GTM fundamentals for founders · Episode 02

PMF vs ICP: product-market fit and your ideal customer profile, decoded

PMF is something you find. Your ICP is how you write it down.

Jamie Sullivan · Arc 1: Who you sell to ·

Prefer YouTube? Watch it there

PMF and ICP, in plain words

Two acronyms that get used as if they’re the same thing. PMF, product-market fit, is when a particular kind of customer wants what you’ve built badly enough to pay for it and keep paying. ICP, your ideal customer profile, is a description of that kind of customer, written down so someone else can find more of them.

So PMF is something you find. Your ICP is how you write it down.

Why it matters now

Early on, who you sell to is mostly a feeling, plus three or four customers you know really well. Your first seller doesn’t have that feeling. Without it written down, they’ll work whoever picks up the phone, and that’s how good sellers end up with a full pipeline of the wrong deals.

Fit also isn’t all-or-nothing. A team can have fit with one kind of customer and not with another. Fit is with a kind of customer, not with “the market”, and the ICP is where you say which.

Going deeper

An ICP has four parts

  1. Firmographics: the facts about the company. Industry, size, location, stage or funding. These are things you can check without a conversation.
  2. Triggers: an event that makes now the time, such as a recent raise or a first sales hire. A trigger has a window: a raise from two years ago isn’t a trigger any more.
  3. Disqualifiers: the explicit “not a fit” rules.
  4. Customers you’d clone: your anchor customers. Three to five real customers you already have and would clone, not wishlist logos. They’re the evidence behind the other three parts.
1

Firmographics

The facts you can look up: industry, size, location, stage.

2

Triggers

What makes now the time. Each one has a window.

3

Disqualifiers

The explicit “not a fit” rules, checkable from the outside.

4

Customers you’d clone

Three to five real ones. The evidence behind the rest.

+ a pass/fail line what would prove it wrong, written down first

An ICP in four parts, plus the line that says when it’s wrong.

A persona is different, and it isn’t a fifth part. The ICP is which companies; a persona is which people inside them: who decides, who champions it, who uses it. I note it separately.

Disqualifiers first

The thing that changed how I write an ICP: the disqualifiers matter more than the description. “Fewer than five staff” or “no one selling yet” are checkable from the outside in ten seconds, before anyone sends an email. That’s why they save more time than the criteria do.

Vague or usable?

Side by side: “B2B SaaS companies, 10 to 500 employees, North America” fits almost anyone and rules no one out. The one that works has the four parts and a pass/fail line.

Bad looks like“B2B SaaS companies, 10–500 employees, North America.”

Good works likeThe four parts: firmographics, triggers (each with a window), disqualifiers and three to five real customers you’d clone. Plus a pass/fail line.

Why it mattersThe vague version fits almost anyone, so a seller can’t rule anyone out.

Bad looks likeCriteria only: who you’d like to sell to.

Good works likeDisqualifiers first: “fewer than five staff”, “no one selling yet”.

Why it mattersA disqualifier is checkable from the outside in ten seconds, before anyone sends an email.

Bad looks likeWishlist logos as evidence.

Good works likeAnchor customers you already have and would clone.

Why it mattersLogos you don’t have yet tell you nothing about fit.

Bad looks like“They raised money” (two years ago).

Good works like“Raised in the last few months” or “just made a first sales hire”.

Why it mattersA trigger has a window; an old one isn’t a reason to buy now.

Bad looks likeThe ICP is written once and never checked.

Good works likeIt’s a hypothesis with a pass/fail line, such as “strong-fit leads convert at least twice as often as the rest”.

Why it mattersIf nothing can prove it wrong, it never gets revised (episode 4).

A hypothesis with a pass/fail line

I treat an ICP as a hypothesis, not a description, so I write down what would prove it wrong. The line I use is a rule of thumb: strong-fit leads should convert at least twice as often as the rest. Without a line like that, the ICP never gets revised. Episode 4 is where you check it against your own numbers.

One dataset is worth knowing here. Ebsta and Pavilion’s “2024 B2B Sales Benchmarks: H1 Update” reports win rates 3.1x higher when opportunities match the ICP, and that only 18% of 2024 pipeline matched the business’s ICP. The sample was 4.7 million opportunities across 566 companies. It’s a vendor dataset drawn from its own customers, and it shows correlation, not cause, so I read it as a reference point rather than a target.

Narrowing pays: the Superhuman example

One well-documented case is Superhuman. Its founder, Rahul Vohra, described in First Round Review how they used Sean Ellis’s survey question: “How would you feel if you could no longer use this?”

