning every single day. Claude and ChatGPT help us generate variations of subject lines and email styles. We feed yesterday's open rates and response rates back in, and the agent suggests what to try next. It's connected to our CRM in HubSpot so we can track who opened, who responded, and what they said. Google Analytics tells us if they clicked through to our site and what they did there.

The technical setup isn't complicated, but it took time to get right. The agent doesn't just blast emails. It thinks about each recipient based on what we know about them and what's been working lately.

Testing every day but keeping the big picture in mind

We use each day as a data point. That's the whole game. If a subject line gets a 25% open rate on Tuesday, we test a variation of it on Wednesday. If a more casual tone gets responses from logistics managers but formal language works better for directors, we remember that.

The key is isolating one variable at a time. Change the subject line but keep the body the same. Change the CTA but keep the opening the same. Otherwise you have no idea what actually moved the needle.

But here's what most people miss: different people respond to different kinds of emails. You can't just find one winner and use it forever. Some people want a short, punchy email. Others want context and detail. Some respond to humor, others want pure business. The agent helps us remember these patterns and adjust for each segment.

Building slow and strong so emails actually hit the inbox

The biggest lesson wasn't about automation speed. It was about delivery.

We went from 2-3% open rates to 30% open rates and 10% response rates. That didn't happen because we sent more emails. It happened because we built something slow and strong that actually hits the inbox instead of spam.

Positive engagement improves your chances of reaching the inbox. Every time someone opens your email or clicks a link, it tells email providers that people want to hear from you. Every time your email sits unopened or gets marked as spam, it hurts you.

We started small. Sent to a handful of people each day. Watched the results. Gradually increased volume as our engagement rates stayed strong. We tailored content to each recipient instead of blasting the same message to everyone.

It was humbling. I thought the point of automation was to move fast. But moving fast to spam folders is worse than not moving at all.

The agent we built now sends fewer emails than MailChimp would have blasted out, but ten times as many people actually see them. That's what matters. You can have all the AI and automation you want, but if your emails don't land in the inbox, you've built nothing.

Written by Kiran Panjwani with Yapship.

More on Sales and AI