Use AI to Build a Campaign in 2 Hours — With a Message You’re Actually Proud To Send

Welcome back! In Part 1 we dismantled the biggest AI myths. Now its time to build something real — a campaign you can run today.

Most sales campaigns are slow to ship, often add noise, and contain about as much nutritional content as a rotten Philly cheesesteak. And what has AI done for us? Well now we can feed these rotten hoagies to our leads EVEN FASTER. THANKS AI!

Now in Part 1 we told you about all the thinking traps to avoid, but you’ve done that now and your mind is clean. Part 2 is about the juice, how you actually go and build (yourself) a campaign with AI in 2 hours. Something you’re proud to send!

Now I’m assuming you have come armed with specific knowledge on your customers. Even if it’s just, “I now could write a compelling case for a customer that just bought and their options,” that’s fine.

And when you see this emoji 🟢, it means actionable tip. Prompts are highlighted in orange.

Let me give you an example of a great insight. 

EXAMPLE CAMPAIGN:

I was talking to a window washer and he often wants to work with realtors as a referral partner and he had an incredible insight, (his customer’s secret):

“They don’t need clean windows when they list, because that’s already been done. But if a realtor has a home on the market for 6 months, they often lose the home. And it turns out the windows are dirty by then and Christmas lights help sell a home. So we do both for free to build a relationship.”

🟢So we can build a small list very, quickly, “build me a list of the biggest realtors in San Francisco Bay Area, find properties of theirs that have been on the market for 5 months, are worth a ton of money, and then give me everything I need to send a message and write a three sentence email about the property + my offer to do this for free.”

Scaling this is surprisingly easy, as all of this data is on the public web, but discovering the insight wasn’t likely easy as his offer:

There’s not much you need. Buy $20/month Claude, buy $20/month ChatGPT.
Setup Claude Code (ChatGPT has CLI, which also works).

🟢Here’s when to use each

This is possibly the hardest problem of all, and it’s extra hard for horizontal SaaS. 

Think about this company’s customers and the variables that go into finding a message that can apply to a group of leads. Here are the variables that might change the message:

If we bucket these intro categories, here’s what likely changes the message:

ChatGPT tells me that just taking these 5 variables and the combinations in the screenshot, there are 1,260 unique messaging categories. JESUS H. CHRIST pin me up now—there’s a lot of pain in figuring this out.

This is why the message looks for almost everyone looks like:

“We are company X and for {department} we have {products} for {situations} for your industry and here are our features you might care about {features}. 

Do you want ANY of this I sell?”

To weed through this you need human intuition + customer conversations + data. You need to narrow the set of these 1,260 messages and turn these into customer problem segments.

The next section will teach you how. Vertical SaaS companies likely won’t have this problem.

You need a way to weed through all of these 1,260 combinations of message segments.

🟢Here are the 5 best heuristics to narrow the set:

AI’s value is not marketing to your entire market, it’s about building message-constrained segments backed by customer insights very very quickly, matching those segments to lists you can build with AI, and deploying those insights in a few days.

Once you have all the work done just drop all the data from the limited set above (which customers passed all 5 criteria) drop it all into Claude Code (good model for this) or Gemini (for longer contexts) and say, “Help me identify campaigns where these customer’s problems meet as many of the 5 criteria above and give me everything I need from the data to the message to the insight to help me organize these. Think from the lens of my customer and where they get the most value when doing this. Come up with a report that tells me which segments I should start with.”

Once you have some concepts, just write a message yourself it’s very simple, here are the guidelines:

Subject: 2-4 words, incomplete thought, where the intent is vague. 

Good: “pizza order”
Bad: Owner.com helps you avoid DoorDash and Uber fees on pizza orders
Good: “It’s terrible”
Bad: AI SDR for Blueprint

Body: Use the PEA framework, preview, engagement, and ask

Preview:
The preview should also be personal and slightly vague, you’re only looking for an open, you’re looking to NOT sound like a sales person. You can test previews here.

Engagement:
This is the meat of it all and you might not literally be able to do this step until Step 5, but you can take an existing customer where you have insider baseball and ideally this should be 1-2 sentences that is either a PQS or PVP. This is either about describing their situation (PQS: Pain Qualified Segment) based on the list build. Or a PVP where you’re crunching useful data for them from internal or external sources.

You’ve weirdly done all of the hard stuff now, let’s take note:

You’re ready to go now–this is where we get to vibe code.

🟢 CHATGPT PROMPT
(only use ChatGPT for this with with thinking + heavy thinking on in the web app):

This process is not hard, but it’s hard to do the first time. You can watch me do this live for WingWork or Texada to see this thing live.

The reason we fail deeply to deploy AI into our organization is that leaders write notes like this (Micha Kaufman – Fiverr CEO):

So here is the unpleasant truth: AI is coming for your jobs. Heck, it’s coming for my job too. This is a wake-up call.

It does not matter if you are a programmer, designer, product manager, data scientist, lawyer, customer support rep, salesperson, or a finance person – AI is coming for you.

I mean, he’s not wrong. But also, what am I supposed to do here? As a GTM leader, you have a responsibility to dive into your team’s tasks, figure out what AI can and can’t do today by getting your hands dirty, and then deploy tools and processes to erase that work.

This article is the culmination of 5 years’ worth of my work doing this…and I run my business almost entirely alone, because AI takes over more and more for me each day. 

If you want to get the most out of AI, you need to understand what your team is doing, try automating those tasks with AI, and then deploy actionable workflows internally.

Or AI will come for your job, too.

Toodles,
Jordan Crawford

Fun fact: Illustrations are hand drawn from creatives at the non profit Big Wonder. Want to help counteract the teen mental health crisis by helping young people find inspiration? Donate here

This week has been full of interesting conversations…

Someone wrote to me recently, “easy money is always the most expensive.” I love that line. Saying no to bad revenue is one of the hardest things we do as GTM operators. I’ve seen it everywhere, and I’ve done it myself, taking on customers outside the ICP for use cases that don’t fit the strategy, all in the name of growth. The result is always the same: scattered revenue streams that pull you off course.

I was talking this week with the CRO of a $50M company debating whether to enter new verticals, expand into new geographies, or sell more to existing customers. I told him to pick the last one. Over the past two years, median SaaS growth has fallen from around 30–40% to the mid-teens, while CAC payback for public companies now sits at roughly 32 months—barely improved despite all the talk of efficiency. In an environment where growth is slow and acquisition expensive, focusing on your ICP and deepening existing relationships remains the surest path to durable progress.

Become a Topline insider by joining our Slack channel.

We want to hear your feedback! Let us know your thoughts.

Sam’s Corner written by Sam Jacobs.