You don't find your ICP in a workshop, you find it in the reply column: how to test an ICP with cold email
You don't find your ICP in a workshop. You find it in the reply column.
I've sat through persona sessions with sticky notes and invented names like "Growth Greg." They feel productive. They're guesses with nicer handwriting. The market answers the question faster and more honestly, if you let it.
Two calls on May 5 made that obvious to me again.
Real estate showed up, doctors didn't
The first was a weekly with a coaching client running a podcast campaign. We'd aimed invitations at several kinds of business owners at once. When we went through the results, the split was stark. In the client's words: "I've got, you know, the real estate people, the ones who are really showing up." Real estate professionals had the highest response by a distance, and an investor group was on the calendar.
Doctors? Zero responses. Food and health? Zero.
Nobody would've predicted that in a workshop. If you'd asked this client to pick an ideal customer before the campaign, doctors would've looked great on paper: high income, a real problem, budget. The market said no. Real estate, which might've been a third or fourth choice, said yes.
That's the whole case for testing. Your best guess about your buyer is usually a story you like about your buyer.
Brute force through quantity
The second call that day was Charles with a seed-funded commercial real estate startup. They were already doing their own outbound, cold calls and cold email, and getting about six or seven leads a week. Respectable. The founder believed he knew who his buyer was.
Charles pitched the opposite of certainty. His phrase was ICP testing as a brute-force approach through quantity: test multiple ICPs, with various offers, at the same time, and let the replies decide.
The setup he described is simple to picture. A hundred mailboxes on separate sending domains. A seven-day warming period. First leads from day eight. On a .com domain rather than the .edu addresses they'd been sending from, because a dedicated setup scales and protects the main brand. The founder's old setup could never have run a real test. Six or seven leads a week from one channel is enough to keep a pipeline alive, not enough to compare five audiences.
The point of that much capacity isn't to blast one audience harder. It's to run five audiences at once and find out in weeks what would take a quarter if you tested them one at a time.
How to structure a multi-ICP test
This is how we set one up.
- Pick four to six segments that each could plausibly be your buyer. Make them genuinely different: different industries, or different roles in the same industry.
- Give each segment its own campaign and its own list. Never mix them, or you won't know who replied to what.
- Keep the copy structure the same across segments and change only the details that have to change. If the copy differs wildly, you're testing copy, not the audience.
- Judge on replies and booked calls, nothing else. Opens lie.
Then wait for real volume before you call it. A segment that's been emailed for three days hasn't told you anything yet. One that has had real volume and produced nothing has.
Offers, not just audiences: test both
The part people miss: sometimes the audience isn't wrong, the ask is.
A doctor might ignore "want a call about growing your practice" and say yes to "want to come on a show and talk about how you built it." The coaching client's campaign was already using a podcast invitation as the offer, which is why the real estate people responded at all. That's an offer test hiding inside an audience test.
So test in two dimensions. Several segments, and for your two or three strongest segments, two different offers. The same week, Charles was pitching three offer types to post-PMF companies: ICP testing itself, a partnership program aimed at others who sell to the same buyer, and podcast invitations instead of sales calls. Different audiences answer different asks.
When to kill an ICP
Killing a segment is harder than starting one. Founders get attached. "Doctors should work. Let me try new copy."
My rule: if a segment has had a fair volume of sends with the offer that's working elsewhere, and it's produced zero replies while other segments are producing, it's done for now. Don't rewrite the copy five times trying to resurrect it. Take the mailboxes and the budget and pour them into the segment that's already showing up.
The coaching client didn't need a better pitch for doctors. They needed more real estate people.
The follow-up is where the test pays off
One more lesson from that client weekly. Even with real estate showing up, the next priority wasn't more lists. It was the follow-up sequence. As the call put it, no one closes on first contact. A test that finds your ICP and then lets those leads go cold has wasted the finding.
So the full loop is: test wide, kill fast, double down where replies come from, and build the follow-up for that segment before you scale it.
The cheapest market research you'll ever buy
Every week you spend arguing about your ICP internally is a week the market could have answered for you. Set up the infrastructure once, run several segments and a couple of offers side by side, and read the reply column like it's the only research that counts. Because it is.
If you want to run that test with a show as the offer, start with your own show.
Frequently asked questions
How many ICPs should I test at once?
We usually run four to six genuinely different segments side by side, each with its own list and campaign. Fewer than that and you're mostly confirming your own guess.
How many emails do I need to know if an ICP works?
Enough that zero replies means something, not a few days of sends. Judge on replies and booked calls, and compare each segment against the others running at the same time.
Should I change the offer or the audience first?
Test the audience first with your strongest offer, then test a second offer in your two or three best segments. Sometimes the audience is right and the ask is wrong.