Stop narrowing your ICP down to one industry before you have any data. New agency owners do this constantly. They pick a niche off a course, go all in, and wonder why the emails aren't landing. Here's what recommend instead. Build the exclusion list first. Remove the industries you genuinely don't want. Hate working with legal? Cut legal. Everything else stays open. That keeps your total addressable market wide while you're still figuring out what actually resonates. Then send the broad campaign. Same offer, every industry that's left. Give it two months. By then the replies tell you where the money is. One or two industries will be answering way more than the rest. Now you narrow. Double down on the niche that's replying. Write copy just for them. Build the next list around them. You didn't guess the niche. The market picked it for you, and you've got the reply data to prove it. Which industry ended up being your niche without you planning it?

Generating a lead is only the first step. The real opportunity starts after the form is submitted: Lead captured → response → qualification → conversation → appointment A strong backend can turn the same traffic into more opportunities without constantly increasing ad spend. Where are your leads dropping off right now?

I usually send around 21 emails per day from one mailbox, divided into 7 new emails and 14 follow-ups. This setup works well for me. I send different types of emails and usually get a good number of replies. However, some of my emails are currently landing in spam. What are the best Instantly settings I should use to improve deliverability and avoid emails going to spam? I’m already sending a relatively low number of emails compared to the usual limits, so I’d like to understand what settings I should adjust.

An Insider member asked me something on the AMA this week that I get a lot. Their campaigns are working. A little over 2% reply rate on a couple of them. Zero booked calls. They'd built a half-working reply bot and wanted to know how to speed things up. I gave them the answer they were hoping I wouldn't give. Park the bot. Do the replies by hand for the first two to four weeks. The positive replies are usually the hardest part of cold email. They already solved that. What's left is turning a sure into a booked call. And you can't automate a conversion you've never done yourself. Those first weeks are your training data. What do people actually ask. What answer moves them to the calendar. You'll split test faster than any prompt you write blind. The replies themselves are simple. Short. Direct. Answer the question if they have one. Push them to the calendar. If they don't book, follow up in a day. Then a day. Then a day. Shoot them a text. Shoot them a LinkedIn message. Stay on it until they act. Once you've got a few weeks of real conversations, train the agent on those. Before that it's guessing. And long, wordy replies are the opposite of what converts.

Most people never notice. Every prompt goes into one bucket. Every sequence drafted. Every subject line tested. Every revision made. When that bucket fills up, the model starts cutting corners. Generic output. Forgotten instructions. Copy that feels like it was written on autopilot. There is a name for it: context rot. Two tools built specifically to fix it. Went straight to GitHub. Checked every fork. Found the dominant versions by star count. GSD: 65,000 stars. Oh My Claudecode: 37,000. 102,000 combined. Almost nobody in this community is using either. I had both installed before dinner. GSD structures what goes into Claude's context at each stage of your project. Instead of flooding the window with everything, it feeds Claude exactly what it needs, when it needs it. One command installs 68 global skills that activate automatically. Oh My Claudecode handles multi-agent orchestration. You describe the task. It splits the work across specialized agents and runs them in parallel. Automatically. Both free. Both open source. GSD: npx get-shit-done-cc@latest Oh My Claudecode: /plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode Jay built this community around test-iterate-improve. Most people apply that to sequences and subject lines. Almost nobody applies it to the AI layer itself. You have already felt the drop in quality on longer sessions. Now you have a name for it. And a fix. Ian Kirk