Road21 (aka productledgrowth.net)
← All posts

How to find your ICP when you only have a few customers

By · · 5 min read

Search for how to define your ideal customer profile and you get the same advice everywhere: look at your best customers and find the pattern.

That’s fine advice if you have 200 customers. You have three.

Three customers don’t have a pattern. They have a coincidence. So what do you do?

The standard advice breaks at small numbers

Golden Section’s ICP guide says to build it from your best existing customers, scoring each one on things like contract value, churn, support load and sales cycle. It also lists a case where this doesn’t apply: when “fewer than a handful of paying customers exist,” it says to use interviews and a projected scorecard instead. That’s the honest answer. But the guide spends almost all its length on the other case.

Outbound Republic has the better framing for an early company. Its post on building an ICP with almost no customers says early customers are “usually a reflection of your network, not the market.” Then: “at this stage you don’t have an ICP – you have hypotheses.”

I agree with that. But most of what ranks for this question stays vague on the steps. Here are mine.

Step 1: write down why your best customer bought

Not their industry. Not their headcount. The reason.

What was broken before they found you? What happened that week that made them act? Was there a deadline, a bad quarter, a new hire?

That’s the trigger, and it’s the most useful thing you can know. Two companies of the same size with different triggers are different customers. Two companies of different sizes with the same trigger are the same customer.

April Dunford’s book Obviously Awesome starts its positioning process in the same place. Step 1, “Understand the Customers Who Love Your Product,” says:

“The first step in the positioning exercise is to make a short list of your best customers.”

(Quoted from the book as excerpted by Commoncog’s summary and in a reader’s Kindle highlights, Kindle location 711.) She defines best customers as the ones who understood the product quickly, became raving fans, referred you and acted as a reference. With a few customers, that list is short. Good. Start there.

Step 2: ask the people who love it

You have a few customers. Call them. Ask who else they think would get the most out of what you do, and why.

Superhuman’s Rahul Vohra did a version of this at scale. In his First Round Review write-up he asked users how they’d feel if they could no longer use the product. He then looked only at the “very disappointed” group. His reasoning:

“you could just narrow the market with preconceived notions of who you think the product is for, but that won’t teach you anything new.”

(The page has a stray space in “you”; the quote above is otherwise exact.)

The point: let the people who value you tell you who you’re for. Don’t guess from a slide.

Wes Bush at ProductLed gives a blunt way to spot them. In his piece on ideal user profiles he writes:

“You can spot your ideal users by looking at which customers use the product the most, upgrade the quickest, and leave the most testimonials.”

Note he says users, not buyers. His point is that the person who uses the product daily is not always the one who signs the check. With three customers, that’s a good prompt: who actually uses what you built, and who only approved it?

You won’t run a survey with three customers. A 20-minute call with each does the same job. Ask what they’d do without you. Listen to how they describe the problem. Their words will beat anything you write.

Some colleagues of mine were selling an agentic solution to manufacturers: user agents that could potentially do almost anything, or at least be set up to. They led with blueprint error detection, request for quotation, and order intake. But when they talked to the manufacturers, they found the companies had very different pains. They had thousands of SKUs, and their prices were tracked in an Excel sheet. That sheet needed pricing updates more than five to six times a year. So they built that for them.

Step 3: pick two or three segments, not one

Now you have a trigger and a few guesses. Turn them into two or three segments. Specific ones. “B2B SaaS companies moving from sales-led to self-serve” is a segment. “Startups” is not.

On how narrow to go, Dunford is blunt in a post on her blog (not the book) about selecting a target market she writes:

“In my experiences the smaller and tighter you can get on your segmentation, the easier it is to get early traction. You can always go broader later.”

She also says “SMBs” is “just too big to be a practical target market for a startup.” Same goes for “startups” or “SaaS.”

Outbound Republic suggests lists of 50 to 100 contacts per segment, run at the same time, so you can compare. I think that’s about right. Small enough to be quick. Big enough to hear back from.

Step 4: let the market answer

Send each segment a short, specific message about the problem. Then read the replies, not the open rates.

Outbound Republic’s test is a good one: when you’ve hit the right segment, prospects “reference the problem without you having to explain it.” When you haven’t, conversations are polite but slow. It suggests running each segment for two to three weeks.

This is the step most founders skip. They spend months debating the ICP in a doc. Three small campaigns would have told them in a month.

A caveat that I’d add and the sources don’t: small tests are noisy. If one segment replies twice and another replies once, that proves nothing. Look for a difference you can’t explain away. Prospects who answer fast and ask real questions are signal. A slightly higher reply count is not.

Step 5: write the one-page version, then revise it

When one segment pulls ahead, write it down on one page:

  • Who they are (industry, size)
  • The trigger
  • The role you talk to
  • Who looks like a fit but isn’t

The last one is the one people skip. Saying no to a prospect who looks fine on paper is what makes the ICP real. When your pipeline is thin, it’s also the hardest thing to do.

Treat the page as a guess you keep improving. After ten customers, redo it with real numbers.

The short version

With three customers, don’t analyze. Talk to them. Find the trigger. Test two or three segments with small outbound runs. Keep the one that answers fast. Write one page. Revise it.

If you want help running those tests, with the lists, the inboxes and the sequences, book a growth call.

Sources