Why does guessing at a filter usually fail?

Because a title alone hides more than it reveals. 2 people with the same title can sit in completely different buying situations, and a filter built on title and industry alone catches both the real buyer and a lot of noise that looks similar on paper.

The people who already bought are the only ground truth you have. Everything else, including the ICP slide in your deck, is a hypothesis until someone pays you.

Where they’re at in the buying cycle is perhaps the most important factor. For example, at what year in their experience do they typically run into the problem that you solve? That’s how deep you should be thinking. 

Or, are they new to their role so they have a fresh budget with which to work? If you’re targeting other founders, the answer here is almost always no budget. Always target Series A and beyond, unless your ticket price is less than $500 or so and saves them a ton of time, which equates to “happiness” in fit-rich-happy marketing psychology.

What does building the filter backward actually look like?

Open the LinkedIn profiles of your last 5 to 10 paying customers side by side. Look past title for what actually repeats: Company stage, a specific function they report into, a tenure range, even the kind of content they engage with. Use Claude Fable 5.1 or ChatGPT’s new Astra to help you deduct what stage they are at, or any other filtering information you need. 

Easy peasy, just copy and paste their profile, their company’s profile or website (or both), and any other information you can dig up on the backend of LinkedIn through Sales Navigator. Have it tell you whether it’s the exact type of person you’re looking for. 

You’re going to be doing this a lot so make sure you create a Claude/ChatGPT project for this so it’s repeatable and you can continuously refine by routinely updating project instructions.

This is the same logic LinkedIn’s own ad platform is built on now. It retired lookalike audiences in 2024 in favor of Predictive Audiences, which start from your own converted customers and expand outward from there rather than a static demographic guess. That’s an ads feature, but the underlying principle is identical for manual outreach: Your own converters are a better seed than a title you picked from a whiteboard.

Landing a secondary buyer while missing your primary one can look like traction on a dashboard. It’s a false signal. If the people replying to your outreach don’t match the pattern in your actual paying customers, you’re getting responses from the wrong folks, and no volume of connection requests fixes that. You’re effectively creating more busywork, which means you’re effectively putting yourself out of business slowly but surely 

What do you do with the pattern once you have it?

Rebuild the search filter around what repeats instead of the job title you started with. Then run a small batch, 20 to 30 sends, before you commit a full week of volume to it.

This is the same discipline behind the fintech founder who booked 12 sales calls in 1 month: A tight, converter-modeled list beat a broad one every time it was tested.

Moves to make this week:

• Pull the LinkedIn profiles and all of their Sales Navigator information on the backend for your last 5 paying customers

• List what actually repeats across them: Stage, function, tenure, budget timing, and where they sit in the buying cycle

• Write down the disqualifiers too, the traits of the customers who cost you more than they paid

• Set up a dedicated Claude project running Fable 5.1, or a ChatGPT Astra project, loaded with that converter pattern so every new prospect gets checked against it the same way

• Rebuild your search filter around the pattern instead of your original title and industry guess

• Run a 20 to 30 person test batch against the rebuilt filter before committing full volume

Steve Lewis
CEO
TORC Executive Marketing

Want the human version of executive and founder personal branding and lead generation?

Start our free 7-day course

Frequently asked questions

What if I only have 2 or 3 customers to model the filter on?
2 or 3 is enough to start, since all you need is a pattern. Add the next customer to the pattern check as soon as they close, and expect the filter to sharpen every time you rebuild it. Keep feeding your Claude project and keep having it give you new instructions to replace the old ones with when new ones come in.

What actually goes in the Claude or ChatGPT project instructions?
3 things: Your converter pattern (stage, function, tenure range, budget timing), your disqualifiers (the traits of clients you don’t want more of), and the output you want back, a yes or no with 1 sentence of reasoning. Update the instructions every time a new customer closes or a prospect surprises you, so Claude Fable 5.1 or ChatGPT Astra gets sharper with every batch instead of repeating your first guess.

Does this replace the target list sizing I’ve already worked out?
No, it feeds it. Sizing tells you how many people your list needs. This tells you which people actually belong on it, so the volume you already calculated is aimed at the right room.

Should I include people who almost bought but didn’t?
Only as a separate, smaller test list, never folded into your core converter pattern. Someone who almost bought often shares the demographic profile of a buyer without sharing the actual motivation, which is exactly the noise this method is built to filter out.