What's an ICP, and Why Do You Need One

Third of six on the steps of go to market.

The ICP standard I work from allows exactly three attributes, constrained by the mechanism: at two attributes you have a market too broad, and at four you have segments you don't have the time to run.

An ideal customer profile is a forcing function, and the work it does is subtraction: plenty of people would benefit from what you've built, and this is the document that decides which of them to focus on… because with no sales team, the resource being allocated is your own attention.

Belkins (opens in a new tab), an outbound agency, sent 7.5 million cold emails through 2025: companies of 0 to 10 employees replied at 0.72%, companies of more than 10,000 at 0.22%. Which is why it’s important to decide which of these to target gets decided before you write a word.

Start from the five customers you already have

Name five companies, people or accounts that are your best examples, where best means highest lifetime value, fastest close, best retention, most referrals or most expansion revenue. Use their actual names. A label like "mid-market SaaS" hides the detail you need.

Then, against each one: the industry, the stage and size they were at when they bought, the tool or workflow signal that shows up in all of them, who pushed hardest for the change, what triggered them to go looking, and what not solving it was costing them in their own words. Specific enough to target is the bar. "Series B SaaS company" beats "SaaS company", and "Head of Revenue Ops using HubSpot" beats "ops team".

If nobody has bought yet, the five become ten people you already know who have the problem: former colleagues, the two founders in your group chat, whoever you helped for free last year and never invoiced (the free work is a better signal than it looks, because somebody let you near the problem). Rob Fitzpatrick's rule from The Mom Test stands for all ten of those conversations, which is that you ask what they already did and already paid for, never what they would do.

Reachability and urgency are the two that decide it

Every candidate attribute gets scored 1 to 5 on four dimensions: reachability, meaning whether you can reliably find and reach people who have this attribute; urgency, meaning whether it comes with acute pain and active buying; revenue potential; and win rate. The top three become the ICP, and when a fourth seems like it makes sense, the question is which of the current three is weakest.

Revenue potential and win rate are the two founders lead with, and at pre-revenue they're the two you can't evidence. I score reachability and urgency first and let the other two break ties.

Reachability at a company of one is whether you can personally hold every conversation the sale needs. TrustRadius (opens in a new tab) asked 2,164 technology buyers in March 2024 how many people were involved in their last software purchase, and at small businesses 53% had a buying group of two to three people. Two or three conversations fit in a week.

Win rate is the dimension to be careful with. Ebsta and Pavilion (opens in a new tab) analyzed 3.2 million sales opportunities and published the numbers in February 2023: small deals did best when the seller had one to three relationships inside the account, at a 29% average win rate. That is the strongest cell in their table.

Keep the candidates you rejected. They're what stops you drifting back to everyone who could buy, in the week when nothing has replied and the plan is hard.

The sentence you're trying to produce

One sentence: "We sell to [attribute 1] [attribute 2] [attribute 3] companies where [buying trigger]." For founder-led or non-company audiences the logic is the same: "We sell to [attribute 1] [attribute 2] [attribute 3] people when [buying trigger]." Underneath it goes the champion: "The person who finds us is the [role] who [cost of inaction]."

Filled in, this looks like: we sell to Series A B2B SaaS companies running revenue operations on HubSpot, where the trigger is a VP asking why forecast accuracy is still 60% after three quarters. The person who finds us is the Head of Revenue Operations who has been trying to build it themselves for months.

The trigger slot is where drafts go soft. "They want better forecasting" is a wish; a VP asking that question of someone on their team is the point in time that you need.

A persona and an ICP do different jobs

The three attributes are company-level: market, stage, size, stack, motion, budget pressure, an operational trigger. That is what makes qualification possible, because somebody looking at an inbound lead can tell from observable information whether it fits.

The archetype is the person, and it's a separate definition. What they want to be true in two years, what they're quietly worried about, what changed that sent them looking, and why they'd pick you. The line I use in the generator is that this isn't demographics, it's who answers your email at 7am and what they were doing at 6:55.

Aspiration and fear are the two forces that move a buying decision, and with only one of them you're selling to half a person. The test for the whole exercise is whether the persona changes your content, your offer or your channel; if it doesn't, it's decorative (I have written personas with a favorite podcast in them and no fear, and they were decorative). Company constraint first, because it decides who goes on the list. The archetype comes after, and it decides what the first line says.

Five checks before you use it

The profile isn't ready until it passes all five.

• Can you name 10 buyers, accounts or conversations that fit right now?
• Would you know from observable information whether an inbound lead fits
• Would your first message change meaningfully outside this profile?
• Can you reach them through one specific, repeatable channel?
• When you have lost a buyer who fit, was it a product gap?

A no on any of them comes with a diagnosis: too broad, too narrow, too behavioral to observe, unreachable, or ahead of what the product can currently prove.

When you're allowed to widen it

Geoffrey Moore's Crossing the Chasm is a skip on the scorecard of the commercial frameworks, because it was written for funded teams with a sales org. One residue survives for solo founders: pick an absurdly narrow niche and dominate it before expanding.

When a new segment turns up, score it green, yellow or red. Green is the same painful problem, reachable buyers, proof they'll believe, budget that exists, and implementation you can handle alone. Yellow is promising with one or two signals missing, and red is a different problem or an unclear buyer.

Paul Graham's do-things-that-don't-scale is the gate underneath all of it: founder-led selling stays non-negotiable to roughly $20K MRR. Widening before that adds segments to a motion that hasn't proved itself in one.

If you can’t choose

At pre-revenue you may not know enough to choose; this can be a real risk, or not one at all. What staying open turns into at one person is four conversations with four segments and no pattern in any of them. Picking badly shows up fast - not picking can show up in six months or longer.

The log

Every version of an ICP is worth tracking with five fields: the date, the three-attribute sentence, what triggered the change, what improved as a result, and a confidence score from 1 to 5.