At the start, 22% of users said they’d be “very disappointed” without it. They then looked only at the users who loved it most, busy professionals who live in email, and the score for that group was about 33%, with no product change. They split the roadmap: half on doubling down on what those users loved, half on fixing what held the others back. Within three quarters the score nearly doubled, to 58%.

  1. At the start

    22%

    All users, summer 2017.

  2. Segmented

    ~33%

    Only the users who loved it most. No product change.

  3. Built for them

    58%

    Within three quarters.

Share of Superhuman users who said they’d be “very disappointed” without the product, as described by founder Rahul Vohra in First Round Review. One startup’s own survey of its own users, not a benchmark.

What it teaches for an ICP: the first jump came from writing down who more precisely, not from building more. It’s one startup’s survey of its own users, reported by its founder, so it’s a worked example of the method rather than a number to aim at.

PMF signals worth watching

The signals I watch are plain ones: customers staying, customers asking for more, referrals, and deals getting easier with the same kind of buyer.

Sean Ellis’s question is one named way to test it. He used 40% “very disappointed” as the line for product-market fit. That’s one practitioner’s threshold, not a law, so I treat it as one input, not a verdict. Both it and the win-rate figure above are blended reference points; your own segmented baseline matters more (episode 8).

Try it this week · Workbook

Try this week

The four parts, how a seller checks each, and which customers it’s seen in.

  1. 01Name three to five real customers you’d clone. For each, write what happened just before they bought: usually your trigger.
  2. 02Write three disqualifiers you can check from the outside in ten seconds, before anyone sends an email.
  3. 03Fill in the firmographics, and give each trigger a window. Note the persona separately.
  4. 04Write the pass/fail line before you look at the data, and the date you’ll check it.
  5. 05If you run the Sean Ellis survey, compare your best-fit customers’ score with everyone else’s, and note how many answered.

Your ICP on one page

Your answers stay in this browser. Nothing is sent anywhere.

Firmographics
Triggers, each with a window
Disqualifiers
Customers we’d clone

Start with the disqualifiers. They save more time than the description does.

Worked example

Tallybird, filled in

Tallybird is a made-up company: a 12-person startup, four months past its Series A, selling month-end close software to small in-house finance teams. Every detail here is made up.

  • Firmographics

    B2B software, 50–300 staff, US or UK, Series A or B.

    All things you can look up without a conversation.

  • Triggers

    Raised in the last 9 months; or hired a first controller in the last 6 months; or a board or auditor asked for faster month-end reporting.

    Each one has a window. A raise from two years ago isn’t a reason to buy now.

  • Disqualifiers

    Fewer than 30 staff on LinkedIn; finance run by an outsourced bookkeeping firm, with no in-house finance hire; public company or 500+ staff; not B2B software.

    Every one is checkable from the outside in ten seconds.

  • Customers we’d clone

    A 140-person HR software company, a 90-person developer-tools company and a 220-person logistics software company.

    The triggers above come from what happened just before each bought: a first controller hired, a Series A and a board wanting monthly numbers within 10 days, an auditor flagging a slow close.

  • Pass/fail line

    Strong-fit leads reply at least twice as often as the rest.

    Wrong if they don’t. Checked in episode 4.

The persona sits outside the ICP: here, the head of finance or controller, with the CFO (or the CEO, if there’s no CFO) as budget owner.

Watch out for

  • Blended scores. A single “very disappointed” number across all users can hide a segment that already loves you. That’s the Superhuman lesson.
  • Vendor datasets. The 3.1x win-rate figure comes from one vendor’s customers and shows correlation, not cause.
  • Survivorship. Famous turnaround stories get written up; the ones where segmenting didn’t help mostly don’t.
  • Small samples. A survey with a dozen answers, or a segment with five customers, is “not enough yet” (episode 3).

Decoded this episode

The glossary

PMF (product-market fit)
A kind of customer wants it enough to pay and stay.
ICP (ideal customer profile)
That kind of customer, written down for others.
Persona
The people inside: who decides, champions, uses it.
Firmographics
Facts you can look up: industry, size, location, stage.
Trigger
An event that makes now the time to buy, with a window.
Disqualifier
An explicit “not a fit” rule, checkable from outside.
Anchor customer
A real customer you’d clone; the ICP’s evidence.
Pass/fail line
What would prove the ICP wrong, written down first.

Your turn

What’s your best disqualifier, the one rule that saves you the most time?

Answer in the YouTube comments